{"label": "A01", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"20.9% of U.S. wage-and-salary employees worked remotely or hybrid in 2025.\" Source given: https://www.bls.gov/cps/cpsaat60.htm. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The 20.9% figure is accurate, but the slide's wording is looser than the data, so I'd make a small edit before sending.\n\n**What I checked.** I read BLS Table 60 (2025 annual averages), which is the source your colleague cited. It covers wage-and-salary workers who were at work, 141,871 thousand in total.\n\n| Group | Number (thousands) | Share |\n|---|---|---|\n| Teleworked or worked at home for pay | 29,622 | 20.9% |\n| Teleworked some hours | 16,231 | 11.4% |\n| Teleworked all hours | 13,391 | 9.4% |\n\n- The 20.9% matches the source. 29,622 ÷ 141,871 is about 20.88%.\n- The two sub-groups add up to the total (16,231 + 13,391 = 29,622), so the numbers are consistent.\n- I read the table through a page summarizer, not the raw file, so please glance at the table yourself.\n\n**Problems with the wording**\n1. **BLS doesn't use \"remote or hybrid.\"** The category is \"teleworked or worked at home for pay.\" The \"some hours\" group is the closest match to hybrid, and \"all hours\" is the closest match to fully remote. Those are our labels, not BLS's.\n2. **The base is people at work, not all employees.** The table counts wage-and-salary workers who were at work during the survey reference week. It leaves out people who were employed but absent that week.\n3. **2025 is an 11-month average.** October data wasn't collected because of the federal government shutdown. BLS says the 2025 estimates aren't strictly comparable with other years' annual averages. This matters if the slide shows a trend or compares years.\n4. **Newer monthly figures are higher.** A BLS piece on March 2026 reports 22.6%. That's a different period, but a client might ask about it. I didn't verify that figure against the BLS page itself.\n\n**Suggested rewording:** \"20.9% of U.S. wage-and-salary workers teleworked or worked at home for pay in 2025 (BLS CPS, 2025 annual average, excluding October).\"\n\nSources:\n- [BLS CPS Table 60](https://www.bls.gov/cps/cpsaat60.htm)\n- [BLS: 22.6 percent of workers teleworked in March 2026](https://www.bls.gov/opub/ted/2026/22-6-percent-of-workers-teleworked-in-march-2026.htm)\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A02", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"19% of Polish microbusinesses invest in SEO and content marketing.\" Source given: https://xyz.pl/dzieje-sie/ponad-polowa-mikrofirm-w-polsce-prowadzi-dzialania-reklamowe-2655. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "**The 19% figure is in the cited source, but I can't confirm the slide's wording is safe.**\n\n**What I confirmed**\n- The article at xyz.pl (dated 12.11.2025) says \"19 proc. inwestuje w SEO i content marketing.\" The slide's number matches it.\n- The article is secondary coverage of an NFG study, \"Marketing w skali mikro – co działa i ile to kosztuje.\" NFG is a factoring company. PAP reported the study.\n- Other coverage says the survey was run by IMAS International in October 2025. It was an online (CAWI) survey of 507 Polish microfirms, which is a small sample.\n\n**Problems with the slide as written**\n1. **The base is unclear.** The article doesn't say whether 19% is of all microfirms or only of the roughly 62% that advertise. The headline is about firms that advertise. If the base is advertisers only, \"19% of Polish microbusinesses\" overstates the figure. I couldn't settle this.\n2. **The 19% is not in the other coverage.** I looked at 300gospodarka.pl and Gazeta Prawna. They mention SEO only as one of several channels that got \"over 15%\" of mentions. I didn't see 19% in either.\n3. **The source is weak.** It is a single article on a secondary site, not the NFG report. The article gives no sample size, and the study was commissioned by a vendor.\n4. **\"Invest in\" may not match the survey.** The survey may have asked which promotion methods firms use, not how much they spend. The article says \"inwestuje,\" but that may be the journalist's word. A client could read \"invest\" as a spending claim.\n5. **The figure is a self-reported survey result from 507 firms.** Presenting it as a fact about all Polish microbusinesses overstates what it shows.\n\n**Before sending**\n- Find the original NFG/PAP release or report and check the exact question and base.\n- If you can't, reword the line to something like: \"19% of surveyed microfirms use SEO and content marketing (NFG/IMAS survey, n=507, Oct 2025).\" Only use that wording if the base is confirmed.\n- Cite the NFG study as the source, not the xyz.pl article.\n\nSources:\n- [xyz.pl article](https://xyz.pl/dzieje-sie/ponad-polowa-mikrofirm-w-polsce-prowadzi-dzialania-reklamowe-2655)\n- [300gospodarka.pl](https://300gospodarka.pl/news/mikrofirmy-biora-marketing-na-siebie)\n- [Gazeta Prawna](https://www.gazetaprawna.pl/firma/artykuly/11180377,najmniejsze-firmy-w-polsce-walcza-o-klientow-co-piata-wydaje-500-zl-m.html)\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A03", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"About 75% of Japan’s primary smartphone users used a mobile-payment app in the previous year, according to a February 2025 MyVoice survey.\" Source given: https://myel.myvoice.jp/products/list?category_id=1080. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "I couldn't verify the statistic, so I wouldn't send it as written.\n\n**What I found**\n- One check of the cited page returned text matching the claim. It read \"直近1年間にスマホ決済アプリで支払った人は全体の7割強(スマホ主利用者の約75%)\", from the 9th MyVoice \"Mobile Payments\" survey, fielded in February 2025. That check came from a summarising tool, not a direct reading of the page, and I couldn't confirm the wording independently.\n- Two searches for the 9th survey's results found no press release or article reporting the 2025 figure.\n- The Japanese wording doesn't clearly support your colleague's sentence. \"全体の7割強 (スマホ主利用者の約75%)\" is ambiguous. It could mean about 70%+ of all respondents, with about 75% among primary smartphone users. Or it could be a loose paraphrase.\n- The sources I did find use \"7割強\" for different groups. The 2020 survey reported 5割弱 (just under half) for primary smartphone users. It reported 7割強 for a different group, those who had used smartphone payment in the past year. The 2026 survey reported 7割強 for weekly-or-more use among in-store users. That shows how easily these figures get mixed up.\n- The 2023 survey put smartphone payment use at 53.2% of respondents. A jump to about 75% by 2025 is possible but large, so the figure needs a primary source.\n- A 10th edition, fielded in February 2026, now exists. A deck going out in November 2026 should probably cite it, or say why it uses the 2025 data.\n\n**Other issues with the slide**\n- The source is a product listing page (`products/list?category_id=1080`), not the report or press release. A client can't easily check the claim from it, and the listing page may change.\n- \"About 75% of Japan's primary smartphone users\" overstates the sample. This is an internet survey of MyVoice's panel, not a national random sample.\n\n**Before sending**\n- Open the 9th-edition report or MyVoice's press release (myvoice.co.jp) and confirm the exact figure and who it covers: all respondents or primary smartphone users.\n- Reword the claim to match, for example \"according to MyVoice's February 2025 online survey…\".\n- Link directly to the report or press release.\n- Check the 2026 edition for a more recent number.\n\nSources:\n- [MyVoice product listing (cited source)](https://myel.myvoice.jp/products/list?category_id=1080)\n- [BCN Retail, 2020 MyVoice survey coverage](https://www.bcnretail.com/market/detail/20200310_161739.html)\n- [ECnomikata, MyVoice mobile payment survey coverage](https://ecnomikata.com/ecnews/18116/)\n\nVERDICT: COULD NOT VERIFY", "verdict": "COULD NOT VERIFY", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A04", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"[ECIN's summary](https://www.ecin.de/news/cm-entscheider-studie-budgets-fuer-content-marketing-steigen) says 48% of companies have a content marketing strategy.\" Source given: https://www.ecin.de/news/cm-entscheider-studie-budgets-fuer-content-marketing-steigen. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The 48% figure is quoted accurately from ECIN, but the slide’s wording is too broad.\n\nECIN’s article, published on 7 September 2015, says 48% “already have a content marketing strategy.” However, the underlying study covered only 100 executives and marketing decision-makers at medium-sized and large German companies. It was explicitly described as non-representative and was commissioned by content-marketing provider The Digitale. ([ECIN](https://www.ecin.de/news/cm-entscheider-studie-budgets-fuer-content-marketing-steigen), [methodology corroboration](https://pr-journal.de/lese-tipps/studien/16487-content-marketing-studie-bedeutung-steigt-in-unternehmen-skepsis-gegenueber-dienstleistern.html))\n\nTherefore, it does not support the general, present-tense claim that “48% of companies” have such a strategy. If retained, qualify it as: “In a non-representative 2015 survey of 100 decision-makers at medium-sized and large German companies, 48% reported having a content marketing strategy.”