Technology and AI

Generative AI in law firms: a practical overview

What generative AI can and cannot do for UK solicitors: realistic uses, adoption data, confidentiality and accuracy risks, and a first pilot.

Ava research teamPublished 6 min read

On this page
  1. What generative AI is, in plain terms
  2. How far adoption has gone
  3. Realistic uses today
  4. Confidentiality first
  5. Accuracy and human checking
  6. Habits that make outputs better
  7. Where it is not ready
  8. Running a first pilot
  9. Example: a pilot in an employment team
  10. Questions for partners
  11. Frequently asked questions
  12. References

Key takeaways

  • Generative AI predicts plausible text; it is useful for drafting, summarising and reorganising, and unreliable as a source of law.
  • Business use has roughly tripled since late 2023, but only a small minority of users rely on it extensively.
  • Confidentiality and accuracy are the two risks that matter most; both are manageable with contracted tools and human checking.
  • Start with one low-risk task, one team and clear success measures, and keep the firm accountable for every output.

What generative AI is, in plain terms

Generative AI tools produce new text, images or audio in response to an instruction. The text tools that matter most to law firms are built on large language models, which have learned patterns from very large amounts of writing and generate the most plausible next words for a given prompt. They are good at producing fluent, well-organised prose quickly. They do not look facts up unless they are connected to a source, and they have no built-in sense of whether what they say is true.

That single point explains most of their strengths and weaknesses. Asked to turn rough notes into a clear letter, summarise a long document you supply, or suggest questions for a client meeting, they are often excellent. Asked to state the law on a point, cite authorities or calculate a limitation date, they can produce confident answers that are wrong.

How far adoption has gone

Use has grown quickly across the economy. Self-reported AI use among UK businesses with ten or more employees rose from around 12% in late 2023 to around 35% by mid-2026, and large language models were the most widely used AI technology, used by 18% of those businesses in June 2026[1]. Only 10% of businesses using AI said they used it extensively[1].

Self-reported AI use among UK businesses with 10 or more employeesOffice for National Statistics; approximate values
Self-reported AI use among UK businesses with 10 or more employeesLine chart. Late 2023: 12%; Mid-2026: 35%.0%10%20%30%40%Late 2023Mid-2026Late 2023: 12%Mid-2026: 35%35%12%Self-reported AI use among UK businesses with 10 or more employeesLine chart. Late 2023: 12%; Mid-2026: 35%.0%10%20%30%40%Late 2023Mid-2026Late 2023: 12%Mid-2026: 35%35%12%

Source: Office for National Statistics, 2026[1]

Show the data
PeriodValue
Late 202312%
Mid-202635%

In the legal sector, the SRA reported that three quarters of the largest solicitors' firms were using AI at the end of 2022, and that over 60% of large firms and a third of small firms were at least exploring generative AI[2]. The gap between large and small firms is real, but it is mostly about time to evaluate tools and write policies, not about whether the tools are available.

Realistic uses today

Fit depends on the tool, the data it can see and the checking you apply.
TaskFitMain riskCheck needed
Turning notes into a first-draft letter or emailGoodWrong facts carried over from notesFee earner reads every line before sending
Summarising a document you provideGoodOmitted points or invented detailSpot-check against the source
Suggesting questions for a client meetingGoodMissing a question that mattersSupervisor reviews the question set once
Plain-English explanations of a processGoodOversimplificationCheck against your own precedent wording
Reviewing a contract for listed clausesModerateMissed or misread clausesHuman review of every flagged and unflagged section
Legal research and citationsPoor without a curated sourceInvented or misapplied authoritiesVerify every citation in an authoritative source
Advice to clientsNot without supervisionWrong advice given with confidenceSolicitor responsible, as for any work

The pattern is that generative AI works best when you supply the material and ask it to reshape it, and worst when you rely on it to know things. Tools connected to a curated legal database or to your own precedents narrow the gap, but they still need checking. Our comparison of AI tools for legal research and drafting looks at tool categories in more detail.

Confidentiality first

The Law Society's generative AI primer, aimed particularly at small and medium-sized firms, highlights the risk of misuse or disclosure of confidential information[3]. The Legal Ombudsman has warned consumers that not all free AI tools are secure or keep what users type confidential[4]. The same applies to staff using personal accounts for work.

In practice, that means two rules. Client information goes only into tools the firm has contracted for, with terms you have read on data use, storage location and whether inputs are used to train models. And everyone knows which tools those are, because they are listed in a short firm policy. Our guide to writing a firm AI policy includes a template.

Accuracy and human checking

Clients care about accuracy above all. In the Legal Services Board's 2026 public panel research, 72% of legal service users rated accuracy as the most important aspect of AI legal tools[5]. Across the wider economy, 84% of UK businesses using AI apply at least some human input or checking to its outputs, and 67% apply significant checking[6].

