ChatGPT and Gemini do not keep a private directory of approved businesses. When someone asks for a recommendation, the assistant either recalls what it learned from the public web during training or runs a live search in the moment, then names the businesses it can describe with confidence from sources it treats as reliable: search results, maps listings, review platforms, and each business's own website. If those sources say little about you, or contradict each other, the assistant quietly skips you and recommends a competitor it can verify.
That is the whole mechanism in one paragraph. The rest of this playbook unpacks it, because once you see how the choice is made, the fix becomes obvious.
The two ways an assistant answers
When someone asks ChatGPT or Gemini for "a good dental clinic in Limassol", one of two things happens. Either the model answers from memory, drawing on the snapshot of the public web it absorbed during training, or it runs a live web search and reads the results before replying. Questions about local businesses increasingly trigger the second path, because the assistant knows its memory goes stale.
Both paths lead to the same place: the open web. Training memory is the web as it was months ago. Live retrieval is the web as it is today. Either way, the assistant never inspects your business. It reads what the internet says about your business and summarises it.
What the assistant actually reads
In our experience, the sources behind AI recommendations are the same ones a careful human researcher would use:
- Search results. Live retrieval usually starts with a search query, so businesses that rank for "service plus town" get read first.
- Maps and business profiles. Your Google Business Profile supplies the facts assistants like to cite: address, hours, phone, services, reviews.
- Review platforms. Reviews give the assistant something no brochure can: third-party evidence, with dates and detail.
- Your own website. Assistants quote pages that answer questions in plain language: what you do, where, for whom, and roughly at what price.
- Directories, press, and local mentions. Each independent mention corroborates the others.
The signals that tip the choice
Assistants are cautious recommenders. Naming the wrong business is a bad outcome for them, so they favour businesses they can describe with confidence. Three things build that confidence:
- Consistency. The same name, phone, address and services everywhere they look. Conflicting details make a business risky to recommend, so it gets left out.
- Specificity. A page that says "we repair accident damage on all makes, in Nicosia, free estimate" hands the model a sentence it can reuse. A page that says "we deliver excellence" hands it nothing.
- Third-party confirmation. Reviews, mentions and listings the business does not control. Anyone can praise themselves; assistants weigh what others say more heavily.
Language matters too in Cyprus. Assistants tend to answer in the language of the question and lean on content written in that language. A business with clear pages in both Greek and English is easier to recommend to both audiences.
Why quiet businesses get skipped
Notice what is missing from that list: the quality of your actual work. The assistant cannot visit your workshop or taste your food. If your web presence is thin, dated or contradictory, the model has nothing safe to say, so it says nothing about you and fills the answer with businesses it can verify. To a machine, being excellent but invisible looks exactly like not existing.
A quick test: ask ChatGPT and Gemini for your own service in your own town, in Greek and in English. If you are absent, or described wrongly, the leak is showing itself. We keep a simple walkthrough in how to check if you are invisible to AI.
What to do about it
The playbook follows the mechanism: give assistants clean sources, specific answers, and independent confirmation.
- Fix the record. One consistent name, address and phone across your site, Google Business Profile and every directory.
- Publish answerable pages. Plain-language pages for each service and area, with real details and prices where you can commit to them.
- Earn reviews steadily. Recent, detailed reviews are the third-party evidence assistants trust most.
- Structure your site for machines. Clear headings, FAQ sections and schema markup make your pages easy to quote.
This work has a name, answer engine optimisation (AEO). It overlaps with classic SEO but is not the same discipline. The step-by-step version lives in how to get recommended by ChatGPT, and we do the work itself as a service: AEO for Cyprus businesses.
We test all of this on Pinelaki, our own 35-year auto-body business in Nicosia, before recommending any of it to anyone else. Nothing on this page is theory we have not run ourselves.
If you want to know where you stand today, start with a free diagnosis. We ask the assistants about your business, check the sources they read, and show you exactly what they see.