For the Legal Sector

40 Questions Legal Should Ask Before Trusting AI

When Harvey AI drafts a motion or Westlaw Edge surfaces a case, your professional liability depends on whether you caught what the system missed. These questions help you verify AI outputs before they leave your firm.

These are suggestions. Use the ones that fit your situation.

Download printable PDF

Citations and Legal Authority

1 When Lexis+ AI or Casetext returns a case citation, have you manually pulled that case to confirm it exists and the holding matches the summary provided?
2 Has the AI system told you which cases it considered but rejected, or are you only seeing the results it chose to highlight?
3 For the primary authorities cited in Harvey AI's draft advice, can you verify the current status of each case in your jurisdiction's appellate rules?
4 When ChatGPT generates a legal argument alongside citation, does that citation appear in your actual legal database or only in the AI's output?
5 Has the AI distinguished between binding authority in your jurisdiction and persuasive authority from other states or circuits?
6 For statutes cited by your AI tool, have you checked whether amendments or recent repeals have changed the text since the AI's training data?
7 When Westlaw Edge ranks cases by relevance, do you understand the weighting system it uses, or are you assuming relevance equals legal strength?
8 Has the AI flagged cases that contradict the position it is arguing, or is it presenting only supporting authority?
9 For any regulation or rule cited by the system, have you verified it against the current Code of Federal Regulations or your state's administrative code?
10 When you see a recent case cited as controlling law, have you checked whether a higher court has since overruled or limited it?

Professional Conduct and Liability Exposure

11 If you file a motion containing an AI-generated citation that proves false, who bears responsibility under your firm's professional liability policy?
12 Does your engagement letter tell the client that junior associate work is being replaced by AI, and does it disclose the risks that comes with?
13 Have you documented your review process for AI outputs in a way that would satisfy a bar ethics inquiry or malpractice defence?
14 When a contract management tool flags a risk, what happens if you miss that flag because you were skimming an AI summary rather than reading the original clause?
15 Does your firm's quality assurance process require a second attorney to review AI-drafted sections before they go to a client?
16 If Casetext misses an adverse case that a competent research would have found, can you demonstrate that your search strategy was adequate?
17 Have you told your malpractice insurer that your firm now uses Harvey AI or Westlaw Edge, and do they require specific review protocols?
18 When e-discovery AI flags documents as privileged, do you have a human review step, or are you relying on the system's classification for production decisions?
19 If a junior lawyer learns contract review only through AI suggestions and never performs independent clause analysis, what happens when they need to spot an issue the AI missed?
20 Does your conflict checking process confirm that the AI tool has access to your full conflict database, or could it suggest a firm member with an undisclosed interest?

Research Quality and Legal Reasoning

21 When Lexis+ AI returns ten cases ranked by relevance, how many of the bottom five have you actually read to verify they are truly less helpful than the top five?
22 Has the AI explained which facts in your client's situation it found most significant, or has it simply returned results based on keyword matching?
23 For a contract drafting task, can you articulate what risk the AI was designed to protect against, or are you accepting its clauses because they sound professional?
24 When you notice the AI's research misses an entire line of cases, is that a gap in the system's training data or a gap in how you phrased your research question?
25 Does the AI summary of a case accurately capture the court's reasoning, or has it simplified in ways that omit a crucial limitation?
26 For Harvey AI's draft advice on a statutory interpretation question, have you worked through the relevant canons of construction independently to test its conclusion?
27 When Westlaw Edge surfaces a case, does the system explain why it is relevant to your specific issue, or do you have to make that connection yourself?
28 Has the e-discovery AI been tested on your firm's prior document sets to establish its accuracy on finding privilege or confidentiality markers?
29 For a contract management flagged risk, have you traced that alert back to the specific clause language, or are you trusting the system's categorisation?
30 When an AI tool claims a case is analogous to your situation, can you articulate the material facts that make that analogy sound, or are you assuming professional-grade reasoning?

Client Communication and Expectation Management

31 Have you agreed with your client on a timeline for legal advice that accounts for your actual review time, or have you promised AI-speed delivery while doing full human verification?
32 When you bill a matter where AI research reduced the hours required, does your rate reflect the time actually worked or the traditional estimate the client expected?
33 If your e-discovery process using AI misses key documents, does your engagement letter allow you to claim you performed due diligence, or does it promise document completeness?
34 Have you disclosed to the client that junior lawyers will be doing less foundational research, which may affect the quality of their development and your bench for future matters?
35 Does your client know that contract review clauses are AI-generated, or have you positioned them as attorney analysis without qualification?
36 When a client asks why your AI-assisted research costs the same as traditional research, can you justify the fee based on the value of your judgment, or is price pressure forcing you to skip review steps?
37 If the client later discovers that a critical case or regulation was missed, can you explain your research methodology in a way that shows reasonable diligence?
38 Have you set expectations about the review timeline for AI-drafted documents, or have clients begun expecting same-day turnaround on complex matters?
39 Does your engagement letter specify which tasks will be handled by AI and which require direct attorney involvement, or have you left that ambiguous?
40 When you tell a client you used AI to reduce their legal costs, have you also explained the risks you are managing to ensure quality, or have you implied cost savings without trade-offs?

How to use these questions

Related reads

The Book — Out Now

Cognitive Sovereignty: How To Think For Yourself When AI Thinks For You

Read the first chapter free.

No spam. Unsubscribe anytime.