Skip to content
Call wiki

Can AI tools see Google Docs comments and suggested edits?

Last updated 2026-08-17 · Confidence: documented — Google’s Docs and Drive API guides and Anthropic’s Drive integration page; the suggestions-read-as-prose failure is reported.

Google’s APIs expose both: suggested edits with exact positions, and comments with the passage they quote. What’s blind to them is claude.ai’s Drive connector — so the fix is an agent that calls the APIs (a script in Claude Code, or a Google Docs MCP server), not a better prompt.

Claude’s connector strips both

Section titled “Claude’s connector strips both”

Anthropic’s Drive integration page is explicit: Google Docs come through as “text extraction only,” and Comments/Suggestions are “not extracted.” Worse than absence, suggested text can arrive flattened into the prose with no markup, so the model critiques crossed-out sentences as if someone wrote them. The chat-level workaround is screenshots: vision sees the strikethrough and margin comments the text extraction drops. For how fresh the text it does read is, see Doc freshness.

Docs API: suggestions, with exact positions

Section titled “Docs API: suggestions, with exact positions”

documents.get takes a suggestionsViewMode: SUGGESTIONS_INLINE returns pending text tagged with suggestedInsertionIds/suggestedDeletionIds on each text run, at usable document indexes; two preview modes render the doc with all suggestions accepted or all rejected. The current guide also documents the write side — writeMode: SUGGEST turns an agent’s edits into suggestions, and AcceptSuggestionRequest/RejectSuggestionRequest resolve them — the suggest-mode leg of the Docs write path; editors built around AI suggestions are at Inline editing tools.

Drive API: comments, anchored by their quote

Section titled “Drive API: comments, anchored by their quote”

comments.list returns each comment’s text, author, replies, resolved flag, timestamps, and quotedFileContent — “the file content to which the comment refers.” The anchor field is stored but ignored for Docs: “Google Workspace editor apps treat these comments as un-anchored comments.” Location therefore comes from matching the quoted passage against the doc text — exact when the quote is unique, ambiguous when it repeats or was edited since.

Q&A from calls

Section titled “Q&A from calls”

Can Claude fetch a Doc’s comments with their locations — and does track changes work at all? Through the claude.ai connector, no to both: comments and suggestions are documented as not extracted. Through the APIs, yes to both: suggestions carry exact positions, and comments carry the passage they quote, which pins the location in all but repeated-text cases.

Sources

Section titled “Sources”