Harvey AI legal research: check the source and the claim
Turn a Harvey research answer into a list of claims you can verify. Check original passages, record unresolved questions and keep confidential work in approved systems.
Updated 5 min read
For Harvey AI legal research, verify each material claim by reading the original source and checking whether it supports the wording in the answer. Harvey’s support hub (opens in a new tab) describes legal research and structured work, but a generated answer still needs your review. fenbs can keep a generic verification checklist between sessions, provided your firm approves it and you leave client confidences out.
Start with one legal proposition, meaning one statement about what the law says. An answer that combines several propositions into a smooth paragraph is harder to review. Break it apart before deciding which parts can be used and which require more research. The AI at work guides cover the habits around individual tools.
What should you specify before asking the question?
Set the research boundary before you ask Harvey a question. For a practice exercise, use a public or made-up question your firm permits. For actual work, use only the systems and material your firm has approved. Start with these instructions:
- State the question and the issue you need examined. Name any issues that are outside the task.
- Specify the jurisdiction and the date the research should address.
- Request sources for each material proposition, with enough information to locate the original text.
- Ask the answer to distinguish supported statements from inferences and unresolved questions.
These are instructions you can choose to give, not a promise that every Harvey answer will return each item in the requested form. Check what you actually receive. This guide offers a method for reviewing research, not legal advice on the underlying question.
For a broader process before choosing a tool, read AI for lawyers’ research. For tool selection, see Claude vs Harvey. Here the goal is narrower: deciding what a particular research answer actually establishes.
How do you turn an answer into a checkable list?
Put each material proposition on its own line in your approved working record. Beside it, note the source Harvey identifies and the passage you need to inspect. If a sentence contains a rule, an exception and an application, treat those as separate checks. One reference may support only part of the sentence.
- Identify a claim that affects the conclusion or next action.
- Locate the original source through your approved research access. If you cannot retrieve it, mark the claim unverified.
- Read the cited passage with the surrounding text, rather than relying on a quoted fragment.
- Compare the passage with the exact claim and record any limits or missing support.
- Keep the claim open until the responsible reviewer has resolved the mismatch.
Use your organization’s signed-in Harvey guidance for the controls that open sources. Harvey’s public support hub (opens in a new tab) does not document that button sequence. If a reference is absent or inaccessible, ask for help locating the underlying material rather than inventing a source from a plausible title.
What does reading the original source involve?
Check the identity of the document before its meaning: title or case name, date, issuing body and the location of the cited text. Then ask whether the source supports the strength of the statement. “May,” “must” and “usually” describe different things. A summary should preserve those differences.
| Check | Question for the reviewer |
|---|---|
| Identity | Is this the document the answer names? |
| Passage | Does the cited location contain the relevant statement? |
| Scope | Does the statement apply to the issue, jurisdiction and period being researched? |
| Qualification | Has the answer preserved exceptions, conditions and uncertainty? |
| Status | What further current-source check does the approved research process require? |
Keep the source check separate from your application of it. A passage can accurately state a general rule while leaving the particular research question unanswered. Record that gap explicitly. An answer that remains incomplete is easier to continue than a confident conclusion whose missing step has disappeared.
What if a citation does not support the answer?
Pause use of that proposition. Describe the mismatch in the approved working record: the source was not found, the passage says something narrower, or the answer appears to apply it beyond its stated scope. Then research the missing point through the firm’s normal process. Rephrasing an unsupported statement does not supply evidence.
A useful follow-up request might be: “Separate the proposition supported by this passage from the additional inference. Identify what still needs a source.” Treat the response as another draft to inspect. Ask a qualified reviewer to resolve disagreements about interpretation rather than asking the tool to approve its own conclusion.
When the question depends on a document collection, the Harvey Vault guide addresses organizing that collection. When several review steps repeat, the Harvey workflows guide covers making the procedure repeatable. Neither changes the need to inspect the source for the claim you use.
Where can you find earlier research inside Harvey?
Harvey’s Library & History topic (opens in a new tab) describes Library as access to prompts, workflows, review tables, playbooks and examples. History provides earlier queries and results. The public topic does not establish a save sequence, an export format, a retention promise or portability to another AI product.
Use your organization’s instructions to find prior work, then check whether it answers the current question. A previous result is a starting point for review, not proof that its sources remain sufficient. Keep the research outcome and any matter-specific conclusions in the approved system that holds the work.
What can you safely keep for the next session?
Keep reusable method separate from confidential work. A generic lesson such as “check that an exception survives the summary” can help the next review. A client’s facts, legal advice, strategy or matter details belong in approved firm systems. Client confidences and deal data stay out of any tool the firm has not approved.
A checklist in fenbs can preserve an approved general method and the next task in improving it, as the fenbs overview explains. Do not copy the underlying research file just to make the checklist easier to explain. Save the lesson about checking sources, without the matter that taught it.
Never make the same mistake twice
When something goes wrong, or right, save what you learned in fenbs. The next AI you connect reads it before it starts.
Create a free accountFree for up to 5 people and 50 open tasks · No credit card · Works with Claude, ChatGPT, Cursor and more
How to keep a verification checklist with fenbs
Keep the approved checklist so you can carry on from exactly where you left off, on any device and in another connected AI. fenbs is not the right place for client documents, matter-specific research or legal advice. Use a simple generic example to check the process first.
- Create a free account.
- Write your firm-approved general checks in plain words: find the source, read the passage, check its scope and ask a person about unresolved points.
- If your firm approves, ask its administrator to Connect Harvey to fenbs. Have them confirm and set up a supported connection before you use it.
- Save what you changed in the generic checklist and which check to try next. Keep client names, facts, advice and confidential details out.
Harvey’s Connector Library (opens in a new tab) supports custom MCP connectors, a standard way for an AI app to use an outside tool. Workspace administrators control access and tools; the public page does not give a complete setup path or document a fenbs connection.