ChatGPT for lawyers: what it can and cannot do

Is there a ChatGPT for lawyers?

There are two kinds of answer. General assistants such as ChatGPT, Claude and Gemini can be used for legal work, and their makers increasingly offer business plans and integrations. Separately, legal AI tools build on the same kinds of models but add legal sources, document workflows and controls designed for legal work.

OpenAI itself has moved in this direction: Legal IT Insider reported on 17 September 2026 that OpenAI had unveiled a legal configuration of its newest model, with a legal search index and a Word integration, in an early-access programme. Check OpenAI’s own announcements for current availability in the UK and EU.

What a general chatbot does well

• Turning rough notes into a clear first draft of a letter, email or summary.
• Explaining a concept in plain English, or restructuring an argument.
• Summarising a long document you provide, as a starting point for your own reading.
• Brainstorming issues, questions to ask a client, or counter-arguments.

Where it falls short for legal work

• Sources. An answer may state the law without tying it to a provision you can open, and can produce citations that do not exist. See why AI hallucinates citations.
• Currency. The model’s knowledge stops at a training date and may not reflect amendments; it may not say which version of the law it used.
• Jurisdiction. Unless told precisely, a general model may blend English, Scottish, US or EU rules.
• Confidentiality. Consumer accounts may keep conversations and, depending on settings and plan, use them to improve models. Read the provider’s terms before entering client material.
• Document workflow. Suggested edits arrive as text in a chat, not as tracked changes in the client’s document, so applying them is manual and error-prone.

Can ChatGPT review contracts?

It can read a contract you paste in and point out clauses worth a closer look, which is a reasonable first pass. It cannot know your client’s position, your playbook or the deal context unless you give it them, and its comments are not checked against the governing law. Treat the output as a list of things to look at, not a review.

For contract work, three things make the difference: the review is tied to the law it relies on, edits arrive as tracked changes you approve one by one, and the whole review is kept with the matter. See contract review.

Using a chatbot safely

1. Keep client data out of consumer toolsUse a business plan with terms you have read, or anonymise the material first.
2. State the jurisdiction and dateSay which law applies and the date that matters, and ask the model to say what it does not know.
3. Verify every authorityOpen each provision or case at the official source before relying on it.
4. Keep the decision humanReview every proposed change yourself before it reaches a client document.

How a legal workspace differs

Consilena uses large language models, including models from the Claude, GPT and Gemini families, inside a workspace built for checking. Agents search the legal sources it holds and record each step; every claim carries a numbered source that opens the provision with its status and date; edits to Word documents arrive as tracked changes; and each question shows a price estimate before you send it. It can still be wrong, and its output is not legal advice.