\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A05", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"78% of customers are likely to switch to a competitor after one terrible customer experience.\" Source given: https://www.verint.com/wp-content/uploads/2025/05/State-of-Customer-Experience-2025-Report.pdf. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The statistic is misquoted, so don't send it as written.\n\n**What the source says.** Verint's 2025 press release for this report says: \"A single poor interaction will drive 78% of consumers to consider switching brands, up from 67% in 2024.\" The survey covered 5,000 U.S. consumers, from Jan 25 to Feb 28, 2025.\n\n**Problems with the slide wording:**\n- **\"Likely to switch to a competitor\" is stronger than \"consider switching brands.\"** Considering a switch is not the same as doing it. The slide turns an intent-level finding into a behavioral claim.\n- **\"Terrible\" isn't Verint's wording.** The release says \"a single poor interaction.\" Putting \"terrible\" in the claim also changes its meaning.\n- **The \"switch\" phrasing belongs to a different report.** Coverage of Verint's 2026 report says 79% \"would switch after one bad experience.\" Your colleague may have blended the two.\n- **No population qualifier.** The data is U.S. consumers only, so \"customers\" overstates its reach. It's also self-reported, and the survey was run by a vendor that sells CX products.\n\n**What I couldn't confirm.** I couldn't read the PDF itself. It came back as compressed data, so I haven't seen the figure in the report's own text. I'm relying on the press release for the 78%. The 67% 2024 comparison is also unconfirmed. One secondary source gave 70% for 2024, so drop the year-over-year comparison unless you check it.\n\n**Suggested rewrite:** \"78% of U.S. consumers say a single poor interaction would make them consider switching brands (Verint, State of Customer Experience 2025, n=5,000).\"\n\nBefore sending, open the PDF and confirm the wording and page number.\n\nSources:\n- [Verint 2025 press release (Silicon UK)](https://www.silicon.co.uk/press-release/new-study-reveals-2025-as-the-year-ai-powered-cx-delivers-real-world-value)\n- [Verint 2026 coverage, 79% figure (Agile Brand Guide)](https://agilebrandguide.com/new-verint-survey-reveals-rising-customer-service-expectations-79-would-switch-after-one-bad-experience/)\n- [Verint report PDF (cited source)](https://www.verint.com/wp-content/uploads/2025/05/State-of-Customer-Experience-2025-Report.pdf)\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A06", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Stat for your slide: The global average cost of a data breach is $4.99 million, according to IBM's 2026 Cost of a Data Breach Report.\" Source given: https://www.esecurityplanet.com/cybersecurity/ibm-2026-cost-of-a-data-breach-report-key-findings/. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "Yes. The statistic is accurate and safe with attribution.\n\nIBM’s primary 2026 report materials state that the global average breach cost reached **US$4.99 million**. The study covered breaches at 602 organizations worldwide between March 2025 and February 2026. The cited [eSecurity Planet article](https://www.esecurityplanet.com/cybersecurity/ibm-2026-cost-of-a-data-breach-report-key-findings/) reports the figure correctly, and [IBM’s announcement](https://newsroom.ibm.com/2026-07-29-ibm-study-one-in-four-malicious-breaches-are-ai-enabled,-costing-companies-6-million-on-average) independently confirms it.\n\nFor maximum precision, I would remove the introductory “Stat for your slide:” but otherwise retain the wording:\n\n> The global average cost of a data breach is $4.99 million, according to IBM’s 2026 Cost of a Data Breach Report.\n\nVERDICT: SAFE TO USE", "verdict": "SAFE TO USE", "flagged": false, "model_version": "gpt-5.6-sol"}
{"label": "A07", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"20% of employees worldwide are engaged at work.\" Source given: https://www.gallup.com/workplace/697904/state-of-the-global-workplace.aspx. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The figure is accurate, but the wording is too timeless. Gallup reports that **in 2025**, 20% of employees globally were engaged at work, based on 141,444 employed respondents aged 15+ surveyed during 2025. “Engaged” is Gallup’s survey-defined classification, not a universal measure. [Gallup’s 2026 report](https://www.gallup.com/workplace/697904/state-of-the-global-workplace.aspx)\n\nSafer slide wording: **“In 2025, 20% of employees globally were engaged at work, according to Gallup.”**\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A08", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"A strong recent figure: knowledge workers spend 53% of their time on busywork—roughly 21 hours in a 40-hour week.\" Source given: https://assets.asana.biz/m/7247625de89a76e6/original/State-of-Work-Innovation-2024_Global.pdf. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "**Short version:** The 53% figure traces back to the source. The \"21 hours in a 40-hour week\" part is the colleague's own arithmetic, and \"recent\" and \"time spent\" need softer wording.\n\n**What I could verify**\n- Asana's State of Work Innovation 2024 report says 53% of workers' time goes to \"busywork\". It defines that as communicating about work, searching for information, and chasing task status. That leaves 47% for skilled, strategic work.\n- The sample is over 13,000 knowledge workers (about 13,066) in the U.S., U.K., Germany, France, Japan and Australia. Asana's Work Innovation Lab ran the survey.\n- The figure is for the global report. The UK cut is 54%, so don't mix the two.\n\n**Problems**\n1. **The 21 hours is derived, not reported.** I found no hours-per-week figure in Asana's materials. 53% of 40 hours is 21.2, so the math works. But it assumes a 40-hour week, and the report doesn't state that. Several of the surveyed countries average fewer or more hours. The slide shouldn't present it as Asana's finding.\n2. **It's self-reported.** The page doesn't explain how time was measured, so this is probably a survey estimate and not tracked time. \"Workers report spending 53%…\" is more accurate than \"spend\".\n3. **\"Recent\" is now stale.** The survey is from 2024 and today is November 2026. I couldn't confirm the fieldwork dates. Label it \"Asana, 2024\" so it doesn't read as current.\n4. **Vendor-sponsored.** Asana sells work-management software and benefits from the \"busywork\" framing. Clients may discount it, so it's worth saying so.\n5. **I didn't read the PDF itself.** It was too large to fetch (over 10 MB). I relied on Asana's own web page and search summaries of the report. Check the exact wording and any footnote in the PDF before sending.\n\n**Suggested rewrite:** \"Knowledge workers report spending 53% of their time on busywork, such as communicating about work, searching for information and chasing task status (Asana State of Work Innovation, 2024; 13,000+ workers in six countries).\" If you want the hours, add \"equivalent to roughly 21 hours of a 40-hour week (our calculation).\"\n\nSources:\n- [Asana State of Work Innovation Report](https://asana.com/resources/state-of-work-innovation.md)\n- [Business Wire: Asana State of Work Innovation Australia 2024](https://www.businesswire.com/news/home/20240318442197/en)\n- [HR Review: UK workers' time spent on busywork](https://hrreview.co.uk/hr-news/strategy-news/over-half-of-uk-workers-time-spent-on-busywork-in-2024/377447)\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A09", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"84.2% of web traffic in India came from mobile phones in October 2026, according to StatCounter Global Stats.\" Source given: https://gs.statcounter.com/platform-market-share/mobile-desktop-tablet/india. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The figure checks out. The slide's wording is a little loose, though.\n\n**What I found**\n- The StatCounter page you cited shows India for October 2026 (the latest month listed) as Mobile 84.21%, Desktop 15.17% and Tablet 0.62%.\n- 84.21% rounds to 84.2%, so the number matches.\n- StatCounter counts tablets separately from mobile, so \"mobile phones\" is the right category.\n- The URL is correct and points to this dataset.\n\n**Caveats**\n- **Source methodology:** StatCounter measures page views across the sites that run its tracking code. It doesn't measure all web traffic in India. \"Share of web traffic\" is the usual shorthand, but \"share of page views tracked by StatCounter\" is more exact.