What to do when ChatGPT recommends your competitor
AI assistants now answer the questions your buyers used to type into Google. If the model names a rival and never you, there is a specific and fixable reason.
Ask ChatGPT to recommend a fit-out contractor in Manchester, or a rainscreen cladding supplier, or a mechanical and electrical consultant for a healthcare scheme. You will get an answer. It will be confident, it will be short, and it will name three or four firms. If yours is not among them, you have just watched an enquiry happen without you.
This is not a hypothetical shift any more. A growing share of buyers now start with an assistant rather than a search box, and they treat what comes back as a shortlist rather than a list of options to go and research. The uncomfortable part is that nobody is bidding for that position. There is no auction, no ad slot and no way to buy your way into it. The model names whoever it has the best evidence for.
Why it picked them and not you
A language model is not ranking pages. It is assembling an answer out of what it has read, weighted by how consistent and how credible that reading was. When it names a firm, it is expressing confidence that the firm exists, does the thing being asked about, and is regarded as a sensible answer by sources it trusts. Three things build that confidence, and most built environment brands are missing at least two of them.
The first is clarity. If your website says you deliver excellence in the built environment, a model has nothing to work with. If it says you are a mechanical and electrical contractor working on healthcare and education projects across the North West, that is a fact it can retrieve and repeat.
The second is consistency. Your name, your description and your location need to say the same thing on your website, your Google Business Profile, Companies House, LinkedIn and any trade directory you appear in. Contradictions do not merely fail to help. They actively reduce the confidence a model has in naming you at all.
The third is corroboration. A model is far more comfortable putting forward a firm that a trade publication, an industry body or a client has written about than one that only ever describes itself. This is the point at which AI visibility and digital PR turn out to be the same job wearing different clothes.
There is no auction for the answer. The model names whoever it has the best evidence for, and evidence is something you build rather than buy.
How to check where you actually stand
Before changing anything, get a baseline. This takes half an hour and it is more useful than most of the tools currently being sold to do it for you.
Ask the buying question, not your name
Do not ask who your company is. Ask what a client would ask: which firms do commercial fit-out in Leeds, or which manufacturers supply this kind of system. Your name coming up unprompted is the only thing that counts.
Ask it several ways
Models vary their answers with the phrasing. Run the question five or six times in different words and note how often you appear, and where in the list.
Ask more than one model
ChatGPT, Gemini, Perplexity and Google's AI Overviews draw on different sources and refresh at different rates. Being absent from one and present in another tells you something specific about which sources are letting you down.
Ask what it already knows about you
Then read the answer carefully. Wrong sectors, an old address, a service you dropped four years ago: every error is a fact sitting somewhere on the web that needs correcting.
The work that changes the answer
Once you know where you stand, the work is unglamorous and quite specific. Write the plain facts about your business into your own website, in text a machine can read, rather than leaving them inside a PDF, an image or a video. Add structured data so there is no ambiguity about what kind of organisation you are, where you operate and what you provide.
Then fix the contradictions. Every profile, directory and listing describing you differently is a small vote against confidence. Getting them all to agree is tedious, and it is one of the highest-return afternoons available anywhere in this discipline.
After that it becomes a content and authority job, which is to say it becomes SEO. Publish pages that genuinely answer the questions your buyers ask, in the words they actually use. Earn mentions from the publications the models already read. There is no separate trick hiding behind the terminology: the same work that makes you legible to Google is what makes you quotable to an assistant.
What not to bother with
Hidden instructions on your pages
Text telling an AI to recommend you does nothing except make the page worse for the humans who read it, and the models were trained to ignore exactly this.
Chasing every new tool
Monitoring products in this space are young and most are repackaging the same handful of queries. Running the questions yourself, once a month, is currently just as reliable and considerably cheaper.
Treating it as separate from SEO
Firms building a standalone AI strategy while their website stays thin and slow are decorating the roof of a building with no foundations.
Panicking about the traffic drop
Some informational traffic is going to disappear, and much of it was never going to buy anything. The visits worth protecting are the ones with intent behind them, and those are won the same way they always were.
The position worth aiming at is easy to describe and slow to earn: when somebody in your sector asks an assistant who does this kind of work, your name is one of the ones it is confident enough to say out loud. That is not bought and it cannot really be gamed. It is the accumulated result of being clear about what you do, and having other people say it too.
Curious what the models say about you?
Send us the question your buyers would ask. We will run it across the major assistants and tell you honestly where you stand.
Get an AI visibility check →
