--- title: "Claim Ledger: Did Your AI Actually Read the Doc?" description: "How to give ChatGPT and Claude the same project notes in one Google Doc, and test whether a new chat can use them." author: "Harry Floyd" publication: "The Durability Curve" canonical: "https://durabilitycurve.com/claims/did-your-ai-read-the-doc/" essay: "https://durabilitycurve.com/blog/did-your-ai-read-the-doc/" substack: "https://harryfloyd.substack.com/p/did-your-ai-read-the-doc" published: "2026-10-16" ledger_date: "2026-10-16" claims: 38 removed_before_publication: 1 --- # Claim Ledger: Did Your AI Actually Read the Doc? *How to give ChatGPT and Claude the same project notes in one Google Doc, and test whether a new chat can use them.* Essay: https://durabilitycurve.com/blog/did-your-ai-read-the-doc/ Ledger (canonical, cite this): https://durabilitycurve.com/claims/did-your-ai-read-the-doc/ Published 2026-10-16 · last verified 2026-10-16 ## What the essay claims Chats forget, and different assistants cannot see each other's chats. Keeping the project's instructions and decisions in a couple of files fixes both, and GitHub is the free place for those files that the main assistants can connect to, with a record of every change. But a connection only lets the AI read; it does not make it read. So the setup is worth nothing until a fresh chat has proved it read the file, by naming it. For several assistants, the same files work, and the missing part is a doorbell: the repository stores the work but tells nobody it is there. ## The claim ladder | Rung | Claim | Evidence | Whose behaviour | Scope limit | |---|---|---|---|---| | 1 | A private repository, files and issues are free and need only a browser | GitHub Docs + pricing (ledger) | GitHub's product | Does not establish that a first-time user finds it easy; no novice panel was run | | 2 | ChatGPT and Claude can each be connected to a private repository | OpenAI and Anthropic help pages (ledger) | the vendors' products, on the plans named | Plan tiers, modes and read/write vary; does not reach Gemini or other tools unless verified | | 3 | A connection lets the AI read the files; it does not make it read them | vendor pages on when the connection is used (ledger) | the assistant's behaviour | Not a measured rate of skipped reads; the article offers the test, not a statistic | | 4 | The fresh-chat test tells "read" from "guessed" | the method itself: a question only the file answers, plus the file's name | the reader's own check | A named file can still be misquoted; the reader checks the answer against the file | | 5 | The repository stores work but wakes nobody; each assistant needs its own trigger | engine TEAM.md (first-party) + GitHub's coding-agent docs (ledger) | our five seats; GitHub's agents | First-party is one team's setup; no claim about time saved or ROI | | 6 | Text inside an issue can carry instructions an AI will follow | Invariant Labs write-up (ledger) | an agent with GitHub access | One documented incident; not a rate | ## What would make it wrong The piece is wrong, and does not ship as written, if any of these holds at publish time: ChatGPT or Claude cannot read a private repository on the plans the article names (C3, C4); any step needs a terminal or a paid plan for the core setup (C1, C2); the assistant's own Projects feature already shares one project across ChatGPT and Claude (it would make GitHub unnecessary for the main reader). ## The evidence, row by row Labels: VERIFIED = checked against the primary source itself by a checker who did not draft the piece, usually with the exact words, where they sit and the date read · EXECUTED = a run or observation done for the piece (code, a query, a count, or a check made live in an app); the claim is what it returned; run records are not published · CHECKED = checked against its source by a separate checker in the older ledger format, which recorded no quote, locator or date · REPORTED = not verified word for word against a primary source (secondary, not openable in full, or supported only in part); the essay words it accordingly · STRUCK = drafted, checked and removed before publication · EXCLUDED = considered and deliberately left out. ### C1. GitHub's free plan includes unlimited private repositories - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C1 - Quote: "Unlimited public/private repositories" - Source: GitHub pricing, Free plan card - URL: https://github.com/pricing ### C2. Files are created and saved in the browser: Add file → Create new file → Commit changes; editing a file uses the pencil and Commit changes - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C2 - Quote: "Click Commit