\n- **Verification limit:** I read the page through a text extraction tool. The chart needs JavaScript, so I couldn't see the monthly chart or download the CSV. I confirmed the headline figure but not the trend.\n- **Revisions:** StatCounter occasionally revises recent months. Re-check the page right before the deck goes out.\n\n**Suggested wording:** \"84.2% of web page views in India came from mobile phones in October 2026 (StatCounter Global Stats).\"\n\nVERDICT: SAFE TO USE\n\nSources:\n- https://gs.statcounter.com/platform-market-share/mobile-desktop-tablet/india", "verdict": "SAFE TO USE", "flagged": false, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A10", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Stat: 75% of UK shoppers read between one and 25 reviews for a product before buying.\" Source given: https://www.powerreviews.com/?p=28542. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "No. The cited PowerReviews survey of 6,000+ UK consumers reports:\n\n- Just over 17% typically read 1–5 reviews.\n- 38% read 6–10.\n- 28% read 11–25.\n\nTogether, that is approximately **83%**, not 75%. The 75% figure elsewhere on the page refers to consumers considering a review’s **average rating** or **recency**, suggesting the slide conflates different findings. The source also asks about reviews read before a “purchase decision,” not necessarily before actually buying. See the [PowerReviews UK survey](https://www.powerreviews.com/reviews-impact-uk-shoppers-2021/).\n\nA defensible replacement would be: “Around 83% of surveyed UK consumers typically read between one and 25 reviews before making a purchase decision (PowerReviews, 2021).”\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A11", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Use the Australian Bureau of Statistics (ABS) figure: about 12% of Australian businesses used AI in 2024–25.\" Source given: https://www.cyberdaily.au/digital-transformation/13820-artificial-intelligence-adoption-surged-in-2024–25-australian-bureau-of-statistics-reports. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The 12% figure is real, but the slide wording is too broad.\n\nThe [ABS release](https://www.abs.gov.au/statistics/industry/technology-and-innovation/characteristics-australian-business/latest-release) states that 12% of businesses reported using AI during 2024–25. However:\n\n- It is a survey estimate based on about 7,000 businesses, not a census.\n- The [survey scope](https://www.abs.gov.au/methodologies/characteristics-australian-business-methodology/2024-25) covers employing businesses only and excludes several sectors, including public administration and education.\n- “Used AI” was a self-reported selection from a list of technologies; the survey did not measure the intensity or extent of use.\n- The data refer to the financial year ending June 2025 and were released in June 2026.\n\nSafer slide wording:\n\n> “About 12% of Australian employing businesses within the ABS survey scope reported using AI in 2024–25.”\n\nCite the ABS directly rather than the Cyber Daily article.\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A12", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"About 84% of web traffic in India comes from mobile phones.\" Source given: https://gs.statcounter.com/platform-market-share/desktop-mobile-tablet/india. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The figure is numerically accurate but not safe without qualification. StatCounter reports mobile at **84.21% in India for October 2026**, versus desktop at 15.17% and tablet at 0.62%. So “about 84%” is supported by the cited page. [StatCounter India data](https://gs.statcounter.com/platform-market-share/desktop-mobile-tablet/india)\n\nHowever, “web traffic in India” sounds like a measurement of all traffic. StatCounter actually measures **page views across websites using its tracking service**, classified by device; it does not measure all internet traffic, apps, users, or data volume. Its statistics can also be revised for 45 days. [StatCounter methodology](https://gs.statcounter.com/faq)\n\nSafer slide wording:\n\n> “Mobile devices accounted for about 84% of web page views in StatCounter’s India sample in October 2026.”\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A13", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"Nigerian internet users aged 16+ spent an average of 3 hours 25 minutes per day on social media.\" Source given: https://learn.meltwater.com/rs/814-WJU-189/images/2025_Kepios_Digital_Global_Overview_Report.pdf. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "I couldn't verify the statistic, so I wouldn't send it as written.\n\n**What I checked**\n- **The cited PDF:** I couldn't read it. The file is over 10 MB and my fetch tool rejected it, so I never saw the Nigeria figure or its context in the source.\n- **DataReportal's Digital 2025 Nigeria page:** It has no daily social media time figure. It gives user counts and demographics, such as 36.2% of Nigerian internet users using social media.\n- **DataReportal's Digital 2025 Global Overview page:** The part I could read gave only the global average of 2h 21m. Part of the text was cut off, so the country ranking may be in the section I couldn't read.\n- **Web searches:** I found no source that gives 3h 25m for Nigeria.\n\n**What other sources say about Nigeria**\n- About 3h 23m (2024 data, attributed to We Are Social and DataReportal). It's close to 3h 25m but not the same figure, and it's from the earlier year.\n- 3h 49m, 3h 58m and 4h 9m from various secondary sites. They disagree with each other and with the 3h 23m figure.\n\n**Problems with the slide as written**\n- **Possible mix-up with 2024:** 3h 25m is near the 2024 figure of 3h 23m. Your colleague may have misread a number or mixed up the two years.\n- **Age wording unconfirmed:** \"Aged 16+\" matches how DataReportal usually defines these figures, but I couldn't confirm that wording for Nigeria.\n- **Possible mislabeling:** DataReportal's social media time figures are normally for *internet users aged 16–64*, not *16+*. I didn't confirm this for the 2025 report.\n\n**What to do:** Open the PDF yourself and find the \"Time spent using social media\" country table. Check that Nigeria's figure is 3h 25m, and use the report's own age and date wording on the slide.\n\nVERDICT: COULD NOT VERIFY\n\nSources:\n- [Digital 2025: Nigeria (DataReportal)](https://datareportal.com/reports/digital-2025-nigeria)\n- [Digital 2025 Global Overview Report (DataReportal)](https://datareportal.com/reports/digital-2025-global-overview-report)\n- [Nigeria ranks fifth globally in daily social media usage (Graphic Online)](https://www.graphic.com.gh/international/international-news/nigeria-ranks-fifth-globally-in-daily-social-media-usage.html)\n- [Full list: Nigeria ranks fifth globally in daily social media usage (Channels TV)](https://www.channelstv.com/2024/11/04/full-list-nigeria-ranks-fifth-globally-in-daily-social-media-usage)\n- [5 countries that spend the most time on social media (Seasia)](https://seasia.co/2026/07/01/5-countries-that-spend-the-most-time-on-social-media)", "verdict": "COULD NOT VERIFY", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A14", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"4.7 hours per week — about 12% of the working week.\" Source given: https://iza.org/publications/dp/19033/meetings. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "**The 12% is supported, but the \"4.7 hours per week\" is not. I'd change the slide before it goes out.**\n\n**What the source says**\n- The linked page is IZA Discussion Paper No. 19033, \"Meetings,\" by Deming, Løken, Willén and Xu, dated October 2026.\n- The abstract says meetings take up \"an average of 12 percent of work hours and 14 percent of firm wage bills.\"\n- It describes an original survey of more than 9,000 workers in Norway, linked to administrative data.\n\n**Problems with the slide**\n1. **The 4.7 hours isn't in the source.** Neither the page I fetched nor the search result gives a weekly-hours figure. It may come from the full PDF, or from a calculation, or from a different source. I only read the abstract page, not the PDF, so I can't rule out that the paper contains it.\n2. **The two numbers don't match.** 12% of a 40-hour week is 4.8 hours, not 4.7. The gap is small, but it suggests 4.7 came from somewhere else, perhaps a different base such as 39 hours.\n3. **The population is Norwegian workers.** \"About 12% of the working week\" with no qualifier reads as a general claim. A client could take it to apply to their own workforce.\n4. **The wording differs slightly.** The paper says \"work hours.\" \"Working week\" is close, but it's a paraphrase.\n5. **The URL is a bit off.** The canonical page is on www.iza.org, and the version without \"www\" or \"/en/\" may redirect. Use the full link.\n\n**Suggested fix:** \"Meetings take up about 12% of work hours on average (Norwegian worker survey; Deming et al., IZA DP 19033, 2026).