How ChatGPT, Gemini, Claude and Perplexity Decide Which Business to Recommend

Try this. Open ChatGPT and type “Recommend a [your service] company in [your city]”. Then ask Gemini, Claude and Perplexity the same thing. You will probably get four different lists. The question is whether you are on any of them.

I have been doing SEO since 2004 and have lived through a few rule changes: meta keywords, then links, then content, then mobile and speed. This shift is different. You are no longer competing for a spot in a list of ten blue links. You are competing to be mentioned in the answer.

Two ways an AI “knows” about you

Simplified, an AI assistant answers in one of two ways:

1. From memory. The model was trained on a vast amount of text. If your name appears often and consistently, it has a sense of who you are. If not, it either leaves you out or confuses you with someone else.
2. From live search. ChatGPT, Gemini, Claude and Perplexity can search the web, read a handful of pages and assemble an answer with sources.

Memory is built slowly, over years of presence. Live search is something you can improve this month.

Where each assistant looks

There is a practical difference here that many businesses miss:

• Gemini builds on Google. Strong Google visibility gives you a head start.
• ChatGPT relies heavily on Bing’s index when it searches. If Bing does not know you, ChatGPT will struggle to find you.
• Perplexity runs its own crawler and shows its sources openly, so you can see exactly which pages it uses.
• Claude can also search the web and read the pages it reaches.

The takeaway is simple and widely ignored: verify your site in Bing Webmaster Tools, submit your sitemap and check that your key pages are indexed. Plenty of businesses never look at Bing because its search share is small. It is now, however, the front door to ChatGPT.

The signals that carry weight

1. Entity clarity

AI models think in entities: a person, a company, a product, a place. It must be unambiguous who you are, what you do and where.

An example close to home. There is another Bozhidar Bonchev, a ceramic artist. If I do not help machines tell us apart, an assistant might blend our biographies. The fix is unglamorous but effective: a consistent name, a clear About page, Person structured data and links to my profiles confirming they all describe the same person.

2. Schema markup

Schema.org markup tells machines, in their own vocabulary, “This is an organisation, here is its address, here are its services, here is who runs it”. The most useful types for most businesses:

• Organization or LocalBusiness with name, address, phone and sameAs links;
• Person for founders and key people;
• Service and Product for what you sell;
• FAQPage for common questions.

3. Consistent mentions

If your website says “Dynomica Ltd”, one directory says “Dynomica” and another lists an old address, a machine sees three slightly different things. Pick how you present yourself and be identical everywhere: directories, social profiles, press coverage, partner sites.

What others say about you outweighs what you say about yourself. A trade publication article, a podcast appearance, a listing in a professional directory: each is independent confirmation.

4. Reviews

When someone asks “Which is a good company for…”, the assistant looks for evidence that you are good. Reviews on Google, industry platforms and directories are exactly that. Not just the stars, but the words: a review describing what you actually did carries far more information than “Great!”.

5. Answer-style content

AI assistants assemble answers. If your site contains a clear, short answer to the question a customer is asking, the odds that it gets used go up. In practice:

• headings that are real questions;
• the first sentence under each heading is the answer, with detail after it;
• concrete facts: prices, timelines, who the service is for and who it is not for;
• FAQ sections built from questions customers actually ask, not invented ones.

6. Technical access

The best content is useless if the bot cannot read it. Check that robots.txt is not blocking AI search crawlers, that important text is in the HTML rather than appearing only after JavaScript runs, and that pages load quickly. It sounds basic, yet I regularly find sites that have locked the door themselves without realising it.

What does not work

• Keyword stuffing. Models understand meaning; they do not count words.
• Mass-produced content with no facts. A hundred articles saying the same generic thing do not make you an authority.
• Hiding prices and details. If the information is missing, the assistant will recommend someone who published it.

Where to start

1. Ask all four assistants about your niche and city, and note what they say about you and your competitors.
2. Verify your site in Bing Webmaster Tools and Google Search Console.
3. Add Organization, Person and Service schema.
4. Clean up your name and address across every directory.
5. Rewrite key pages so they answer customer questions directly.
6. Ask happy clients for specific, detailed reviews.

None of this is magic, and none of it guarantees a recommendation tomorrow. But every step makes you easier for a machine to understand. And a machine that does not understand you will not recommend you.

How I work on this

My AI search optimization service combines classic SEO with these newer signals: entities, schema, mentions, content and Bing. If the Google fundamentals are still shaky, we start with SEO, because AI search is built on the same foundation.

The good news: what helps an AI recommend you is what helps a person choose you. Clarity, evidence and specifics.

If you want to find out what AI assistants say about your business today, get in touch.