changes..." … "Click Commit changes or Propose changes." - Source: GitHub Docs, Creating new files, steps 7 and 11 - URL: https://docs.github.com/en/repositories/working-with-files/managing-files/creating-new-files ### C13. A repository is created from + (upper right) → New repository: type a name, choose a visibility, optionally create a README; a file is edited with the pencil in the file view - Status: VERIFIED (read 2026-10-16) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C13 - Quote: "In the upper-right corner of any page, select , then click New repository." … "Choose a repository visibility." … "You can create a README" … "Click Create repository." … (commit box) "decide whether to add your commit to the current branch or to a new branch" … "In the upper right corner of the file view, click to open the file editor." - Source: GitHub Docs, Creating a new repository; Editing files, "Creating a new repository from the web UI" steps 1, 5, 6; "Editing files in your repository" step 2 (the + and the pencil are icons in the docs, so the quotes show a gap) - URL: https://docs.github.com/en/repositories/creating-and-managing-repositories/creating-a-new-repository ### L1. Navigation links lead where their text says: chatgpt.com (ChatGPT), claude.ai (Claude, sign-in when signed out), docs.new (a new Google Doc), github.com/signup (GitHub sign-up: "Create your free account"), github.com/new (new repository; signed out it goes to sign-in with return_to=/new) - Status: EXECUTED (read 2026-10-16) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-L1 - Quote: page titles and redirects - Source: First-party browser check (built-in browser, signed out) + curl - URL: https://chatgpt.com ### G1. On a Google Workspace account an admin can block third-party apps from Google data; a blocked user is told so, and the admin can trust the app - Status: VERIFIED (read 2026-10-16) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-G1 - Quote: "If a user tries to install (or sign in to) an app you haven't trusted that accesses a restricted service, they're notified that the app is blocked." … access levels "Trusted, Limited, Specific Google data, or Blocked" - Source: Google Workspace Admin Help, Control which apps access Google Workspace data, restricted services; App access control - URL: https://support.google.com/a/answer/7281227 ### L2. Claude's connector settings are at https://claude.ai/customize/connectors - Status: VERIFIED (read 2026-10-16) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-L2 - Quote: the address is linked from Anthropic's own pages (4, 6 and 3 times) - Source: Claude Docs (Google Drive; GitHub integration) + Anthropic Help (Google Drive), href in page - URL: https://claude.com/docs/connectors/google/drive ### D9. Claude's Google Drive connector reads a file's content, not its comments or suggested edits - Status: VERIFIED (read 2026-10-16) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D9 - Quote: "Comments and suggestions : Claude can't read a file's comments or suggested edits, only its content" - Source: Claude Docs, Google Drive connector, "Two limits apply to every type" - URL: https://claude.com/docs/connectors/google/drive ### D10. On Team and Enterprise, Google Drive appears only after an Owner adds it - Status: VERIFIED (read 2026-10-16) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D10 - Quote: "On Team and Enterprise plans, Google Drive doesn't appear in your connector list until an Owner adds it for your organization." - Source: Claude Docs, Google Drive connector, "Connect Google Drive" - URL: https://claude.com/docs/connectors/google/drive ### C4f. Claude can add a repository to a chat (+ → Add from GitHub, once GitHub is connected) as well as to a project - Status: VERIFIED (read 2026-10-16) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C4f - Quote: "By the end of this page you have added repository content to a chat or a project" … "select + at the lower left of the message box … Select Add from GitHub" - Source: Claude Docs, GitHub integration, "In a chat" tab - URL: https://claude.com/docs/connectors/github ### C3a. ChatGPT connects to GitHub at Settings → Plugins → GitHub; you pick the repositories on GitHub - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C3a - Quote: "To connect GitHub, open Settings > Plugins and select GitHub" … "then select the repositories ChatGPT can access" - Source: OpenAI Help, Connecting GitHub to ChatGPT, "Connect GitHub to ChatGPT" - URL: https://help.openai.com/en/articles/11145903-connecting-github-to-chatgpt ### C3b. ChatGPT's GitHub app only reads - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C3b - Quote: "The GitHub app in ChatGPT only lets you read from your repositories" - Source: OpenAI Help, same page, FAQ "Can I push code…" - URL: https://help.openai.com/en/articles/11145903-connecting-github-to-chatgpt ### C3c. Availability varies by plan; some plans have GitHub in deep research or agent mode but not ordinary chat - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C3c - Quote: "Some plans may support GitHub in deep research or agent mode but not in standard chat." - Source: OpenAI Help, same page, FAQ "Why is GitHub available in deep research…" - URL: https://help.openai.com/en/articles/11145903-connecting-github-to-chatgpt ### C3d. A new repository takes about five minutes to appear; ChatGPT searches the repository from your prompt - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C3d - Quote: "Repositories generally take about 5 minutes to appear as available in ChatGPT." … "ChatGPT forms search queries from your prompts" - Source: OpenAI Help, same page, "Find missing GitHub repositories"; "How ChatGPT works with GitHub" - URL: https://help.openai.com/en/articles/11145903-connecting-github-to-chatgpt ### C3e. In a chat you can pick an installed plugin with an @ mention, or ChatGPT uses it when relevant - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C3e - Quote: "ChatGPT automatically uses your installed plugins when they're relevant to your request. You can also select a plugin directly with an @ mention" - Source: OpenAI Help, Plugins in ChatGPT, "Use a plugin" - URL: https://help.openai.com/en/articles/20001256-plugins-in-chatgpt ### C3f. GitHub's app install screen offers All repositories or Only select repositories - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C3f - Quote: "select All repositories or Only select repositories" - Source: GitHub Docs, Installing a GitHub App from a third party, installation steps - URL: https://docs.github.com/en/apps/using-github-apps/installing-a-github-app-from-a-third-party ### C4a. Claude's GitHub Integration is on every plan including Free, adds repository files to a project's knowledge via + → GitHub → Add files, and reads file names and contents only - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C4a - Quote: "It's available on every plan, including Free" … "it doesn't see commit history, pull requests, or issues" … "Select Add files." - Source: Claude Docs, GitHub integration, intro; "In a project" tab - URL: https://claude.com/docs/connectors/github ### C4b. Claude can be given access to only specific private repositories - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C4b - Quote: "allow Claude access to all of your repositories or only to specific ones" - Source: Claude Docs, same page, "Grant Claude access to a private repository" - URL: https://claude.com/docs/connectors/github ### C4c. Claude keeps a copy; Sync now fetches changes - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C4c - Quote: "select Sync now on the repository in your project knowledge to fetch the latest changes" - Source: Claude Docs, same page, "Keep repository content current" - URL: https://claude.com/docs/connectors/github ### C4d. Repository content can be added only to a project you have not shared - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C4d - Quote: "You can add repository content only to a private project, one you haven't shared with other people." - Source: Claude Docs, same page, "In a project" tab - URL: https://claude.com/docs/connectors/github ### C4e. Claude's integration reads only; editing and pull requests go through Claude Code - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C4e - Quote: "If you want Claude to edit code, run commands, and open pull requests in your repository, see Claude Code" - Source: Claude Docs, same page, intro note - URL: https://claude.com/docs/connectors/github ### C6. ChatGPT and Claude each have Projects that keep files and instructions across chats inside that app - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C6 - Quote: ChatGPT: projects keep related chats, files and instructions together (paraphrase; page copy fc/oai_proj.md). Claude: projects are self-contained workspaces with their own chats and knowledge (paraphrase; fc/cl_proj.html) - Source: OpenAI Help; Claude Help, overview sections - URL: https://help.openai.com/en/articles/10169521-projects-in-chatgpt ### C7. A committed secret should be treated as compromised and revoked; deleting it in a later commit does not remove the exposure - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C7 - Quote: "You should consider any leaked secret to be immediately compromised" … "Simply removing