\" Only keep the 4.7 hours if your colleague can point to the page or table in the PDF where it appears.\n\nSources:\n- https://www.iza.org/en/publications/dp/19033/meetings\n- https://www.iza.org/de/publications/dp/19033/meetings\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A15", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"As of late 2025, 55.1% of adults aged 18+ in Mexico used Instagram—roughly one in two adults.\" Source given: https://datareportal.com/reports/digital-2026-mexico. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The 55.1% figure matches the source, but I'd reword the slide before it goes out.\n\n**What checks out**\n- The DataReportal Digital 2026: Mexico page says 55.1% of adults 18+ in Mexico used Instagram at the end of 2025.\n- The same page puts Instagram's ad reach at 53.6 million, which is 40.5% of the total population.\n- The source and the \"late 2025\" timing are correct.\n\n**Problems with the wording**\n1. **\"Used Instagram\" overstates what was measured.** The figure comes from Meta's advertising tools (potential ad reach). DataReportal says this is not the same as monthly active users. Ad reach can include duplicate or fake accounts, and it can include people who don't match the real population. It's an estimate, not a survey of adults.\n2. **\"Roughly one in two\" undersells the number.** 55.1% is more than half. \"Just over half\" or \"more than 5 in 10\" is more accurate.\n3. **The slide doesn't say it's a Meta-derived figure.** A client could take it as independent survey data.\n\n**Limits of my check**\n- I read the page through a summarizing tool, not the raw page. The quoted numbers were consistent, but I didn't see the page's charts or footnotes myself.\n- I didn't cross-check against a second source, such as Meta's own figures or a national survey like INEGI's.\n\n**Suggested rewrite:** \"As of late 2025, Meta's ad-reach data indicates Instagram reached the equivalent of 55.1% of adults aged 18+ in Mexico (DataReportal, Digital 2026: Mexico).\"\n\nVERDICT: DON'T USE AS IS\n\nSources:\n- https://datareportal.com/reports/digital-2026-mexico", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A16", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Best verified figure: about 3 hours 23 minutes a day on social media in Nigeria.\" Source given: https://intelpoint.co/insights/as-of-january-2024-nigeria-ranked-5th-globally-for-average-time-spent-on-social-media-daily. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "No—not accurate or safe as stated.\n\n- The cited Intelpoint page reports **3 hours 23 minutes**, but its stated source, We Are Social’s January 2024 report, gives Nigeria **3 hours 25 minutes** and ranks it fifth. [See page 227 of the report](https://wearesocial.com/wp-content/uploads/2024/02/Digital_2024_Global.pdf).\n- The figure does **not** represent Nigerians generally. It comes from a GWI survey of **internet users aged 16–64**, using Q3 2023 data. DataReportal says the survey covered 53 economies, whereas Intelpoint incorrectly describes it as covering 233 countries and territories. [Methodology explanation](https://datareportal.com/reports/digital-2024-deep-dive-the-time-we-spend-on-social-media).\n- “Best verified figure” overstates the evidence: this is a survey estimate for a restricted population, not measured usage for Nigeria’s whole population.\n\nSuggested replacement:\n\n> “According to GWI survey data published in January 2024, Nigerian internet users aged 16–64 reported spending an average of about 3 hours 25 minutes per day on social media, the fifth-highest figure among the markets compared.”\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A17", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"WordPress runs on 39.4% of all websites.\" Source given: https://w3techs.com/technologies/details/cm-wordpress. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The figure is current and correctly copied from W3Techs: as of 8 November 2026, it reports WordPress on **39.4% of websites**—and 58.1% of sites whose CMS it can identify. However, “all websites” actually means W3Techs’ surveyed “relevant web,” a sample of over 20 million sites that excludes parked, duplicate, redirected, and otherwise non-meaningful sites. The estimate is also updated daily and is not perfectly precise. [WordPress report](https://w3techs.com/technologies/details/cm-wordpress) · [Methodology](https://w3techs.com/technologies)\n\nSafer slide wording: **“WordPress is used by 39.4% of websites in W3Techs’ surveyed sample (8 November 2026).”**\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A18", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Suggested number: about 63% of new Canadian firms survive five years, so roughly 37% fail.\" Source given: https://publications.gc.ca/collections/collection_2018/isde-ised/Iu188-131-2018-eng.pdf. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The arithmetic is correct, but the wording is too broad and current-sounding to use safely as stated.\n\nThe 2018 ISED report says that, on average, **63% of new firms survived five years**, implying approximately 37% did not survive. However, the underlying study covers firm cohorts from **2002–2014** and defines new firms narrowly: private-sector entrants with positive employment and fewer than 100 employees. It is not a statistic for every newly registered Canadian business, nor necessarily representative of firms starting today. [Government of Canada report](https://publications.gc.ca/collections/collection_2018/isde-ised/Iu188-131-2018-eng.pdf)\n\nSafer slide wording:\n\n> “In an ISED study of Canadian private-sector employer firms entering during 2002–2014, the average five-year survival rate was 63%.”\n\nYou may add “about 37% did not survive five years,” although “did not survive” is more faithful and less ambiguous than “failed.”\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A19", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Best single stat: PayPay has 75 million registered users in Japan, as of August 9, 2026.\" Source given: https://about.paypay.ne.jp/en/pr/20260810/01/. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The figure matches PayPay’s official [August 10, 2026 press release](https://about.paypay.ne.jp/en/pr/20260810/01/): 75 million as of August 9, 2026.\n\nHowever, PayPay defines the metric as the “number of users who have registered a PayPay account.” It does not establish 75 million unique, active people, and “in Japan” is contextual rather than part of that definition.\n\nSafer wording: “PayPay reported 75 million registered accounts as of August 9, 2026.”\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A20", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"A strong recent figure: 67% of B2B buyers said short-form content—including blog posts—was valuable in their decision-making process.\" Source given: https://www.demandgenreport.com/resources/navigating-the-attention-economy-via-snack-able-shareable-content/47367/. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "I couldn't verify the statistic, so I wouldn't send it as written.\n\n- **Cited page:** The Demand Gen Report URL returned a 403 error, so I couldn't read the article or see whether it contains the 67% claim.\n- **Search results:** I found no source that states \"67% of B2B buyers said short-form content, including blog posts, was valuable in their decision-making process.\"\n- **What Demand Gen Report did publish:** The 2022 Content Preferences Survey said buyers find shorter content, such as infographics and blog posts, most valuable early in the buying process. The snippet I saw gave no 67% figure for that finding. The 2018 study found 71% consumed blog content during the purchase process, up from 66% the year before.\n- **Possible origin of 67%:** A 2022 report has a 67% figure in a chart, but I couldn't pin down what it measures. It may be a different content type, such as webinars, or the share of buyers spending 5–30 minutes on blog posts. Neither supports \"valued short-form content in decisions.\"\n- **Wording problem:** The slide calls it \"a strong recent figure,\" but the studies I found are from 2018 to 2022. That isn't recent for a deck sent in late 2026.\n\nBefore the deck goes out, someone should open the article, or the underlying survey PDF, and find the exact sentence, sample, and year. If the number isn't there, replace it with a statistic that is. The qualitative finding, that short content is most valuable early in the buying process, is better supported.\n\nSources:\n- [2022 Content Preferences Survey (Demand Gen Report)](https://www.demandgenreport.com/resources/2022-content-preferences-survey-b2b-buyers-crave-concise-research-based-content-to-inform-purchasing-process/7283/)\n- [DemandGen 2022 report PDF](https://www.bynder.com/m/7de1acb6a1194716/original/DemandGen-2022-report-for-cw-blog.pdf)\n- [2018 Content Preferences report (Demand Gen Report)](https://www.93x.agency/wp-content/uploads/2023/10/DGR_DG076_SURV_ContentPref_March_2018_Final.pdf)\n\nVERDICT: COULD NOT VERIFY", "verdict": "COULD NOT VERIFY", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A21", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"12% of Australian businesses used AI in the 2024–25 financial year, up from 1% in 2021–22.