ChamberOne

When clients use ChatGPT before calling you

By the ChamberOne team 5 minute read

A client opens the first meeting by turning their phone towards you. On the screen is a confident, well-formatted answer about their dispute, and it is wrong in two places that matter. What you do in the next two minutes shapes the whole relationship. Dismiss the printout and you teach the client to stop telling you what they have read. Accept it and you inherit its mistakes. There is a third way, and it is mostly a matter of curiosity.

Key points

  • Ask what the client asked and what they relied on before you correct anything.
  • Correct specifics, not the tool or the client: show where the real rule or date comes from.
  • Read their prompts as a map of what worries them, and let it focus your advice.
  • Check whether confidential documents were pasted into public tools or anything was sent to the other side.
  • Agree simple ground rules instead of banning AI, and give an honest answer about your own use.

Why this is now an ordinary part of intake

Clients have always arrived with partial information: a neighbour's story, a forum thread, a half-remembered television programme. General-purpose AI tools are the latest source, and they are more persuasive than most because they answer in fluent, tidy prose that sounds as if a professional wrote it.

The tools are not useless to the client. They can help someone understand the vocabulary, organise their facts and work out what to ask. The trouble is that they usually answer in general terms, rarely know the local procedure, and cannot tell when a detail in the client's story changes everything. They also tend to sound equally sure whether they are right or wrong.

So the useful stance is neither alarm nor contempt. The client did some homework. Your job is to mark it.

Ask what they relied on before you correct anything

Open with questions, not rebuttals. You want to know what the client typed, what the tool said, and which parts they found convincing. People often paste a long answer but have really latched onto one sentence, and that sentence is the one to deal with.

  • What did you ask, and in what words? The framing of the prompt often explains the answer.
  • Which tool was it, and did you give it any documents or names?
  • What part reassured you, and what part worried you?
  • Has it changed what you have already done, such as ignoring a letter, signing something or contacting the other side?

Correcting without condescension

Separate the tool from the person. Saying 'these tools often give general answers that miss local rules' lands better than 'that's rubbish'. It tells the client they were reasonable to look, and it explains the gap without blaming them.

Be specific about why a point does not apply. If the answer assumed a different country's rules, say so. If it missed a time limit, show the client where the real date comes from and what you will do to protect it. A concrete correction is more persuasive than a general warning, and it models how you want them to treat any future advice they find online.

Where the tool was right, say so. Clients trust a lawyer more when they see that the lawyer will concede a good point.

Suppose a client has been told by a chatbot that a landlord must return a deposit within a fixed number of days, and the number is from another country. You might say: 'That is a reasonable thing to look up, and the rule it describes does exist in some places. Here is how it works here, and here is what we will do about your deposit.' Nobody has been humiliated, and the client now knows where the advice comes from.

Lands badly
  • Dismissing the answer as rubbish
  • A general warning with no specific reason
  • Saying nothing where the tool was right
Lands better
  • Saying these tools often give general answers that miss local rules
  • Showing where the real date comes from
  • Conceding the points the tool got right
Two ways to correct a client's AI answer

Use their questions to focus the advice

A prompt history is a rough map of what the client is afraid of. Someone who asked three variations of 'can they sack me without notice' is telling you what keeps them awake, even if your engagement is formally about something else. Reading the questions can shorten the first meeting and make your advice land on the points that matter to them.

Consider asking clients to send what they asked and what they were told before the first meeting. Treat it as you would any other document the client volunteers: read it, note the assumptions in it, and record in your file what you said about it.

Protect the client from what they may already have done

The real risks are practical. A client who has pasted the other side's letter, a contract or a medical record into a public tool may have sent confidential material to a third party. A client who has drafted a reply with AI help may have conceded something in writing. Ask, without drama, and then deal with it.

If privileged or sensitive documents went into a consumer tool, think about what that does to confidentiality, and tell the client plainly how to avoid repeating it. If a message has gone to the other side, find out what it said before you decide your next move. Check the guidance from your regulator or bar if you are unsure how this affects your duties.

Set a simple expectation for the rest of the matter

Do not forbid AI. Clients will use it anyway, and a ban only means they stop telling you. Agree instead on a few ground rules that you can state in a sentence each.

  • Tell us if you have used an AI tool on this matter, and show us what it said.
  • Do not paste documents from the matter into public tools without checking with us first.
  • Do not send anything drafted with AI help to the other side or to a court without letting us read it.
  • If we disagree with what a tool told you, we will explain why, and you can still ask us questions.
Do
  • Tell the firm if you have used an AI tool, and show what it said
  • Expect an explanation if the firm disagrees with a tool, and keep asking questions
Don't
  • Paste matter documents into public tools without checking first
  • Send AI-drafted material to the other side or a court without the firm reading it
Ground rules to agree with a client who uses AI

Where the firm's own use of AI fits in

Some clients will ask whether you use these tools too. Have an honest, short answer ready. If you do, explain that a lawyer reads and is responsible for every output, and say what the firm does about confidentiality. If you do not, say so without sneering. Either answer is fine; evasion is not.

The point of the exercise is trust. A client who feels heard on the printout is more likely to bring you the next uncomfortable fact, and that is worth more than winning the argument about the chatbot.

This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.

Read next

When clients use ChatGPT before calling you | ChamberOne