Most Lawyers Now Use AI. Few Firms Have Built It Into How the Work Gets Done.

Ask a room of litigators whether they use AI and nearly every hand goes up. Ask who in the firm owns it, and what happens to the output before it leaves the building, and most hands come down. That second question is the one that matters.

Using AI and running on AI are different things. The first is a fee earner pasting a paragraph into a chat window. The second is a firm that has decided where AI is allowed in the workflow, who checks what comes out, and where the evidence of that check is kept. The first is common. The second is rarer, and it is where small firms are most exposed.

What the numbers say, and what they do not

LexisNexis published a survey of 543 legal professionals in early September 2026. The headline: 94% use AI for legal work, 74% at least weekly and 34% every day.

The figure that matters more is a different one. LexisNexis says 29% of organisations now have AI “embedded in strategy and operations”, which it reports as up from 17% in January 2026. Use is close to universal. Embedding is a minority position, though it is growing.

A follow-up LexisNexis UK report, covered by Legal Futures on 14 October, looked at who owns it. It was based on 543 lawyers polled in July 2026. The researchers said almost a fifth of lawyers did not know who managed AI governance at their firm. Asked separately about ownership, 18% said there was no clear owner. Another 40% named a single owner, and 39% said several teams were responsible.

Two limits to keep in mind. The results are not broken out by firm size, so we cannot say how a ten-person practice compares with a large one. And the September release does not state fieldwork dates. Read these as a picture of the profession, not of a high street firm.

A separate LexisNexis report on small law firms, published in May 2026, adds the other half. Of the small firms surveyed, 41% planned to increase their use of AI in research, drafting and review over the next 12 to 18 months, and 52% named administrative tasks as their biggest workflow problem. The release did not state a sample size.

So small firms want AI, they want it for repetitive work, and many cannot say who is responsible for it. That is where unchecked use is most tempting, and most risky.

What “embedded” looks like in a small firm

Embedding does not mean buying a big system. It means three things are true.

Someone owns it. A named person is responsible for where AI is used in the firm and what is allowed. In a small firm that is often a partner or the head of a department, not an IT lead.

There is a checked step. AI output goes to a human before it goes anywhere that matters, and the check is part of the workflow rather than a good intention. If the output is a witness statement, a letter or a schedule, someone qualified reads it against the source.

There is a record. You can show what the tool produced, what was changed and who approved it. When a court, a regulator or a client asks, the answer is in the file.

Most of what has gone wrong publicly with AI in legal work has failed one of those three. We looked at one example in our piece on the SDT strike-off for AI-invented citations, where the question was who signed.

Where small firms get stuck

Admin and case management are what firms say they find hardest. They are repetitive, high volume and low glamour, which is exactly where someone will quietly start pasting things into a free chatbot because nobody told them not to.

We wrote about the data side of that in what happens when client documents go into ChatGPT. The short version: if there is no approved route, people make their own.

The fix is rarely a ban. It is giving people an approved route that is quicker than the unapproved one, with the checks built in.

What we see when a firm does own it

Recently we made changes to a housing disrepair letter of claim for SBW Law, one of our development-partner firms. The feedback came back quickly and it was specific. That is the pattern we keep seeing with them.

That is what a firm looks like when someone has been given the job of owning how the tool is used. The person giving feedback knows what the document is for and what a good one looks like, so the changes are small, fast and right.

The tool is not the point. The owner is.

Three questions for any AI tool you already use

Before buying anything new, ask these of what you already have.

1. Who in the firm is accountable for this tool’s output? Not “who uses it”. Who answers for it.
2. What happens between the tool producing something and the work leaving the firm? If the answer is “the fee earner has a look”, write down what “a look” means.
3. Could we show, six months from now, what it produced on this file and who approved it? If not, the record is missing.

If you are weighing up a new tool, our earlier piece on five questions to ask before you buy an AI product for your firm covers the buying side. This piece is about what to put around it afterwards.

Where to start

Pick one repetitive task your firm already uses AI for, even informally. Write down who owns it, what the check is and where the record lives. If any of the three is missing, that is your first job. If you would like to see how a checked, recorded workflow looks in practice, book a demo.

© LegalDocs Assist. www.legaldocs-assist.co.uk