the secret from the codebase, pushing a new commit, or deleting…" - Source: GitHub Docs, Remediating a leaked secret, Introduction - URL: https://docs.github.com/en/code-security/tutorials/remediate-leaked-secrets/remediating-a-leaked-secret ### C8. In May 2025 Invariant Labs showed a malicious issue on a user's public repository making their AI agent (GitHub MCP server) leak private-repository data into a public pull request. A researcher demonstration, not a reported real-world attack - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C8 - Quote: "allows an attacker to hijack a user's agent via a malicious GitHub Issue, and coerce it into leaking data from private repositories" - Source: Invariant Labs blog (Milanta, Beurer-Kellner), 2025-05-26, intro; "Attack Demonstration" - URL: https://invariantlabs.ai/blog/mcp-github-vulnerability ### C9. On a paid Copilot plan you can assign an issue to Copilot, which works, raises a pull request and asks you to review - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C9 - Quote: "Copilot will start working on the task, raise a pull request, then request a review from you when it's finished." - Source: GitHub Docs, Using Copilot cloud agent on GitHub; plans, "Assigning an issue to Copilot" (public preview); plans table excludes Copilot Free - URL: https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/use-cloud-agent-on-github ### C10. Claude and Codex coding agents can be assigned work on GitHub, public preview, paid Copilot plans - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C10 - Quote: Third-party coding agents page marks Claude and Codex as public preview on paid plans (paraphrase; fc/tp.md) - Source: GitHub Docs, About third-party coding agents, page note; "Usage costs" - URL: https://docs.github.com/en/copilot/concepts/agent-capabilities/about-agents/about-third-party-coding-agents ### C12. On Plus, Pro, Business, Enterprise and Edu, ChatGPT Work can run a task when pull request activity happens in a connected GitHub repository - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-C12 - Quote: "Event-triggered (webhook-based) tasks run in Work and respond to supported Gmail, Slack, or GitHub activity. They are available to eligible users on Plus, Pro, Business, Enterprise, and Edu plans" - Source: OpenAI Help, Scheduled tasks in ChatGPT, "Create an event-triggered task" - URL: https://help.openai.com/en/articles/10291617 ### F1. We run our own team of assistants from one private repository, and each assistant needed its own way to notice new work: one on new pull requests, one that checks every few minutes, and one that a script feeds - Status: EXECUTED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-F1 - Quote: Our team's own operating file, which names each assistant's trigger - Source: First-party: our team's operating file (private repository, not linked) ### F3. The article's steps were run end to end on 2026-10-15 in Claude (Max) and ChatGPT (Plus): both quoted the codeword and both followed the newer of two conflicting decisions; three UI labels corrected from the run - Status: EXECUTED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-F3 - Quote: "2026-10-15: Test. Codeword: quillfeather-83." quoted by both; Claude path + > Connectors > Add from Google Drive > Add to chat; ChatGPT Plugins in the sidebar - Source: First-party run record ### D1a. Claude connects to Google Drive at Customize > Connectors > Google Drive > Connect, on every plan including Free (Team/Enterprise: an Owner adds it first) - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D1a - Quote: "available for all users on Claude and Claude Desktop" … "Go to Customize > Connectors in claude.ai" - Source: Anthropic Help + Claude Docs, intro; "Connect Google Drive" - URL: https://support.claude.com/en/articles/10166901-using-the-google-drive-integration ### D1b. In a Claude chat, + > Add from Google Drive attaches a Doc, and Claude re-reads its current content every message - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D1b - Quote: "Claude reads the file's current content from Drive each time you send a message, so later edits in Drive reach Claude too." - Source: Claude Docs, Google Drive connector, "Add Drive files" › In a chat - URL: https://claude.com/docs/connectors/google/drive ### D1c. In a Claude project, a Drive file refreshes periodically when you open the project; only unshared projects - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D1c - Quote: "Its content refreshes from Drive periodically when you open the project." … "You can add Drive files only to a private project" - Source: Claude Docs, same page, "Add Drive