A customer opens ChatGPT and types: "I'm in Hertfordshire and need a good aesthetics clinic — who do you recommend?"
ChatGPT answers with three names. Either you're one of them or you're not — and most small business owners have no idea how that list gets made. It isn't random, and it isn't a Google ranking wearing a different hat. Here's what's actually happening, based on what's been measured in 2026.
ChatGPT doesn't rank pages. It verifies businesses.
Google's model is a ranked list of pages. ChatGPT's model is closer to a cautious concierge: it assembles a picture of each candidate business from every source it can reach, and it only names businesses whose picture is complete and consistent enough to stake a recommendation on.
That distinction changes the work. With SEO you optimise pages. With ChatGPT you're optimising the evidence trail about your business — across your site, the indexes, the directories and the review platforms an AI can read.
If no local business crosses its confidence threshold, ChatGPT gives a generic answer — "look for clinics with good reviews and proper qualifications" — and nobody wins. In an area where few businesses have done this work, the first one that does tends to get named again and again. We covered why those AI-referred customers convert so well in our piece on AI search and leads.
Where the evidence actually comes from
1. Bing's index, not Google's
When ChatGPT searches the live web, it searches Bing. A site that's invisible in Bing — not indexed, not verified in Bing Webmaster Tools — is invisible to ChatGPT's web search regardless of how well it ranks on Google. The same applies to Copilot and DuckDuckGo, which also draw on Bing's index. This is the single most common gap we find in audits: businesses that have never once looked at Bing because "nobody uses it". Your customers don't use Bing. The AI they ask does.
2. Location databases for local queries
For "near me" and local recommendation queries, ChatGPT calls external location data rather than guessing from text. One 2026 analysis found the large majority of local results in ChatGPT traced back to Foursquare's database, alongside review platforms. Most UK business owners have never claimed a Foursquare listing. The broader point: AI engines lean on structured location databases, so being present and accurate in them — Bing Places, Foursquare, Apple Maps, your Google Business Profile — is no longer optional housekeeping.
3. Google Shopping feeds for product queries
If you sell products, the pipeline is different again. A March 2026 Semrush analysis of 43,000 products shown in ChatGPT's shopping results found 83% matched Google Shopping's top 40 organic listings — 60% from the top ten. Your Google Merchant Center feed quietly became an AI visibility input. OpenAI has also opened a merchant program for submitting product feeds directly. For UK e-commerce businesses, feed quality is now doing double duty.
4. Reviews — but only the detailed ones
AI engines use reviews as evidence, and the useful unit isn't the star count — it's the sentence. "Five stars, great work" tells a model almost nothing. "They rebuilt our booking flow and answered our Instagram DMs within the hour, and we're in Bishop's Stortford" gives it a service, a behaviour and a place to match against the next question. When you ask happy customers for reviews, ask them to mention what you did and where you are.
5. Consistency across everything
The model cross-references. A business whose name, address, phone number, services and prices match everywhere — site, schema, GBP, Bing Places, directories — reads as verifiable. One whose details contradict each other reads as risky, and risky doesn't get recommended. Inconsistency doesn't just dilute your visibility; it actively lowers the model's confidence.
What's rising in weight: recency and momentum
Analyses through 2026 point the same direction: AI engines increasingly favour businesses with recent corroboration — fresh mentions, updated content, new reviews — over businesses whose evidence trail went quiet two years ago. Some practitioners call this citation velocity. You don't need a PR machine; you need a steady drip. One article a month, a few new detailed reviews, the occasional industry mention. Steady beats sporadic.
What to do this month
- Verify your site in Bing Webmaster Tools and claim Bing Places. An hour of work, and it's the gateway to ChatGPT's live search. You can import straight from Google Search Console.
- Claim your location listings. Foursquare, Apple Maps, and make sure your Google Business Profile is complete. Identical details everywhere.
- Change how you ask for reviews. "Would you mind mentioning the treatment and the town?" turns a star rating into evidence an AI can quote.
- Ship the machine-readable layer. Schema (Organization, LocalBusiness, Service with prices, FAQPage) and an llms.txt file. This is the difference between an AI reading your site and an AI skipping it — the full list of blockers is in 7 reasons ChatGPT isn't citing you.
- Ask ChatGPT about yourself. Fresh chat, your customers' actual questions. The answer is your baseline — and if it's wrong or empty, you now know exactly what the engines can't see.
None of this is magic, and anyone selling guaranteed ChatGPT placement is selling something that doesn't exist. What exists is a checklist, applied patiently, that makes your business the one the model can verify — and the verifiable business is the one that gets named. If you'd rather hand it off, that's our SEO + GEO retainer: Foundations at £600/month, Growth at £1,200/month, every engagement starting with a free audit that shows you exactly where you stand today.
