Part three of our AI search series.

What AI search tracking misses

If your professional services company doesn’t appear when you ask ChatGPT (or another LLM) for the best firms in your field, you don’t necessarily have a visibility problem. Visibility matters, but being recommended and chosen matters more, because AI answers change with who is asking. The generic question goes to the biggest names. The question your dream client asks is one you want to win.

Picture the marketing lead at a mid-sized employment law practice. They type “best employment law firms in the UK” into ChatGPT, read the answer, and see the same big players they see in every league table. Their firm isn’t there. The natural conclusion is that they have a visibility problem, and the natural next step is to find someone who will fix it.

That conclusion is half right. Visibility does matter. But the question they typed is not the question their buyers ask, and being named in an answer is not the same as being picked.

Visible, recommended, chosen: what’s the difference?

There are three steps, and each depends on the one before.

  • Visible means the AI names you.
  • Recommended means it names you with specific reasons why.
  • Chosen means the reasons it gives are the particular reasons the buyer cares about.

You can’t be chosen if you’re not visible. But visibility on its own wins nothing, and most AI search tracking stops there. It counts how often a brand is mentioned rather than just ranked, and treats that count as the result. I worked on a recent AI visibility and reputation report that found that in seven out of ten unbranded category answers, the monitored brand was either missing or mentioned, but not the preferred choice. Being in the answer is the start of the job, not the end of it.

Why doesn’t the Google ranking model carry over?

Most of us learned search through Google, where the aim was a position on page one. Google does adjust results, and its own documentation says it uses your location, search history and settings to do it. But the personalisation works on signals Google infers about you.

An AI assistant works from what you tell it about yourself. Add a sector, a size, a location or a constraint, and the shortlist changes. Different platforms also lean on different sources, so there is no single ranking to climb.

That makes the generic category question, the one most firms benchmark themselves against, a strange thing to measure. It is the question almost nobody with a real need actually asks.

What happens with additional user context?

I tested this in October 2026, asking ChatGPT the same need two ways, twice each.

The first was the generic question: “What are the best employment law firms in the UK?” Both answers were mostly large City firms. Almost every entry was backed by a link to the Chambers and Partners or Legal 500 rankings. One answer ended by asking who needed the solicitor, what the issue was and where they were based, so it could give “a much more useful shortlist”. The assistant itself knew the generic answer was not the useful one.

The second came from a specific buyer: the owner of a UK hospitality group with seasonal and zero-hours staff across 12 sites, asking which employment law firms actually understand businesses like theirs. Both answers named five or six regional and mid-sized firms. Each one was cited from the firm’s own hospitality page, the kind that names the partner who leads the work and lists the cases they handle, from restaurant chains to minimum wage investigations. No directory was cited.

No firm appeared in both answers, in any run. I ran the same test for accountancy. The generic question returned the Big Four and the next tier down, backed by professional bodies and trade press league tables. A finance director at a 60-person architecture practice got four mid-sized firms, each cited from its own site. Again, no overlap.

This is a small test, and AI answers change from one run to the next: about half the names changed between two runs of the same question. The individual names are not the finding. The pattern is: the generic question and the specific question are different contests, won with different evidence.

It matches what we’re seeing happen in real buying decisions. Four businesses have found Midnight through AI search.

A law firm found us using ChatGPT. They were looking for a PR agency based outside London with experience of working with law firms, and our culture and values were a core part of their decision.

An insurance broker found us in Claude after asking a long, specific question about getting into the insurance trade press.

An HR tech company found us through Copilot, and a B Corp business found us in ChatGPT.

One prospect said to Flo “I found Midnight using Copilot and it’s one of the reasons we reached out to you as you clearly have your own GEO sorted!”

None of them asked for “the best PR agency in the UK”.

What does this mean for law and accountancy firms?

Professional services firms sell expertise that looks interchangeable from the outside. Every firm says it is client-focused and commercially minded, and an AI assistant reading those pages has nothing to tell them apart.

The generic question will keep going to the firms with the largest footprint: the most brand mentions, the most coverage, the most awards, the most authority. That won’t change, and the people asking it are rarely close to buying. They are browsing.

The narrow question is different. The person asking it has a specific problem, a sector, a size and a constraint, and they are usually much closer to a decision. That question is winnable by a mid-sized professional services firm if the evidence exists to answer it.

What can you do about it?

Write down the question your best client would actually ask. Use their words, not yours. “Employment lawyers who understand hospitality businesses with seasonal staff” is a question. “Best employment lawyers” is a category. Then ask it yourself, in two or three AI assistants, and see who comes back.

Get in the cited listings, and know what they do. In the test, Chambers and the Legal 500 shaped every generic answer and were not cited once in the specific ones. These lists give you authority and visibility, and that matters. If you’re on a top list, say so on your own site, but don’t expect it to win the specific questions for you.

Build the page the specific answer is built from. The firms that won the hospitality question had a sector page that named the partner, the type of client and the issues they handle. Count the named proof points on your own sector pages. “A leading firm” is meaningless without evidence to back it up. Specific examples are helpful. Something like: “we advised 14 multi-site operators on holiday pay last year” is a fact a machine can repeat. Earned coverage in the trade press your buyers read adds third-party validation, which is where digital PR comes in. What others say about you is more important than what you say about yourself.

Get your differentiator out of the pitch deck. If the real reason clients choose you lives in a partner’s head or a pitch deck, it does not exist as far as an AI assistant is concerned. Write it down, on the page that answers the question. We don’t pitch Midnight as the best B2B PR agency. Our homepage says it is the PR agency for the responsible economy, building the reputations of purpose-driven businesses in insurance and risk, professional advisory and HR. That is a narrower claim. Three of the four buyers above came from exactly those sectors, and the fourth was a B Corp.

Ask to see AI questions and answers with a client’s context in them. If your agency or tool reports your visibility on the broadest term in your market, that tells you whether you have one of the strongest digital footprints. It does not tell you whether you are recommended to the buyers you want. Ask for both.

So what should you be asking?

Don’t stop at “are we visible?”. Ask “are we the obvious choice for the person we most want to work with?”

Do you know which question your best client is asking? If you’d like help finding it, and making sure the answer points to you, start a conversation with Midnight.


More from us on AI search

How to get your brand mentioned in AI answers (not just ranked in search results)

Rik Turner

Rik Turner is the founder of PR for AI, helping PR teams earn AI search visibility. He specialises in content strategy, AI search optimisation (AI SEO), and facilitating PR training workshops. With over 16 years of experience in digital marketing, Rik has worked with leading B2B and B2C brands like M&S, Waitrose, Hotjar, Yell, and giffgaff, delivering measurable business impact across multiple industries.