> For the complete documentation index, see [llms.txt](https://help.filed.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.filed.com/reference/products/ai-review.md). # AI Review ## What AI Review does AI Review reads a draft tax return and cross-checks it against the source documents already in the client's binder. It surfaces discrepancies, missing items, and potential errors as structured findings called leadsheets, which appear in the binder's Leads tab. The AI handles the cross-referencing work; the preparer reviews each finding and decides how to act on it. ## What AI Review checks AI Review compares values on the draft return against source documents in the binder, including W-2s, 1099s, K-1s, payroll summaries, prior-year returns, trial balances, and other supporting materials. It identifies: * Value mismatches between the draft and source evidence * Missing items flagged in the binder * Arithmetic inconsistencies across forms * Form-level anomalies Most findings are anchored to a specific form and line on the return. Filing-level findings, such as a required form that is missing, reference the form only. Supported return types: * Form 1040 (individuals) * Form 1120-S (S corporations) * Form 1065 (partnerships) * Form 1120 (C corporations) * Form 1041 (estates and trusts) * Form 990 (tax-exempt organizations) AI Review automatically scopes its checks to the forms applicable to the client's configured return type. ### Issue severity levels Each finding carries one of four severity levels: **Critical** - The finding indicates a significant discrepancy or missing item that should be resolved before the return is filed. **High** - Strong evidence of a mismatch or inconsistency between the draft and source documents. Warrants immediate attention. **Medium** - A potential issue that warrants review. The evidence is present but the discrepancy may be explainable. **Low** - An informational or low-confidence finding. May reflect a minor inconsistency or an item to verify, but is less likely to require a correction. Severity is assigned by the AI based on the evidence in the binder. The preparer makes the final judgment on each finding. ## Before you run AI Review AI Review uses whatever source documents are in the binder at the time of the run, so a more complete binder produces better findings. Before starting, upload the client's W-2s, 1099s, K-1s, and other supporting documents to the binder. The more evidence available, the more precisely AI Review can cross-check the draft return. ## Running AI Review {% embed url="" %} **Walkthrough: run AI Review** 1. Open **AI Review** from the client hub. On **Step 1 of 2 (Upload the draft 1040 return)**, add the draft return PDF via the drop zone or **Browse files**, then click **Next: Confirm source documents**. ![AI Review step 1: upload the draft return](https://storage.googleapis.com/filed-prod-public-assets/walkthroughs/ai-review/image1.webp) 2. On **Step 2 of 2 (Confirm source documents)**, review the binder preview - the source documents grouped by category and the missing-documents count - to make sure the binder looks complete. Use **Go to Binder** to add anything missing, then click **Start review**. ![AI Review step 2: confirm the binder source documents](https://storage.googleapis.com/filed-prod-public-assets/walkthroughs/ai-review/image2.webp) 3. AI Review runs and shows an in-progress loader ("Running AI Review") while it generates leadsheets. This usually takes a few minutes and updates automatically when ready, transitioning to the completed summary page (see [Reading the results](#reading-the-results)). ![The AI Review in-progress loader](https://storage.googleapis.com/filed-prod-public-assets/walkthroughs/ai-review/image3.webp) ## Reading the results When the review completes, the summary page shows: * The client name, tax year, and total number of findings * A four-tile breakdown with the count of Critical, High, Medium, and Low findings * A written narrative summarizing the review * A section listing findings grouped by form, with per-form counts The "Open leadsheets" button takes you to the Leads tab, where you can work through each finding in detail. ## Working through findings in the Leads tab ## Re-running AI Review You can re-run AI Review at any time, for example after adding new source documents to the binder or when a revised draft is ready. Each run produces a fresh, complete set of leadsheets and becomes the active result. Prior runs remain accessible by selecting them in the sidebar; they are not deleted, but binder flags update to reflect the latest run. ## In this section * [**Understanding flags**](/reference/products/ai-review/understanding-flags.md) - Where a finding's rule comes from, how to turn a check on or off, and a worked example of what AI Review catches that tax software won't. * [**Managing review protocols**](/reference/products/ai-review/managing-review-protocols.md) - Built-in and firm-authored checks, and how admins enable or disable them. * [**Prior-year comparison**](/reference/products/ai-review/prior-year-comparison.md) - How current-year values get compared against last year's return. * [**Sign-off levels**](/reference/products/ai-review/sign-off-levels.md) - What L0-L3 mean and who signs at each level. * [**Focus mode**](/reference/products/ai-review/focus-mode.md) - Keyboard-driven review of fields one at a time, with source evidence expanded alongside. * [**Re-review workflow**](/reference/products/ai-review/re-review-workflow.md) - The fix-in-tax-software, re-export, re-review loop. * [**Downloading work papers**](/reference/products/ai-review/downloading-work-papers.md) - Exporting the signed-off leadsheets as the final CSV or Excel work paper. * **Frequently asked questions** - Answers to common questions about what AI Review checks, supported return types, timing, severity levels, and how findings relate to the Leads tab. Planned future pages: Signing off findings, Supported return types. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://help.filed.com/reference/products/ai-review.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.