How to Verify Eve's Responses
Last updated: September 29, 2026
Overview
This article covers how to confirm that Eve's responses are accurate and grounded in your case documents. Eve gives you two built-in tools for it, quote verification and citations, and they work best combined with a few prompting habits and your own professional judgment.
This applies anywhere Eve generates written content from your matter documents: Talk to Eve, chat agents, drafting agents, and Medical Overviews.
Before You Start
A matter with documents uploaded to it. Quote verification and citations both run against uploaded documents, so text typed directly into Eve, such as a note, isn't covered.
Basic familiarity with asking Eve questions or requesting work product.
Set Yourself Up for Verifiable Responses
Structure the request before you send it. A focused request produces a focused response, and a focused response is much easier to check. State the task, say what you need it for, give Eve the case background it can't get from the documents, and name the format you want back. See 📄 How to Write Effective Prompts for Evefor the full structure.
Ask Eve to back up its assertions with quotes. Add an instruction like "include quotes from the medical records supporting each injury you identify" or "provide direct quotes for each key point." This grounds the response in specific source text instead of generalizations, and every quote that comes back is something quote verification can check for you. You should see inline quotes with citation bubbles throughout the response.
Break complex work into separate sessions. For a complicated matter, or a request that would return a large volume of analysis, create a separate session for each distinct task rather than running everything through one long conversation. Focused tasks produce better responses and make it far easier to verify each piece systematically.
Read the Verification Signals
Quote verification runs automatically on every quote Eve provides. Each quoted passage is checked against the documents in context and marked with a color before the response reaches you.
Green means the quote is verbatim. Eve found the text in your source documents exactly as quoted. Click the highlight to jump to that spot in the document. Green is your standard-review state: open the source when the exact wording is carrying weight, such as language headed into a filing or a demand, and move on otherwise.
Yellow means Eve found similar or paraphrased language rather than an exact match. Yellow appears less often, and it's common with scanned records, OCR text, and documents where formatting has shifted. Click the highlight and it takes you to the source document the same way green does, so you can compare the passage against the quote in the response. The information is usually still accurate with slightly different wording, but confirm it yourself before relying on it. If you want more detail on what Eve matched and why, ask Eve directly in chat.
Yellow is not a failure state. It's a signal to read the original, not a signal that Eve invented something. When you're deciding where to spend review time, start with the yellow ones.
Click citation bubbles to check context. For any quote supporting a critical fact or legal argument, click the numbered citation bubble next to it to read the surrounding passage. Do this even on green, because an accurately reproduced quote can still mislead if the context around it changes the meaning. Citations rank passages by relevance rather than listing every mention, so treat the list as the most relevant support rather than an exhaustive one.
In a draft, the equivalent move is to highlight the sentence and select Show Sources.
Go Deeper When the Stakes Are High
Ask Eve to check its own work. If something in a response gives you pause, follow up with: "check your work and make sure all of your results are verified from the documents I provided." Eve re-examines the response against your source documents and flags what it can't verify.
Cross-reference major work product against the originals. For demand letters, chronologies, and medical summaries, open the source documents alongside Eve's output. Confirm the critical facts came across and that nothing appears that isn't in your case file. You should be able to trace every significant claim back to a specific document.
Check the arithmetic yourself. Nothing in the verification tools checks math. Totals, date spans, and running balances are worth confirming directly.
Apply your professional judgment to every response. Treat Eve's output as a strong first draft, not final work product. Look for gaps, logical inconsistencies, and claims that feel unsupported even when they're cited. If something doesn't square with your understanding of the case or the law, dig further. The output should meet your standards before it goes to a client or into a filing.
What These Tools Don't Cover
Knowing the edges is part of using the tools well.
Sources outside your matter documents. Quote verification runs against documents in the matter and your Central Library. Quotes drawn from case law search, web search, or data synced from your case management system aren't checked the same way, so verify those against the source directly. Case law citations open the opinion inside Eve, and every web result carries its URL, title, publish date, and excerpt.
Conclusions, as opposed to quotes. A correctly quoted passage can still be summarized, characterized, or applied in a way you'd argue with. Verification confirms the text, not the reasoning built on it.
Omissions. Eve can be accurate about everything it says and silent about a document nobody uploaded. Checks run against what's in context, so a document excluded from matter context can't be cited, quoted, or verified against.
If you can't find a source for something important, ask Eve where it came from, or search the documents directly, before relying on it.