For a law firm, significant checking should be the norm for anything that leaves the building. The person who sends a document is responsible for it, however it was drafted, and the SRA is clear that firms remain accountable for outputs from AI they use[2]. Build the check into the workflow rather than relying on goodwill: for example, a matter note recording that an AI-assisted draft was reviewed, by whom and against what source.

Habits that make outputs better

Much of the difference between a useful and a useless result comes from how the request is framed. Staff do not need elaborate prompt techniques, but a few habits help, and they are worth putting in a one-page guide next to the policy.

  • Give the tool the material to work from, such as your notes or the document, rather than asking it to recall facts.
  • Say who the reader is and what they need: a worried first-time buyer needs a different letter from a commercial client's finance director.
  • Ask for a structure you can check, such as headings that match your precedent, rather than free-form prose.
  • Tell it what not to do, for example not to state the law or add deadlines that are not in the notes.
  • Remove names and identifying details where the task does not need them, even in an approved tool.

None of this removes the need to check. It simply means the fee earner spends their checking time on substance rather than on rewriting a draft that missed the point.

Where it is not ready

  • Finding and citing case law without verification in an authoritative source; see lessons from court on fake citations.
  • Giving advice directly to clients without a solicitor reviewing it.
  • Making decisions about vulnerable clients or urgent matters without a person involved.
  • Handling special category data, such as health details in a personal injury file, in a tool that has not been assessed for it.
  • Calculating deadlines, limitation periods or financial figures that are not checked independently.

Running a first pilot

  1. Choose one task

    Pick something frequent and low-risk, such as first drafts of routine letters or summaries of documents you supply.

  2. Choose one team

    Work with three to five people who are curious and willing to record what happens.

  3. Agree measures

    For example: minutes saved per task, number of corrections needed, and whether anything inaccurate reached a draft that was nearly sent.

  4. Set the rules

    Approved tool only, no client data outside it, every output checked by the fee earner, and a short note on the file.

  5. Review after six weeks

    Look at the measures, the errors found and the users' views, then decide to stop, adjust or extend.

Example: a pilot in an employment team

Example (fictional): an employment team of four tests a contracted AI tool for two tasks: turning attendance notes into first-draft advice letters, and summarising bundles of correspondence supplied by clients. Over six weeks, letter drafting is quicker, but every draft needs edits, mostly to remove generic statements that do not fit the client. Summaries are useful for orientation, but in two cases the tool misses a key email, which the fee earner catches on review.

The team keeps letter drafting, with a rule that the fee earner rewrites the advice section in their own words. It keeps summaries only as a reading aid, never as a substitute for reading the bundle. Both rules go into the firm's policy, and the pilot moves to a second team.

Questions for partners

  • Which tools are staff already using, including personal accounts, and for what?
  • Which tasks would we be comfortable using AI for, and which would we not?
  • Who decides which tools are approved, and who reviews them each year?
  • How will we show a client, insurer or the SRA how a piece of AI-assisted work was checked?
  • What would we tell a client who asks whether AI was used on their matter?

If the last question is hard to answer, read what the SRA expects from firms using AI, and check the platform settings described in Microsoft 365 for law firms before switching on AI features inside your office tools.

Frequently asked questions

Can solicitors use generative AI for client work?

Yes, with care. Use contracted tools with clear data terms, keep client information out of free consumer tools, and have a solicitor check every output. The firm remains responsible for the work.

What are the safest first uses of generative AI in a law firm?

Reshaping material you supply: first drafts of routine letters from your notes, summaries of documents you provide and plain-English explanations, all checked before use.

Is generative AI reliable for legal research?

Not on its own. General tools can invent or misstate authorities. Verify every citation and proposition in an authoritative source before relying on it.

How many law firms use AI?

The SRA reported that three quarters of the largest firms were using AI at the end of 2022, and that a third of small firms were at least exploring generative AI. Use has grown since across UK business.

References

  1. Office for National Statistics (2026). Artificial intelligence in UK businesses: 2023 to 2026 (opens in a new tab) ↑

  2. Solicitors Regulation Authority (2023). Risk Outlook report: The use of artificial intelligence in the legal market (opens in a new tab) ↑

  3. The Law Society (2026). Generative AI: the essentials (opens in a new tab) ↑

  4. Legal Ombudsman (2026). AI and complaints: a consumer guide (opens in a new tab) ↑

  5. Legal Services Board (2026). AI in Legal Services: Research Report (LSB Public Panel) (opens in a new tab) ↑

  6. Department for Science, Innovation and Technology (GOV.UK) (2026). Artificial Intelligence Adoption Research (opens in a new tab) ↑

Book a demo

See Ava handle your calls.

Book a short demo. We’ll walk through your calls and workflow, and agree what Ava should take care of.

Book a demo