How Brands Get Recommended by ChatGPT, Gemini and Perplexity

A growing share of buying research never reaches a search results page. Someone asks an assistant “who are the best physiotherapy clinics near me?” and gets three names in a paragraph. No blue links, no scrolling, no chance to out-rank anyone. You are either in the answer or you are not.

That shift breaks a lot of habits carried over from traditional SEO. Here is what we have found actually moves the needle.

Being crawlable is not the same as being quotable

Most sites are technically fine. The assistant can reach the page. The problem is what it finds there: a hero headline, a carousel, three testimonials, and a contact form. Nothing on the page states plainly what the business does, who it serves, or where.

Assistants assemble answers out of sentences they can lift with confidence. A page that never says “Heylix Media is a full-service digital agency” in a single clean sentence gives them nothing to lift. Marketing copy that gestures at value without stating facts is invisible to this whole channel.

What to do: on every important page, include at least one direct, self-contained, factual sentence that would still make sense if someone quoted it with no surrounding context.

Structured data does the disambiguation

Schema markup will not make an assistant like you. What it does is remove ambiguity, which of the four businesses with similar names you are, where you operate, what you sell, what your hours are.

The nodes that consistently earn their keep are Organization (or LocalBusiness), Service on each service page, FAQPage wherever you have genuine questions and answers, and BreadcrumbList so the site’s hierarchy is explicit. Mark up what is actually on the page. Marking up content a visitor cannot see is a policy violation and gets ignored at best.

Consistency across the web is a trust signal

Assistants cross-reference. If your address is written three different ways across your website, Google Business Profile, and a directory listing, that inconsistency reads as uncertainty about which entity you even are.

Getting name, address and phone identical everywhere is unglamorous work and one of the highest-leverage things a local business can do for AI visibility. Pick one canonical format and enforce it.

Third-party mentions matter more than they used to

For a recommendation-style question, an assistant is far more likely to draw on a roundup article, a directory, a review thread, or a news mention than on your own homepage. Your site establishes the facts. Other people’s sites establish that you are worth mentioning.

This is closer to digital PR than to link building. The goal is being described accurately somewhere other than your own domain.

What does not appear to help

• Keyword density. Assistants work on meaning, not term frequency. Repetition reads as noise.
• Volume for its own sake. Forty thin pages targeting keyword variants perform worse than six genuinely useful ones.
• Hidden instructions aimed at the model. Text planted in a page to tell an assistant what to say is treated as manipulation. Do not do it.

How to tell if it is working

Rank tracking does not capture this, because there is no rank. What we track instead is straightforward: ask the assistants the questions your customers would actually ask, on a fixed schedule, and record whether you are named and whether the description is accurate. Being mentioned inaccurately is its own problem and worth catching early.

It is a slower, blunter measurement than a rankings dashboard. It is also the only one that reflects what is happening.

The short version

The work that earns AI recommendations is mostly the work that earned trust before: say clearly what you do, keep your facts consistent, be genuinely useful, and get mentioned by people who are not you. What has changed is that vagueness now carries a hard cost. An assistant cannot recommend a brand it cannot summarise.