Your firm may have a fee earner who thinks AI is a waste of time. Often it is a senior person with a good track record and a sharp eye for errors. Their doubts are usually worth hearing.
Scepticism is not the problem. A rollout that ignores it is. If the people who do the most careful work stay away, the firm loses valuable reviewers and the tool risks being used badly by a few enthusiasts.
This article sets out a practical way to bring doubtful colleagues on board without pressure, slogans or promises the tool cannot keep.
Find out what the doubt is about
“I don’t trust it” can mean several different things. Each one needs a different answer, so ask before you try to persuade anyone.
Common concerns include:
- Accuracy: they have seen AI give confident, wrong answers and do not want to sign off on them.
- Confidentiality: they are not sure where client information goes or who can see it.
- Time: they expect checking the output to take longer than doing the work themselves.
- Professional standing: they worry about being blamed for a mistake the tool made.
- Billing and job security: they are unsure what faster work means for their hours, targets or role.
Hold short conversations with each person or team. Write down the concerns in their own words. You will use that list to plan the pilot and the training.
Answer each concern with facts, not enthusiasm
Sceptics respond to specifics. Give them the same information you would want before relying on any new supplier or system.
Accuracy
Agree with them. AI output can be wrong, and it remains unchecked until an appropriate person has reviewed it. Show them the review steps the firm expects, such as checking every citation against the source. Our article on checking AI outputs in legal work covers this in more detail.
Confidentiality
Tell them which tool and account configuration have been approved, where data is processed and stored, who can access it, how long it is retained, and whether inputs are used to train the supplier’s models. If you cannot answer those questions yet, the tool is not ready for client work, and the sceptic was right to ask. Our guide to keeping client information safe explains the checks.
Time
Do not promise savings you have not measured. Count the time spent preparing, checking and correcting, not only the time the tool takes to produce a draft. Be clear about the difference between time released and cash saved.
Responsibility, billing and roles
Say plainly that using AI does not transfer responsibility to the tool. The responsible fee earner, appropriate supervisors and the firm retain their existing responsibilities for the work. Explain how AI-assisted time will be recorded, how billing will be handled under the client’s agreed terms, and whether targets will change. If those decisions have not been made, say so and give a date for them.
Quick check: Could you answer each concern on your list in one or two plain sentences today? Any you cannot answer are gaps in your rollout, not in your colleague’s attitude.
Start with their work, not a demonstration
Generic demonstrations rarely convince careful people. A polished example on made-up facts tells them little about the tasks they handle.
Instead, ask the sceptic to pick one repetitive, lower-risk task. Use material the firm is authorised to process in the approved tool, with unnecessary client information removed. A closed matter is still confidential, and removing names alone may not make documents anonymous. Where those safeguards cannot be met, use a realistic fictional or suitably anonymised file.
Consider a hypothetical pilot in which a doubtful litigation partner chooses a closed matter that is suitable for testing under those safeguards. The tool produces a first-draft chronology from the permitted documents. The partner then checks it against the source documents and the chronology their team produced at the time.
This does three things. It tests a task the partner understands. It puts the sceptic in charge of the test. And it makes their critical eye part of the evaluation rather than an obstacle to it.
Keep the test small: one task and one team. You can see examples of everyday tasks on our AI uses page; each still needs a suitability assessment and a review process.
Give sceptics a formal role
People who look for faults are useful during a pilot. Make that official.
- Ask them to review outputs. Give them a simple form: what was wrong, how long it took to fix, and whether they would use the draft.
- Collect the hard examples. Ask them for the awkward documents, unusual facts and edge cases that are most likely to trip the tool up, using the same information safeguards.
- Let them shape the review checklist. Invite them to draft the checks for the task, working with the people responsible for supervision and risk. They will set a high bar, which is what you want.
- Record what they find. Keep a proportionate note of what AI was used for, who checked it and what was corrected, following the firm’s security and retention rules. This becomes evidence for or against wider use.
- Report back honestly. Share results that disappoint as well as those that impress. If the tool does not help with a task, stop using it for that task.
A sceptic who has stress-tested a tool and found it useful for a narrow job can be far more persuasive to colleagues than a supplier presentation.
Avoid the mistakes that harden resistance
Some approaches make doubtful colleagues dig in. Watch for these:
- Making use compulsory before the tool has been tested on suitable examples of the firm’s work.
- Measuring success by logins or prompt counts rather than useful, checked output.
- Talking about AI replacing tasks people take pride in, without explaining what happens to the time released.
- Treating questions about confidentiality or accuracy as negativity.
- Rolling out several tools at once, so nobody becomes confident with any of them.
Some people will remain unconvinced after a fair test, and that is fine. The aim is informed use where it helps, not universal enthusiasm.
Common questions
Should we make AI use mandatory?
Not at the start. It is usually better to require that approved tools are used safely when they are used, and to let results from a pilot build the case. You can revisit this once a task has a tested process and a review checklist.
What if a senior partner refuses to engage?
Ask them to review outputs rather than produce them. Their judgement on quality is valuable, and it does not require them to change how they work.
How long should a pilot run before we decide?
Long enough to cover a realistic mix of examples, including difficult ones. Agree the length and the measures in advance so the decision does not rest on first impressions.
Bring your whole team along with you.
Legal Edge helps firms run small, practical pilots and training that answer the questions careful fee earners ask. Get in touch to talk it through; scope and fees are agreed before any work starts.
Last reviewed: 3 November 2026.