files" › In a project - URL: https://claude.com/docs/connectors/google/drive ### D2. Claude can edit Google Docs live (beta) through a separate Google Docs connector, asking approval before each action by default - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D2 - Quote: "Beta: Claude can edit Google Docs, Sheets, and Slides live in a pane beside the chat" … "each action Claude takes on your behalf requires your explicit approval" - Source: Anthropic Help, "Edit … live (beta)"; "How to use the connectors" - URL: https://support.claude.com/en/articles/10166901-using-the-google-drive-integration ### D3. ChatGPT connects Google Drive under Settings > Plugins (Apps in some versions); availability depends on plan; in a chat use @ or +; changing a file needs Google permission and may not be available - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D3 - Quote: "Open Plugins or Apps, depending on the options available in your version of ChatGPT." … "Google Drive availability depends on your ChatGPT plan" … "mention an available app or plugin with @, or select +" - Source: OpenAI Help, "Connect Google Drive"; "Use an app" - URL: https://help.openai.com/en/articles/10929079 ### D4. Google Docs keeps version history (Last edit, top right; Restore this version); deleted text can usually be recovered unless the owner deletes history - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D4 - Quote: "At the top right, click Last edit" … "Restore this version" … "Deleting version history is permanent" - Source: Google Docs Help, "Restore an earlier version"; "Manage and delete version history" - URL: https://support.google.com/docs/answer/190843 ### D5. ChatGPT can recall details from past chats when Memory / Reference chat history is on - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D5 - Quote: "When Memory is enabled, ChatGPT can remember relevant preferences and details from your chats" - Source: OpenAI Help, Memory FAQ, Overview - URL: https://help.openai.com/en/articles/8590148 ### D6. Claude's memory is on by default on Free, Pro and Max; it can be switched off for one chat from the + menu - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D6 - Quote: "Memory is on by default for Free, Pro, and Max plans" … "turning off "Memory" in the "+" menu before you send your first message" - Source: Anthropic Help, "Manage Claude's memory in projects" - URL: https://support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects ### D7. ChatGPT event-triggered tasks respond to Gmail, Slack or GitHub activity only; Free and Go cannot create them - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D7 - Quote: "respond to supported Gmail, Slack, or GitHub activity" … "Free and Go accounts cannot create event-triggered tasks" - Source: OpenAI Help, Scheduled tasks, "Create an event-triggered task" - URL: https://help.openai.com/en/articles/10291617 ### D8. Once connected, both assistants can search the files the Google account can access, not only the project doc - Status: VERIFIED (read 2026-10-15) - Row link: https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-D8 - Quote: Claude: "lets Claude search your Google Drive, read the files it finds" · ChatGPT: "connects ChatGPT to files and folders your Google account can already access" - Source: Claude Docs; OpenAI Help, intros - URL: https://claude.com/docs/connectors/google/drive ## Removed before publication Claims the essay dropped or left out, each with the reason. - ~~A template is copied with Use this template → Create a new repository → choose visibility → Create repository from template; the copy has its own history~~ (STRUCK 2026-10-15): Removed from the essay, which now gives the two files and the steps itself (row C13) ## Cite - A claim: name the original source (and its locator) first, then the row it was checked in, e.g. ", . Checked in: Claim Ledger, "Did Your AI Actually Read the Doc?", The Durability Curve, row , https://durabilitycurve.com/claims/did-your-ai-read-the-doc/#c-." - The essay: Floyd, Harry (2026). Did Your AI Actually Read the Doc?. The Durability Curve. https://durabilitycurve.com/blog/did-your-ai-read-the-doc/ - This ledger: Floyd, Harry (2026). Claim Ledger: Did Your AI Actually Read the Doc?. The Durability Curve. https://durabilitycurve.com/claims/did-your-ai-read-the-doc/ - Downloads: https://durabilitycurve.com/claims/did-your-ai-read-the-doc.csv · https://durabilitycurve.com/claims/did-your-ai-read-the-doc.json Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Reuse it, quote it or train on it, with credit to The Durability Curve and a link. Open to AI (robots.txt: search=yes, ai-input=yes, ai-train=yes).