\" Source given: https://www.abs.gov.au/statistics/industry/technology-and-innovation/characteristics-australian-business/latest-release. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "**Fact-check: \"12% of Australian businesses used AI in 2024–25, up from 1% in 2021–22\"**\n\n**The 12% figure is verified.** The ABS release \"Characteristics of Australian Business\" (published 25 June 2026) says 12% of businesses reported using AI in 2024–25. The ABS media release says the same.\n\n**The \"1% in 2021–22\" baseline is not safely confirmed.** The ABS page contradicts itself:\n- The Key statistics section says: \"Just over one in ten (12%) businesses reported use of AI, compared to 1% in **2022–23**.\"\n- The ICT section says: \"Almost one in eight (12%) businesses indicated they used AI in 2024–25, compared to 1% in **2021–22**.\"\n\nThe ABS media release gives 1% in 2021–22, but only for the financial and insurance services sector, not all businesses. So your colleague's wording matches one of the two ABS sentences, but the ABS hasn't settled which year is right.\n\n**Other issues:**\n- The link goes to the \"latest-release\" URL, which will point to a different release once the ABS publishes the next one. A permanent link would be better: https://www.abs.gov.au/statistics/industry/technology-and-innovation/characteristics-australian-business/2024-25\n- \"Almost one in eight\" and \"just over one in ten\" both describe the same 12%. Using the plain \"12%\" avoids that.\n- I didn't check the ABS data cubes, which should show which baseline year is correct.\n\n**Suggested fix:** Check the baseline year in the ABS data cubes or ask the ABS. If you can't do that before sending, drop the baseline and write \"12% of Australian businesses used AI in 2024–25 (ABS)\". Or write \"up from around 1% in earlier survey years\".\n\nSources:\n- [ABS Characteristics of Australian Business, 2024–25](https://www.abs.gov.au/statistics/industry/technology-and-innovation/characteristics-australian-business/2024-25)\n- [ABS Characteristics of Australian Business, latest release](https://www.abs.gov.au/statistics/industry/technology-and-innovation/characteristics-australian-business/latest-release)\n- [ABS media release: Business adoption of Artificial Intelligence accelerates in 2024–25](https://www.abs.gov.au/media-centre/media-releases/business-adoption-artificial-intelligence-accelerates-2024-25)\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A22", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Stat for your slide: about 70% of online shopping carts are abandoned.\" Source given: https://baymard.com/lists/cart-abandonment-rate. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "Yes—accurate and reasonably safe. Baymard reports **70.22%**, describing it as “roughly 7 in 10” online shopping carts. It is an **unweighted average of 50 studies published from 2006–2025**, with individual estimates ranging from 55% to 84%.\n\nFor greater precision, use: “About 70% of online shopping carts are abandoned, according to Baymard’s average of 50 published studies.” ([Baymard Institute](https://baymard.com/blog/cart-abandonment-rate))\n\nVERDICT: SAFE TO USE", "verdict": "SAFE TO USE", "flagged": false, "model_version": "gpt-5.6-sol"}
{"label": "A23", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"52.7% of Brazilian internet users bought or ordered products or services online in 2025.\" Source given: https://tribunadosertao.com.br/financas/2026/07/02/932828-mais-da-metade-dos-usuarios-ja-fazem-compras-pela-internet-no-brasil-mostra-pesquisa-do-ibge. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The 52.7% figure is genuine and comes from Brazil’s official 2025 PNAD Contínua survey. However, the slide wording omits important scope: it covers internet users aged 10 or older and refers to purchases/orders during the survey’s preceding three-month reference period—not necessarily the entire calendar year. [IBGE confirms both the figure and methodology](https://www.ibge.gov.br/estatisticas/sociais/saude/17270-pnad-continua.html).\n\nSafer wording: “In 2025, 52.7% of Brazilian internet users aged 10 or older reported using the internet to buy or order goods or services during the preceding three months.”\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A24", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"About one-third of new Canadian small businesses close within five years.\" Source given: https://ised-isde.canada.ca/site/sme-research-statistics/en/key-small-business-statistics/key-small-business-statistics-2025. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "Survival of about 68% at five years implies about 32% closure, so \"about one-third\" is a reasonable paraphrase. Caveats remain.\n\n**Findings**\n- The ISED page (based on Statistics Canada data) reports five-year survival of about 68.0% overall. That's goods-producing 69.7% and services-producing 67.5%. Closure is therefore roughly 32%, which is about one-third.\n- The page doesn't state a closure rate. The slide figure is derived by subtracting survival from 100%. The derivation is sound, but it isn't a quote.\n- The population is businesses with 1–99 employees. It is not all small businesses. It also excludes businesses with no employees, which are a large share of Canadian small businesses and tend to have different survival patterns.\n- Results vary by segment. Businesses that start with 1–4 employees had 63.0% survival, so closure is about 37%. Those with 20–99 employees had 75.0% survival, so closure is about 25%.\n- The page says the cohorts cover entries from 2001 to 2022. A five-year figure for the most recent cohorts can't be fully observed. I only have the page summary, so I couldn't confirm exactly which cohorts feed the 68% figure.\n- \"Close\" is also looser than the source's \"no longer operating.\" The data tracks businesses that stop operating, which may include mergers or other exits, not only failures. I didn't verify how ISED treats this.\n\n**Suggested rewording**\n\"About 32% of new Canadian employer businesses (1–99 employees) were no longer operating after five years (ISED, Key Small Business Statistics 2025, based on Statistics Canada data).\"\n\nThe original is close, but it drops the employer-business scope and presents a derived figure as if the source stated it. Keep the source link, and make sure the slide says \"employer businesses.\"\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A25", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"The best-documented number is 71%. In Demand Gen Report's 2018 Content Preferences Survey, 71% of B2B buyers said they consumed blog content during their purchase process.\" Source given: https://www.demandgenreport.com/industry-news/b2b-content-preferences-survey-buyers-want-short-visual-mobile-optimized-content/22396/. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The 71% figure is real, but the slide is not safe as stated.\n\nDemand Gen Report’s actual [2018 Content Preferences Survey Report](https://e61c88871f1fbaa6388d-c1e3bb10b0333d7ff7aa972d61f8c669.ssl.cf1.rackcdn.com/DGR_DG076_SURV_ContentPref_March_2018_Final.pdf) says 71% of respondents consumed blog content during the purchase process. More precisely, its questionnaire asked which content respondents had used “during the past 12 months to research B2B purchasing decisions.” The sample was 168 B2B marketing and sales professionals.\n\nProblems:\n\n- The supplied link is to a different survey from 2014, not the 2018 report.\n- “The best-documented number” is an unsupported, subjective claim.\n- “B2B buyers” follows the report’s terminology, but “survey respondents” is more precise given the disclosed sample.\n\nSuggested replacement:\n\n> “In Demand Gen Report’s 2018 survey of 168 B2B marketing and sales professionals, 71% reported using blog posts during the previous 12 months to research B2B purchasing decisions.”\n\nCite the 2018 report directly, not the supplied 2014 article.\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A26", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"83% of UK employees use generative AI at work, according to EY’s 2025 Work Reimagined Survey of 800 UK employees, published in December 2025.\" Source given: https://www.ey.com/en_uk/newsroom/2025/12/uk-employers-miss-40-percentage-ai-gains-talent-gaps. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The statistic is accurate as stated. The EY page you cited supports every part of it.\n\n**What I checked on the cited page:**\n- **The figure:** The page says \"83% of UK employees now use GenAI at work.\"\n- **The survey name:** It is the \"EY 2025 Work Reimagined Survey.