How to Get Your Business Mentioned by ChatGPT, Gemini and Perplexity

A growing share of buying decisions now starts with a question typed into an assistant rather than a search box. Someone asks ChatGPT for a dentist in their city, asks Gemini which platform to use for their restaurant menu, asks Perplexity who builds web applications. The assistant answers in a paragraph, sometimes with links, and the businesses it names get the enquiry. Everyone else is invisible, however well they rank in a traditional list.

The good news is that this is not mysterious, and it is not a separate discipline bolted onto SEO. The signals that make a page easy for an assistant to quote are largely the signals that made a page good in the first place. What has changed is which of them matter most, and how unforgiving the process is about facts.

How an assistant actually builds an answer

It helps to know the mechanics, because they explain every recommendation that follows. When a question needs current or local information, an assistant does not recall it from training. It runs one or more searches, fetches a handful of pages live, reads them, and writes an answer grounded in what it found. The models differ in detail, but that retrieve-then-answer shape is common to ChatGPT search, Gemini, Perplexity and Claude.

Two consequences follow immediately. First, your page has to be reachable by the specific agent doing the fetching, which is not the same agent as the one that crawls for training. Second, the text it finds has to be usable as an answer with minimal editing. The assistant is looking for a sentence it can quote, not a brochure it must interpret.

Let the right robots in

This is the failure that silently disqualifies a site, and it is the easiest to fix. Most robots.txt advice concerns crawlers that collect training data. The agents that produce citations are different: they fetch a page at the moment a user asks a question. If they are blocked, the site is not in the answer.

Check that your robots.txt permits the retrieval agents by name as well as the training crawlers. Blocking training while allowing retrieval is a legitimate position; blocking retrieval while chasing AI visibility is simply working against yourself.

• Retrieval agents that produce citations: OAI-SearchBot and ChatGPT-User, PerplexityBot and Perplexity-User, Claude-SearchBot and Claude-User, Google-Extended
• Training crawlers, a separate decision: GPTBot, ClaudeBot, CCBot
• Verify by loading yourdomain.com/robots.txt and reading it, rather than assuming the platform default is sensible

Be consistent everywhere you appear

An assistant assembling an answer about a business is, in effect, checking whether independent sources agree. Your website says one thing, your map listing says another, a directory has an old phone number: the assistant has no way to decide which is right, so it hedges, or it names a competitor whose details line up cleanly.

Consistency is unglamorous work with outsized returns. The exact business name, the same address format, one phone number, the same short description, the same opening hours, everywhere they appear. Where you have profiles on other platforms, link them from your site in structured data so the connection between them is explicit rather than inferred.

Write the answer, not the pitch

The single most useful change to most websites is structural. Take the questions your customers genuinely ask, make each one a heading in the words they would use, and put a direct, complete answer in the first sentence below it. Add the detail afterwards for the humans who want it.

This is why a well-made FAQ outperforms a polished paragraph of positioning. The FAQ maps a question to an answer, which is exactly the shape an assistant needs. Marketing prose that builds to a conclusion three sentences in gives the assistant nothing clean to lift.

The same logic applies to what you claim. Specific, checkable facts get quoted. Adjectives do not, and superlatives actively hurt: an assistant weighing several sources tends to discount the one asserting it is the best and cite the one stating what it does and for whom.

Make the facts machine-readable

Structured data is how you state, unambiguously, what your page is about. It removes the guesswork: this is an organisation, this is its address, these are its services, this is an article published on this date, this is a question and its answer. Schema.org markup in JSON-LD is read by search engines and by the systems that feed assistants.

Two rules matter more than volume. Keep one stable identifier for your organisation and reference it from every other entity, so all your pages describe one business rather than several unconnected ones. And never let the markup say something the visible page does not: if the FAQ schema and the visible FAQ disagree, you have created exactly the inconsistency you are trying to eliminate.

• Organization with address, contact details and links to your other profiles
• Service or Product entities for what you actually sell
• FAQPage whose text matches the questions and answers a visitor sees
• Article with a real publication date on anything written
• One organisation identifier, reused by every page

Speed, structure and languages still count

Retrieval happens under a time limit. A page that takes several seconds to deliver its content may simply be passed over in favour of one that answered immediately. The performance work that helps human visitors on a phone helps here too, for the same reason.

If you serve more than one language, give each language a real page at its own address rather than a script that swaps the text after loading, and declare the relationship between the versions. An assistant answering in Macedonian should find the Macedonian page, not an English page it has to translate and might misquote.

What nobody can promise

No agency controls what an assistant says. There is no submission form, no ranking factor to buy, and the answers vary between models and change as they are updated. Any offer to guarantee that ChatGPT will recommend you is a warning sign, in the same way that guaranteed Google rankings always were.

What can be engineered is every condition that makes a citation possible: reachable by the right agents, consistent everywhere, structured so a machine can read it, written so an answer can be lifted from it, fast enough to be used, and honest enough to survive checking. Do that work and you become the easiest source to cite on your subject, which is the closest thing to a guarantee that exists.

How to check where you stand

Measurement here is less developed than in search, but it is not guesswork. Ask the assistants the questions your customers would ask and record what comes back, by name and by date, so you can see movement rather than relying on impressions.

• Ask each assistant the three or four questions a customer would ask, and note whether you are named and what it says about you
• Repeat monthly, and after any significant change to your site, so you can attribute movement
• Fetch your own robots.txt and confirm the retrieval agents are allowed
• Check that your name, address, phone and description match everywhere they appear
• Correct anything an assistant states about you that is wrong, at the source it is reading