Ask ChatGPT to recommend a supplier in your sector. Then check whether your name comes up. For most UK SMEs it does not, and the reason is rarely that the competitor is better. It is that the competitor is easier for the model to read, verify and trust. So how does ChatGPT choose sources, and how does that translate into which businesses it names? Nobody outside OpenAI can give you the full formula, and anyone claiming otherwise is selling certainty that does not exist. What we can do is describe the mechanism, the signals that consistently correlate with being cited, and what to fix first.
Key takeaways
- ChatGPT does not hold a ranked list of businesses. It retrieves pages when you ask, reads them, and synthesises an answer.
- Being retrievable beats being clever. Clear, self-contained answers get quoted; buried ones do not.
- Entity consistency across your site, listings and profiles removes the doubt that costs you the mention.
- Third-party corroboration carries more weight than anything you say about yourself.
- There is no permanent position. The realistic goal is durable presence in the pool of pages likely to be pulled.
- No agency can guarantee placement. Treat anyone who does with suspicion.
How does ChatGPT choose sources when recommending a business?
ChatGPT does not hold a ranked list of businesses. When you ask, it runs its own searches, pulls a handful of pages, reads them, and writes an answer from what it finds. The businesses named are the ones whose information was easiest to retrieve, easiest to understand and corroborated elsewhere.
Two mechanisms are at work. Without browsing, the model answers from training data, which is older and vaguer about small companies. With browsing, it searches live, then decides which of those sources deserve a citation. For any competitive commercial query, the second mechanism is the one that matters.
That has a practical consequence. You are not competing for a fixed position you can hold for a year. You are competing to be in the set of pages the model is likely to pull on the day someone asks. Retrieval is dynamic, so a page cited this week may not be cited next week.
If the whole discipline is new to you, start with our beginner’s guide to GEO and AEO before going further.
Why does content structure decide what gets quoted?
Because the model lifts passages, it does not index pages. If your answer to a buyer’s question sits in a clean, self-contained paragraph directly under a clear heading, it can be quoted whole. If the same answer is scattered across three sections, a slider and a downloadable PDF, it usually cannot be used at all.
Front-load the substance. Put your credentials, pricing model, coverage area and core proposition high on the page rather than after 400 words of scene-setting. Write headings as the questions your buyers actually ask, then answer them immediately underneath.
Tables, short lists and direct definitions all help, because they survive being extracted from their surroundings. Google’s own guidance on creating helpful, people-first content points the same way, which is convenient: the structural work pays off in classic search and AI answers at once.
There is more detail in our breakdown of content structures that get picked up by generative AI engines.
Why does entity consistency matter so much?
Because the model has to be satisfied that the company on your website, the one on Companies House, the one on LinkedIn and the one in a trade directory are the same organisation. Conflicting names, addresses, service descriptions and founding dates create doubt, and doubt costs you the mention.
Get the basics identical everywhere: legal name, trading name, address, phone number, and a one-line description of what you do and who for. Then support it with LocalBusiness structured data so the information is machine-readable rather than merely present.
Corroboration is the other half. Reviews, industry press, directory entries, podcast appearances and client testimonials published somewhere other than your own site all tell the model that your claims hold up independently. A small specialist with genuine third-party evidence often gets named ahead of a larger generalist with none.
“In seven years marketing facilities management and property firms, prospects arrived having already built a shortlist from directories, referrals and Google. The shortlist is now increasingly drawn up inside a chatbot, and the same fragmented company information that used to cost you a tender slot now costs you the recommendation.” – Marcus, Founder of Lobo Digital
What should a UK SME fix first?
Fix retrievability before you fix anything else. Imagine a 12-person commercial cleaning firm in Croydon: it has a decent website, no reviews outside Google, and a services page that never states which boroughs it covers. It is invisible to the model for the exact query its buyers are typing.
Work in this order.
| Priority | What to do | Commercial payoff |
|---|---|---|
| 1 | Standardise name, address, phone and service description across every profile you own | Removes ambiguity that stops you being matched to a query |
| 2 | Add a plain-English answer to each buyer question, directly under a matching heading | Makes your page quotable, not just readable |
| 3 | Publish coverage area, sectors served and pricing model in text, not images | Lets the model qualify you against specific asks |
| 4 | Earn reviews and mentions on sites you do not control | Supplies the corroboration that supports the recommendation |
| 5 | Implement LocalBusiness and Organization schema | Turns your details into structured, verifiable data |
| 6 | Refresh key pages annually and date them | Keeps you competitive on time-sensitive queries |
None of this is quick, and none of it is a switch you flip once. It compounds, which is why we treat it as an ongoing programme inside our GEO and AEO services rather than a one-off audit.
Frequently asked questions
Does ChatGPT use live search?
Yes, when browsing is active. It runs live searches, retrieves pages and cites what it used. Without browsing, it answers from training data, which is older and far less reliable on individual companies. Most commercial recommendation queries trigger browsing, which is why current, retrievable pages matter more than historical ranking history.
Can I influence which businesses get mentioned?
You can influence the odds, not the outcome. There is no paid placement and no guaranteed slot. What you control is how clear, consistent, structured and independently corroborated your information is. Do that well and you enter the pool of pages likely to be retrieved. You cannot suppress a competitor, only strengthen your own case.
Is this the same across all AI chatbots?
No. Perplexity runs its own index and cites aggressively. Google’s AI features lean on pages already performing in Google search, as its documentation on AI features sets out. Gemini and Claude weigh things differently again. The underlying hygiene transfers, but the emphasis shifts. Our engine-by-engine comparison covers where they diverge.
Why is my business mentioned without a link to my site?
Usually because the model learned about you from somewhere else: a directory, a review platform, a press mention. It cites the source it read, not the company it named. The fix is to make your own pages the strongest and clearest version of that information, so there is a reason to cite you directly.
How long does this take to show results?
Expect months rather than weeks, and expect movement rather than a fixed position. Entity clean-up can register quickly because it is mechanical. Corroboration takes longer, since reviews and mentions accumulate at the pace of real client work. Track mentions across engines rather than watching for a single ranking number that does not exist.
Does this apply to service sectors like facilities management?
Yes, and often more sharply, because buyers use AI to shortlist before they ever request a tender. Sectors with complex service definitions benefit most from clear entity data. We cover the specifics in our guide to GEO for facilities management companies.
Get recommended, not just ranked
The mechanism is not magic and it is not fully public. What it rewards is unglamorous: information a machine can read without guessing, claims other people back up, and answers written where a model can lift them. Most UK SMEs have not done that work yet, which is precisely why there is ground to take now.
Learn about our GEO service →
