What actually makes AI recommend your firm?
There is no secret switch that puts your firm into an AI answer. But there are clear reasons one firm is easier to recommend than another.
A prospective client asks ChatGPT to recommend a firm like yours. A moment later, three names pop up on their screen.
Why those three, and not you?
I’ve written before about why a firm that ranks well in Google can still be absent from AI answers. This blog post takes a look at the problem from the other side. Instead of diving into a critique of your web pages, it shows you what an AI assistant needs before it’s willing to name your business.
Once you know that, it’ll help you decide on fixes that aren’t based on guesswork.
But before we begin, I want to lay my cards on the table. Nobody outside the companies that build these AI systems knows in detail how every recommendation is assembled and ranked.
Plus, different AI assistants work in different ways. ChatGPT works differently from Gemini or Claude. In turn, the different models of each work in distinct ways – and new models are appearing all the time.
So what follows is not a set of formulae. Rather, it’s a practical account of the factors that repeatedly make a firm easier to find, understand and recommend via AI.
In short, it’s meant to be a helpful guide, not a workshop manual.
There is no single AI league table
An AI does not keep laminated lists of the UK’s best accountants, lawyers and insurance brokers. Instead, it builds shortlists that it believes will answer the specific questions you ask it.
So if you ask ChatGPT to “recommend an accountant”, you haven’t really given it anything to work with.
Ask it to “recommend an accountant in East Anglia who can help me handle an HMRC Corporation Tax investigation”, though, and it will build a shortlist that it thinks best answers that question.
When I tried exactly that, ChatGPT didn’t return a generic list. It named two specific East Anglian firms – and a named partner at each – as the best fit for that narrow problem.
The same happens for every profession. Ask Claude who should handle a dilapidations fight and it’s likely to put commercial property solicitors straight onto the shortlist. Ask who should negotiate an investment round and those names disappear.
This is what gets lost in talk of “ranking in AI”. There’s no winner’s podium and no fixed position. There are thousands of possible shortlists, each shaped by the client, the problem, the location and the words used to describe them. Your firm appears – if it appears at all – on a new shortlist for every question.
So the goal is not to be recommended to everyone. It’s to be recommended when the question describes the services you offer, the problems you solve and the work you do best.
An AI assistant reaches an answer by two broad routes
The next thing to understand is that “ask an AI” can describe two different processes. They reward two overlapping – but not identical – things.
1. An AI may search for evidence now
Some answers are built with live or recent web search. The assistant runs one or more queries, retrieves pages and uses them to form its response.
- Perplexity says it searches the internet in real time.
- ChatGPT can search automatically when a question might benefit from information on the web.
- Anthropic says Claude can invoke a search tool, process several sources and ground its response in live web content with citations.
- Google says its AI features may run several searches for a single question, then use the pages they find to build an answer.
That means your pages must be available to find and read. Google can only use a page in its AI answers if the page is already eligible to appear in Google Search. ChatGPT needs permission for its search crawler, OAI-SearchBot, to reach your site.
All this only gets your page considered. It doesn’t guarantee a mention. The page still has to be relevant, reliable and useful for the question being asked.
2. An AI may answer from what it has already learnt
Some answers rely mainly on patterns learnt during model training. They are created without consulting your website at the time of being asked.
This doesn’t mean the model contains a neat stored copy of everything ever written about your firm. It doesn’t. Indeed, OpenAI describes training as learning patterns from a mixture of publicly available information, licensed or partnered sources, and material provided or generated by users and trainers.
Here the AI relies on what it has already learnt about your firm. If different sources repeatedly describe you in the same clear way, it has a better chance of understanding what you do. But that knowledge may be out of date, simply because the AI hasn’t checked your website recently.
However, an answer may use both existing knowledge and a fresh web search. So two things matter: your current pages must be easy to find, and your firm must be described consistently across the wider web.
What makes your firm easier to recommend?
If you want an AI to recommend your firm, it needs enough evidence to connect it with a request, and a reason it can give for naming it. Six things make this easier.
1. A specific fit with the question
“We provide high-quality advice to businesses” gives an AI almost nothing to work with.
Compare it with this:
“We help owners of UK architecture practices plan their succession, value their firm and structure a management buyout.”
Now the AI knows who the firm helps, where those clients are, what problem they face and what work the firm can do. It has a clear reason to include that firm when an architect asks for help passing the practice to its management team.
Being specific may make you relevant to fewer questions. It makes you a much stronger answer to the right ones.
2. Independent corroboration
Your website can call the firm experienced, commercial and trusted. Your competitors can do the same. An AI has no reason to believe either you or them.
Claims like these are more credible when other people back them up elsewhere on the web.
For example, a client might praise your work, a professional body could list your accreditation, or a sector publication might discuss your expertise. If several independent sources connect your firm with the same kind of work, an AI has more reason to believe that connection is real and valid.
But the number of mentions isn’t the point. Ten directories repeating the same company description are effectively a single claim that has been copied ten times. One detailed client example, or one credible sector source, may be worth far more. What matters is who is saying it, how specific they are and whether they’re genuinely independent.
3. A reason the answer can give
An AI needs more than a name. It needs a “because”.
A stated niche, a relevant accreditation, a named piece of work, a recognised award, a useful research project or a documented client result gives it one. “Ranked in the Legal 500 for construction disputes”, or “we recovered £2.1m for a late-paying client”, is repeatable, because it’s specific and someone can check it. “A leading firm you can trust” is just a claim.
The strongest proof is concrete enough for a reader to check and relevant enough to help them choose.
4. One unambiguous identity
An AI may find your website, your LinkedIn page and an entry in a professional directory. Before it can use those sources together, it has to be confident that they all describe the same firm.