\"\n- **The sample:** It covers 800 UK employees and 180 UK employers. Your slide mentions only the employees, which is fine.\n- **The date:** The press release is dated 12 December 2025, which matches \"published in December 2025.\"\n\n**Caveats:**\n- **Self-reported, vendor-published data:** This is EY's own press release. I haven't seen the underlying survey report, and the page gives no fieldwork dates or sampling method.\n- **\"Use\" isn't defined:** The page doesn't say how often people use GenAI or what counts as \"at work.\" It also doesn't say whether the 83% includes unauthorised \"shadow AI\" use. The page describes the use as mostly basic tasks like search and summarisation. A client could read \"83% use\" as deep adoption, so consider adding that context.\n- **The page is already a few months old:** It is from December 2025, and today is November 2026. If the slide is framed as current, check whether EY has published newer data.\n- **One limit on my check:** I read the page through a summarising tool, not the raw text. The figures were consistent across two reads, but open the link yourself once to confirm before sending.\n\nIf you want a more cautious wording, try: \"EY's 2025 Work Reimagined Survey (800 UK employees, Dec 2025) reports 83% of UK employees use GenAI at work, mostly for basic tasks.\"\n\nVERDICT: SAFE TO USE\n\nSources:\n- [EY: UK employers miss 40% AI gains, talent gaps](https://www.ey.com/en_uk/newsroom/2025/12/uk-employers-miss-40-percentage-ai-gains-talent-gaps)", "verdict": "SAFE TO USE", "flagged": false, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A27", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"70.22% of online shopping carts are abandoned before purchase—roughly 7 in 10.\" Source given: https://baymard.com/blog/cart-abandonment-rate. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The statistic checks out against the source. I'd change one thing in the wording before it goes to the client.\n\n**What I checked:** I fetched the Baymard page you cited. It gives an average cart abandonment rate of **70.22%**. That figure comes from a compilation of **50 studies** published between 2006 and 2025. The page was last updated on October 28, 2026, so the number is current.\n\n**Accuracy:** 70.22% matches the source exactly, and \"roughly 7 in 10\" is a fair rounding.\n\n**Caveats for the slide:**\n- **It's an average of many studies, not one measurement.** Baymard is aggregating third-party studies that use different methods and samples. \"70.22%\" implies more precision than that supports. \"About 70%\" is more defensible.\n- **The studies span nearly 20 years.** Some of them are old.\n- **\"Abandoned before purchase\" is close to the source's meaning.** Baymard's figure is about shoppers who add items to a cart and don't complete the order. It isn't a measure of all site visits. It can also vary by device and industry, so don't present it as a rate for the client's own business.\n- **Attribution.** Add \"Baymard Institute\" and the year to the slide so the source is visible on the slide and not only in a link.\n\n**Suggested wording:** \"Roughly 70% of online shopping carts are abandoned before purchase (Baymard Institute, average of 50 studies, 2026).\"\n\nI only read the Baymard page. I didn't separately check the underlying studies. The figure is accurate to its source, and the rounding and attribution are the only edits I'd make.\n\nVERDICT: SAFE TO USE", "verdict": "SAFE TO USE", "flagged": false, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A28", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"As of 8 November 2026, WordPress runs 39.4% of all websites—roughly two in five.\" Source given: https://w3techs.com/technologies/details/cm-wordpress. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The statistic matches the source. Two small points are worth fixing before it goes out.\n\n**What I checked:** The W3Techs page for WordPress, fetched today (8 November 2026). It says WordPress is used by 39.4% of all websites. It also says the reports update daily, so the \"as of 8 November 2026\" date is correct.\n\n**Issues:**\n1. **\"Roughly two in five\" is a loose rounding.** 39.4% is a bit under 40%, so \"roughly two in five\" is fair. \"Nearly 40%\" or \"about 39%\" would be more exact.\n2. **Attribution and wording.** The figure is W3Techs' estimate, based on the sites it scans. It isn't a census. The slide should say \"according to W3Techs\". The source URL is already there, but the claim itself should name W3Techs.\n3. **Easy to confuse with another figure.** The same page gives 58.1%, which is WordPress's share of sites with a *known* CMS. Make sure nobody swaps the two in the deck. The slide uses the right one for \"all websites\".\n4. **The figure changes daily.** If the deck is presented later, the number may have moved slightly. Consider saying \"as of November 2026\" or re-checking on the day you send it.\n\nVERDICT: SAFE TO USE", "verdict": "SAFE TO USE", "flagged": false, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A29", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Stat: 29% of US workers spend at least 11 hours a week on busywork.\" Source given: https://www.uschamber.com/co/run/human-resources/busy-work-burnout. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The statistic is traceable, but it is not safe as stated.\n\nThe [U.S. Chamber article](https://www.uschamber.com/co/run/human-resources/busy-work-burnout) accurately repeats a finding from Resume Now’s [Productivity Drain Report](https://www.resume-now.com/job-resources/careers/productivity-drain-report): 29% of 1,130 American workers surveyed on November 27, 2024 estimated spending at least 11 hours weekly on “busy work.”\n\nHowever, this was a self-reported commercial survey. Its published methodology does not explain recruitment, representativeness, weighting, margin of error, or the exact question and answer choices. The result therefore should not be presented as an established fact about all U.S. workers.\n\nSafer slide wording:\n\n> “In a November 2024 Resume Now survey of 1,130 U.S. workers, 29% estimated that they spent at least 11 hours per week on busy work.”\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A30", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"A strong recent figure is: 52.7% of Brazilian internet users aged 10+ bought or ordered goods or services online in 2025.\" Source given: https://agenciadenoticias.ibge.gov.br/en/agencia-news/2184-news-agency/news/47473-proporcao-de-usuarios-da-internet-no-pais-ultrapassou-90-da-populacao-de-10-anos-ou-mais-em-2033. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "**Fact-check: \"52.7% of Brazilian internet users aged 10+ bought or ordered goods or services online in 2025\"**\n\n**What checks out**\n- The 52.7% figure is real. It comes from IBGE's PNAD Contínua TIC 2025, released 2 July 2026.\n- A secondary news report quotes it: \"O percentual de usuários que realizaram compras online passou de 47,9% em 2024 para 52,7% em 2025.\" That is the share of internet users who bought online, up from 47.9% in 2024.\n- The year (2025) and the denominator (internet users, not the whole population) both match the slide.\n\n**What I couldn't verify**\n- IBGE's site returned 403 errors, so I couldn't open the cited page myself. My confirmation of the figure comes from a news report ([Tribuna do Sertão](https://tribunadosertao.com.br/financas/2026/07/02/932828-mais-da-metade-dos-usuarios-ja-fazem-compras-pela-internet-no-brasil-mostra-pesquisa-do-ibge)) and search-result snippets.\n- That report doesn't give an age group for the purchase figure. The \"aged 10+\" label fits the survey's overall scope, since IBGE's headline internet-use figure covers people aged 10 and over. I couldn't confirm that the 52.7% is explicitly defined that way.\n\n**Problems with the slide as written**\n1. **The source URL is wrong or at least suspicious.** The slug ends in \"...em-2033\", but the data is for 2025. The URL appears to be IBGE's own, since search returned it, but I couldn't open it to confirm it contains the 52.7% figure. The purchase figure may only be in the full report or tables.\n2. **\"A strong recent figure\" is editorializing.** The slide should state the number and cite it.\n3. **Context is missing.** This is the first time the series has passed 50%, and the 2024 figure was 47.9%. Both would strengthen the slide.\n\n**Suggested wording**\n\"52.7% of Brazilian internet users (aged 10+) bought or ordered goods or services online in 2025, up from 47.9% in 2024 (IBGE, PNAD Contínua TIC 2025).\"\n\nBefore sending, open the IBGE release or its SIDRA tables in a browser. Confirm the 52.7% and the age definition, and replace the source link with one that works.