Google’s Deep Review, publicly launched on October 14, 2026, offers two free document reviews each week. Researchers, engineers and people preparing substantial proposals or presentations can use it to seek critique before human review—not a sign-off that the draft is correct.
Choose a question you can check
Google describes Deep Review as a way to examine reasoning, evidence and consistency in complex documents, including design documents and presentations. That makes it a candidate for a substantive question about a draft, rather than a quick punctuation fix.
For example, an engineering proposal author could ask whether the evidence supplied supports a proposed design choice. This is a suggested use: it still needs someone who understands the proposal to judge the response. If nobody can assess the objections, the review creates more unresolved questions rather than an approval decision.
The evidence also has boundaries. Google’s Paper Assistant Tool research, which underpins Deep Review, includes an evaluation of 26 selected scientific papers containing 29 equation or proof errors. That narrow exercise does not establish how reliably today’s service reviews business proposals or presentations.
Check sharing permission before uploading
Google’s current document-handling terms say documents remain private, are retained for up to a year and are not used for model training without permission. The FAQ also describes optional consent for research-team use to improve Deep Review and possible web searches for related work.
Opting out of improvement is not the same as immediate deletion. Before uploading client material, an unpublished joint paper or someone else’s manuscript, check the relevant sharing rules. A service’s privacy promises do not give you permission to disclose the document.
Understand the free and paid routes
Google sign-in is required. The portal’s FAQ lists PDF, text, Google Docs and Google Slides; optional context can identify areas to focus on.
Free reviews use Medium depth and Medium thinking effort; paid access unlocks other settings. Depth changes the number of passes; thinking changes effort within each model call. The FAQ does not specify the weekly reset time.
For extra reviews, Google’s pricing FAQ describes using your Gemini API key, with model calls billed at public rates and no Deep Review surcharge. An estimate appears after upload, before you choose to begin. The FAQ gives no fixed per-document price or guaranteed spending cap.
The consequence: permission to share comes before the spending decision. By the time the estimate appears, the document has already been uploaded.
Start with a review question that matters, rather than assuming the highest paid settings will be worthwhile. More work by the system is not, by itself, evidence that a particular objection deserves to be accepted.
Turn feedback into decisions, not automatic edits
The earlier NeurIPS author programme explicitly separated AI feedback from official peer review and warned that it could miss flaws or challenge correct statements. Those caveats are a reason to examine individual suggestions, not to treat a completed review as proof of correctness.
A useful workflow is to review a technical draft, check the objections, and revise only where the evidence supports a change. For each substantive suggestion:
Find the passage it concerns. Check whether the objection accurately represents what the draft says, including qualifications elsewhere in the document.
Check the evidence. Recalculate a disputed quantity or consult the underlying reference before changing the claim. If the issue needs specialist judgement, take it to the relevant reviewer.
Record a decision. Accept a supported correction, reject an objection you can explain, or leave an unresolved question for human review. Keep the feedback with the draft version it concerns.
Keep the document’s purpose in view, too. A suggestion can sound sensible without helping the intended reader or decision—a distinction explored in why AI advice can miss the real problem.
One possible use of the weekly allowance is an initial critique, followed by a second pass after checking and revising the draft. That is an allocation suggestion, not a promised improvement. Use the second review only if there is a remaining question worth examining; neither pass replaces the person responsible for the final document.