Make the basic facts match wherever your firm appears: its name, location, services, people and specialisms. Keep your profiles on professional bodies’ websites and reputable directories accurate, and remove old descriptions that no longer reflect your business.
Your website can also carry hidden labels, known as structured data (or ‘Schema‘). These tell search engines things such as “this is the firm’s name”, “this is its address” and “this person wrote the page”, and they can help an AI connect the information correctly. What they can’t do is prove that your firm is experienced, trusted or good at what it does.
The problem for many firms isn’t that the AI can find nothing. It’s that it finds conflicting versions: one description on your website, another on LinkedIn and a third in an old directory profile. If the AI can’t be sure who you are and what you do, it’s less likely to recommend you.
5. Pages that are accessible and easy to interpret
In search mode, an AI needs pages it can crawl and passages that clearly answer the question it has been asked.
So the website basics still matter. Pages should be available without unnecessary technical barriers, with descriptive titles and headings, useful internal links, clear authorship, and important claims stated in the visible copy rather than buried deep in the site or in a brochure.
Plain sentences help. So does putting evidence near the claim it supports. This isn’t because an AI simply copies a sentence into its answer. It’s because clear, well-structured information is easier for both search systems and people to interpret correctly. There’s no special “AI copy” that replaces good web writing.
Google’s own guidance says the usual SEO fundamentals still apply to its AI features.
6. Clear signs of what is still current
Newer is not always better. A ten-year-old case study can still prove that your firm knows how to handle a particular kind of work. A three-year-old guide to a regulation that changed last month may be worse than useless.
If a page covers rules, deadlines, services or people, show when it was last checked and keep the information accurate. Update the content, not just the date. Changing “2023” to “2026” without reviewing the advice simply won’t cut the mustard.
An AI looking for an up-to-date answer needs to know the information can still be trusted.
Test the questions your clients would actually ask
Asking “Who are the best firms?” once won’t tell you much.
Write down five to ten questions a serious prospective client might ask. Make them specific. Include the service, problem, sector or location when those details would matter. For example:
“Which UK advisers can help the partners of an architecture practice plan a management buyout?”
Try the questions in fresh conversations, and use two or three different assistants. Record:
- Which firms are named
- Why each firm is recommended
- Which sources the answer links to
- Whether the information is accurate and current
- Whether your firm appears for the work it genuinely fits
Don’t worry too much about any one answer. Look for a pattern. And if possible, log out of your account before asking an AI anything – you don’t want it to know who’s asking the question.
If your firm never appears, the assistant may be unable to find it or understand what it does. If it finds the firm but gives no clear reason to recommend it, your positioning or proof may be too weak. And if it associates you with the wrong work, old or inconsistent descriptions may be getting in the way.
So, when a competitor keeps appearing with a specific reason attached, follow the sources. Find out what gives the assistant the confidence to name them. Then work out how to give the AI similar confidence in you.
Make your firm easier to understand – and recommend
AI hasn’t created a new content problem. It’s made an old one harder to ignore.
If your website is vague, prospective clients will struggle to see why they should choose you. An AI may find your name without finding a clear reason to recommend it. Both need the same answers: what you do, who it’s for, and why anyone should believe you.
I can help you with this in three main ways:
- A content audit that finds vague claims, conflicting descriptions, outdated pages and buried proof
- Website copywriting that turns those findings into clear service pages
- Blog writing that shows what your firm knows, and creates material worth quoting or linking to
Together, that makes the reputation you’ve earned easier for clients and AI assistants to find and understand.
I can’t manufacture a reputation, or promise you a place in tomorrow’s AI answers. What I can do is make sure a good reputation isn’t hidden behind vague copy, inconsistent descriptions and proof that nobody can find.
If your competitors are being named and you are not, they may not be better than you. They may simply be easier to understand, and easier to justify.
That part you can fix.
Practical questions
A few things people ask
A place for the questions that come up about AI choosing which firms to put forward.
Is there a single AI ranking of firms?
No. There is no fixed league table to climb.
An assistant builds a fresh shortlist for each question, shaped by the service, the problem, the sector and the location someone asks about. Your firm can be the obvious answer to one question and irrelevant to the next.
The goal is not to rank everywhere. It is to be the clear choice for the questions that describe your best work.
Why does AI keep recommending my competitors?
Usually because they are easier to understand, not because they are better.
An assistant names the firm it can describe with confidence. If a competitor states plainly who they help and what they do, and other sources back it up, they give the AI a reason to name them.
If your pages are vaguer, the AI has less to work with – even when your work is the stronger.
Can I pay to be recommended by AI?
No. There is no setting that drops your firm into an answer.
Anyone selling you one is overselling. What actually helps is unglamorous: say clearly who you help and what you do, keep it consistent everywhere you appear, and earn genuine mentions from independent sources.
That improves your chances. It does not buy a result.
Do I need to add schema markup to get recommended?
It helps, but it is not the whole answer.
Structured data can help an AI connect your name, address and services correctly, so it is confident it is reading about one firm. What it cannot do is prove that firm is any good.
It is a label, not evidence. Clear, specific copy backed by real proof does the heavier work.
How do I check whether AI recommends my firm?
Ask it the questions your clients would ask.
Write down five to ten realistic prompts – the service, the problem, the sector, the place – and try them in fresh conversations across two or three assistants. Log out first, so the answer is not shaped by who you are.
Then look for the pattern rather than one result: are you named, and is there a clear reason attached?
Will any of this still be true in a year?
The tools will change. The principles will not.
Models, search features and crawlers move quickly, so nobody can promise a lasting position. But being clear about who you help, consistent across the web and backed by real evidence has worked for as long as there has been search to be found in.
It is the safest thing to invest in, precisely because it does not depend on this month’s technology.
Know a firm wondering why AI never puts it forward?
Send them this before they spend on another "AI visibility" fix.