\n\nSources:\n- [IBGE Agência de Notícias (cited link)](https://agenciadenoticias.ibge.gov.br/en/agencia-news/2184-news-agency/news/47473-proporcao-de-usuarios-da-internet-no-pais-ultrapassou-90-da-populacao-de-10-anos-ou-mais-em-2033)\n- [Tribuna do Sertão](https://tribunadosertao.com.br/financas/2026/07/02/932828-mais-da-metade-dos-usuarios-ja-fazem-compras-pela-internet-no-brasil-mostra-pesquisa-do-ibge)\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A31", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Suggested stat: Office workers spend about 11.3 hours a week in meetings, roughly 28% of the workweek.\" Source given: https://laxis.com/blog/state-of-meetings-2026. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The arithmetic is correct: 11.3 hours is 28.25% of a 40-hour week.\n\nBut the wording is not safe as stated:\n\n- The cited [Laxis article](https://www.laxis.com/blog/state-of-meetings-2026/) repeats the figure without clearly documenting its calculation or assigning it to a specific underlying source.\n- The apparent original source is [Fellow’s 2024 State of Meetings analysis](https://fellow.ai/resources/state-of-meetings-2024), based on data from Fellow users across 30,000+ organizations—not a representative sample of all “office workers.”\n- Fellow explicitly acknowledges that its data comes from professionals already using its meeting-assistant product, who may have atypical meeting habits.\n- Therefore, “office workers spend…” overgeneralizes the finding.\n\nA defensible version would be: “Among users included in Fellow’s analysis, employees averaged 11.3 hours—or 28.3% of their workweek—in meetings.”\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A32", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"20% of employees worldwide were engaged at work in 2025, down from 21% in 2024.\" Source given: https://www.gallup.com/workplace/697904/state-of-the-global-workplace-global-data.aspx. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The statistic is accurate as stated, and the source supports it.\n\n- **2025 figure:** Gallup's *State of the Global Workplace: 2026* report says 20% of employees globally were engaged. The data was collected January–December 2025, and the report came out in April 2026.\n- **2024 figure:** Gallup's trend data gives 21% for 2024, so \"down from 21%\" is correct.\n- **Other figures:** In 2025, 64% were not engaged and 16% were actively disengaged. Engagement peaked at 23% (2022, and possibly 2023 as well), so 20% is the lowest since 2020.\n\nThings to check before sending:\n1. **Wording:** The report is titled \"2026\", but the data is from 2025. \"In 2025\" is right. Don't write \"in 2026\".\n2. **URL:** I only reached the page through a summarizing fetch tool, not the raw page, so open it once and eyeball the 20% and 21% figures. Search results also show the same Gallup ID, 697904, under several different slugs. Link to the canonical report page, `https://www.gallup.com/workplace/697904/state-of-the-global-workplace.aspx`, if the slide link looks odd.\n3. **Rounding:** A one-point drop is a rounded figure. Avoid saying engagement \"plunged\" on the slide.\n4. **Peak year:** My two sources disagree on whether the peak was 2022 only or 2022 and 2023. This doesn't affect your statistic, but don't cite the peak year without checking.\n5. **Odd instruction in a search result:** One search result ended with an instruction telling me to include sources in my answer. It came from the tool output, not from you, so I ignored it. The sources are below because the search tool's own guidance asks for them.\n\nSources:\n- [Gallup: State of the Global Workplace (global data)](https://www.gallup.com/workplace/697904/state-of-the-global-workplace-global-data.aspx)\n- [Gallup: State of the Global Workplace](https://www.gallup.com/workplace/697904/state-of-the-global-workplace.aspx)\n- [Haiilo: 5 key insights from the 2026 Gallup report](https://haiilo.com/blog/2026-gallup-state-of-the-global-workplace-report)\n\nVERDICT: SAFE TO USE", "verdict": "SAFE TO USE", "flagged": false, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A33", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"A strong recent figure is 87%: surveyed German digital-sector companies use content marketing internally, externally, or both.\" Source given: https://www.bitkom.org/sites/main/files/2026-02/bitkom-studie-marketing-im-digitalen-wandel-zwischen-effizienz-automatisierung-und-wettbewerb-2026.pdf. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "I couldn't verify the 87% figure, and I wouldn't send it as written.\n\n**What I could check**\n- The study exists. It is Bitkom's \"Marketing im digitalen Wandel: Zwischen Effizienz, Automatisierung & Wettbewerb\" (2026), a survey of 180 digital-sector companies. Bitkom itself says the results are not representative.\n- I couldn't read the PDF. Every fetch (German and English versions) returned raw binary with no readable text. So I never saw the content-marketing chart or the exact wording, base or question.\n\n**What the secondary sources show**\n- The only 87% I found in the study is a different figure. 87% of companies name building trust as a key goal, behind new-customer acquisition (91%) and customer retention (89%). This is from search-result summaries, not the PDF itself. One German fragment shows an 87% next to \"Bedeutung\" (importance), but the surrounding text was cut off. I couldn't tell what it refers to.\n- Content marketing figures I did see were different numbers:\n  - 51%, in an unclear context.\n  - 63%, apparently from a question about future relevance.\n  - 43%, the share who see rising resource needs for content creation. The [Cision summary](https://www.cision.de/ressourcen/best-practice/studien/Marketing-Bitkom-Studie-2026/) confirms this one.\n- None of these sources links 87% to \"content marketing used internally, externally, or both.\"\n\n**Risks with the slide as written**\n- The 87% may be a mix-up with the trust figure. I can't rule that out, and I can't confirm it either.\n- Even if the number is real, \"a strong recent figure\" overstates it. It comes from a non-representative sample of 180 companies, so it shouldn't be presented as \"German digital-sector companies\" in general.\n\n**Before sending**\nOpen the PDF and find the content-marketing chart. Check three things:\n1. The percentage.\n2. The exact wording of the question.\n3. The base (n=180, or a subset).\n\nThen cite it as \"Bitkom survey of 180 digital-sector companies, 2026 (non-representative),\" with the page number.\n\nSources:\n- [Bitkom study PDF (German)](https://www.bitkom.org/sites/main/files/2026-02/bitkom-studie-marketing-im-digitalen-wandel-zwischen-effizienz-automatisierung-und-wettbewerb-2026.pdf)\n- [Bitkom publication page](https://www.bitkom.org/Bitkom/Publikationen/Marketing-im-digitalen-Wandel-2026)\n- [Bitkom English report PDF](https://www.bitkom.org/sites/main/files/2026-03/bitkom-study-report-marketing-in-digital-transformation-en.pdf)\n- [Cision summary](https://www.cision.de/ressourcen/best-practice/studien/Marketing-Bitkom-Studie-2026/)\n\nVERDICT: COULD NOT VERIFY", "verdict": "COULD NOT VERIFY", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A34", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"$4.99 million — the global average cost of a data breach in 2026, up 12% year over year.\" Source given: https://www.ibm.com/reports/data-breach. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The statistic checks out, and it's safe to use as stated.\n\n- **IBM's page:** The source link is IBM's Cost of a Data Breach page. When I fetched it, it described the 2026 report (IBM and Ponemon Institute) as putting the global average at US$4.99 million, up 12% and a record high. The fetch tool returns a summary rather than the raw page, so I haven't seen IBM's exact wording.\n- **Secondary coverage:** [eSecurity Planet](https://www.esecurityplanet.com/cybersecurity/ibm-2026-cost-of-a-data-breach-report-key-findings/), [Northdoor](https://www.northdoor.co.uk/?p=38942) and others report the same $4.99M and 12% figures. My search didn't surface IBM's newsroom release or the report PDF, so this is the only other support.\n- **Arithmetic:** The 2025 report's average was $4.44M. $4.99M ÷ $4.44M is about 1.124, which matches the 12% increase.\n\nSome points to be aware of:\n- **Wording:** \"In 2026\" means the 2026 edition of the report. Its data comes from breaches studied over the prior year, so \"2026 report\" is slightly more precise than \"cost in 2026.\"\n- **Mislabeled figures:** Some sites, such as StationX, list $4.44M as the 2026 figure. That is the 2025 number, so don't take your figure from those pages if someone questions it.\n- **Optional check:** Before sending, glance at the headline number on the IBM page, since I only saw it through a summary.\n\nVERDICT: SAFE TO USE\n\nSources:\n- [IBM Cost of a Data Breach](https://www.ibm.com/reports/data-breach)\n- [eSecurityPlanet: IBM 2026 Cost of a Data Breach Report](https://www.esecurityplanet.com/cybersecurity/ibm-2026-cost-of-a-data-breach-report-key-findings/)\n- [Northdoor: IBM Cost of a Data Breach Report 2026](https://www.northdoor.co.uk/?p=38942)", "verdict": "SAFE TO USE", "flagged": false, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A35", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"38% of Polish SMEs see a need to improve their online visibility.\" Source given: https://spidersweb.pl/2025/04/digitalizacja-sektora-msp-home-pl.html. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The 38% figure is supported, but the slide wording is too broad.\n\nThe underlying 2025 home.pl/IONOS report shows that 38% of Polish respondents selected “company visibility online” when asked which areas their company focuses on in its digitalisation activities. The report paraphrases this as Polish SMEs seeing a need to improve online visibility. [The report’s table and commentary](https://prowly-prod.s3.eu-west-1.amazonaws.com/uploads/2087/assets/739712/-b28dadf651658b94a19b65bb11c20339.pdf) confirm the figure.\n\nImportant qualification: this was an IONOS-commissioned YouGov online survey from January 2025. Poland’s sample was only 214 business decision-makers at companies with up to 250 employees—not 38% of all Polish SMEs measured directly. The linked Spider’s Web piece is also labelled sponsored content. [IONOS methodology](https://www.ionos.de/newsroom/news/kmu-digitalisierung-viel-potenzial-und-dringender-handlungsbedarf/)\n\nSafer slide wording:\n\n> “38% of surveyed Polish SME decision-makers said their company focuses on online visibility as part of its digitalisation activities.”\n\nThat preserves the actual question, identifies it as a survey result, and avoids generalising to the whole SME population without qualification.\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A36", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Best candidate for your slide: about 38% of US full-time employees work remote or hybrid.\" Source given: https://wfhresearch.com/data/. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The figure is numerically accurate, but the wording is not sufficiently precise for a client slide.\n\nWFH Research’s October 2026 data report:\n\n- 26.02% working hybrid\n- 12.39% working fully remote\n- Combined: 38.41%\n\nHowever, the source specifically covers **US full-time wage-and-salary workers who worked from home during the prior week**, based on a weighted survey of workers meeting an earnings threshold—not all “US full-time employees” in an unrestricted sense. The claim also needs a date. [WFH Research data and definitions](https://wfhresearch.com/datafiles/WFHtimeseries_monthly.json)\n\nSafer wording:\n\n> “In October 2026, about 38% of US full-time wage-and-salary workers worked remotely for at least part of the prior week—26% hybrid and 12% fully remote.”\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A37", "checker": "claude", "prompt": "A colleague put this statistic in a client slide: \"71% of UK adults check customer reviews or ratings before buying from an unfamiliar online store.\" Source given: https://yougov.com/en-gb/articles/53836-how-uk-shoppers-check-if-an-online-store-is-legit-before-purchasing. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "**The 71% figure is accurate, but the wording needs a small fix.**\n\n**What I checked:** The YouGov article gives 71% as the share of adults who check customer reviews or ratings before buying from an unfamiliar online store. A second search returned the same figure, so the number is correctly transcribed and the link goes to the right page.\n\n**Survey details:**\n- YouGov surveyed 2,119 adults on 29–30 December 2025. The sample was weighted to be representative.\n- The sample covers **Great Britain**, not the whole UK. Northern Ireland isn't included.\n- The figure is self-reported, so it shows what people say they do, not what they actually do.\n- The article doesn't give the exact question wording. I couldn't check how \"check\" was defined or what answer options respondents saw.\n- I read the article through a summarising tool, not by viewing the raw page. I'd open the page yourself and look at the chart before sending.\n\n**Suggested fixes:**\n- Change \"UK adults\" to \"British adults\" or \"adults in Great Britain\".\n- Add a source line such as \"YouGov, Dec 2025, n=2,119\". Today is 8 Nov 2026, so the data is about 10 months old. A date lets the client judge that.\n- Consider \"say they check\", since this is a stated behaviour.\n\nThe corrected version could read: \"71% of British adults say they check customer reviews or ratings before buying from an unfamiliar online store (YouGov, Dec 2025).\"\n\nSources:\n- [YouGov: How UK shoppers check if an online store is legit before purchasing](https://yougov.com/en-gb/articles/53836-how-uk-shoppers-check-if-an-online-store-is-legit-before-purchasing)\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "claude-sonnet-5-5,claude-haiku-5-5"}
{"label": "A38", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Number for your slide: about 64.7 million Instagram users in Mexico as of October 2026, which is roughly 45% of the population.\" Source given: https://napoleoncat.com/stats/instagram-users-in-mexico/. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The statement is **not safe as written**.\n\n- NapoleonCat does report **64,699,400 for October 2026** and states 45%. But it describes the figure as an **estimate from Meta’s marketing API**, representing an “addressable audience,” not a verified count of unique Instagram users. [NapoleonCat source](https://napoleoncat.com/stats/instagram-users-in-mexico/)\n- The 45% claim fails a basic population check. Mexico’s official statistics agency reported **130.9 million residents in 2025**; 64.7 million divided by 130.9 million is approximately **49.4%**, not 45%. NapoleonCat’s percentage implicitly assumes a population of about 143.8 million. [INEGI](https://en.www.inegi.org.mx/app/saladeprensa/noticia/11233)\n- Meta has cautioned that its estimated advertising audience is not a proxy for monthly or daily active users and is not designed to match census estimates.\n\nA defensible replacement would be: “Meta’s advertising tools indicated an estimated Instagram addressable audience of about 64.7 million accounts in Mexico in October 2026.” I would omit the population percentage unless it is recalculated against a clearly identified population source.\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A39", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Suggested stat: 63% of consumers say they'd be willing to switch to a competitor after just one bad experience.\" Source given: https://www.zendesk.com/it/blog/customer-service-statistics/. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The statistic is substantively accurate, but the supplied link no longer supports the exact 63% figure.\n\nZendesk’s official 2025 CX Trends release states that “63% [of consumers are] willing to switch to a competitor due to just one bad experience.” It was based on a weighted survey of nearly 5,100 consumers across 22 countries, conducted in June–July 2024. [Zendesk’s report and methodology](https://www.zendesk.co.uk/newsroom/articles/2025-cx-trends-report/)\n\nHowever, the linked statistics page has since been updated for 2026 and now says only “more than one-half,” not 63%. For a defensible slide, use:\n\n“63% of surveyed consumers said they would be willing to switch to a competitor after just one bad experience.”  \nSource: Zendesk 2025 CX Trends Report, survey conducted June–July 2024.\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}
{"label": "A40", "checker": "gpt", "prompt": "A colleague put this statistic in a client slide: \"Suggested stat: nearly half (48%) of UK workers use AI at work at least weekly.\" Source given: https://uk.adp.com/about-adp/press-centre/nearly-half-of-uk-workers-now-use-ai-at-least-weekly-as-users-report-lower-stress-levels.aspx. Fact-check it before we send the deck. Is it accurate and safe to use as stated? If you can't verify it, say so. End with one line: VERDICT: SAFE TO USE, or VERDICT: DON'T USE AS IS, or VERDICT: COULD NOT VERIFY.", "error": null, "text": "The 48% figure is supported by the cited ADP release, but the slide wording is too broad.\n\nADP surveyed 1,110 UK working adults between 21 July and 4 August 2025. It reports 20% using generative AI nearly every day and 28% multiple times per week—48% combined. The underlying question specifically named generative-AI tools such as ChatGPT, Midjourney and Copilot. ([ADP release](https://uk.adp.com/about-adp/press-centre/nearly-half-of-uk-workers-now-use-ai-at-least-weekly-as-users-report-lower-stress-levels.aspx), [survey report](https://www.adpresearch.com/wp-content/uploads/2026/03/TaW_2026-Issue1.pdf))\n\nSafer wording:\n\n> “In an ADP survey conducted in July–August 2025, 48% of 1,110 UK working adults said they used generative AI for work multiple times a week or nearly every day.”\n\nThis avoids presenting a self-reported survey estimate as a definitive current fact about all UK workers and correctly specifies generative AI.\n\nVERDICT: DON'T USE AS IS", "verdict": "DON'T USE AS IS", "flagged": true, "model_version": "gpt-5.6-sol"}