How AI Assistants Decide Which Brands to Recommend

When someone asks ChatGPT “what's the best CRM for a small law firm?” or asks Perplexity to “recommend an accountant in Toronto,” a short list of names comes back — and most business owners have no idea why their brand is, or isn't, on it. There's no public ranking to check and no dashboard to log into. But the process is not random, and it is not a black box you can do nothing about. Understanding the few mechanisms that actually drive these recommendations is the first step to influencing them.

The short version: an AI assistant recommends the brands it has seen described, by many independent sources, as a clear fit for the specific thing the user asked about — and, for assistants that search the live web, the brands whose pages it can retrieve and cite at answer time. Presence, consistency, and corroboration matter far more than clever copy on your own homepage.

There isn't one mechanism — there are two

It helps to separate two different ways an assistant can produce an answer, because they're influenced by different things.

1. From memory (parametric knowledge). The model answers from patterns learned during training. Here, a brand surfaces because it appeared often enough, and in the right context, across the public text the model was trained on. You can't edit this directly, and it updates slowly — only when the model is retrained or refreshed.
2. From live retrieval (search and grounding). Assistants such as Perplexity, ChatGPT with search, and Gemini fetch web pages at the moment you ask, then summarize and cite them. Here, a brand surfaces because a page that clearly answers the query was retrievable, relevant, and citable right then. This pathway updates fast — new or improved content can show up within days or weeks.

Most real answers blend the two: the model leans on what it “knows” and reinforces or corrects it with what it just retrieved. That's why the practical playbook is to influence both — the durable, slow training signal and the fast, editable retrieval signal.

The signals that actually move a recommendation

Across both pathways, the same handful of factors keep deciding who gets named.

1. Category association, not just name recognition

Being “known” isn't enough; you get recommended when the model strongly associates your brand with the specific intent behind the question. A bakery that's mentioned all over the web for its storefront but never described as doing “custom wedding cakes” won't surface for that query. The goal is to be unmistakably tied to the exact problem, audience, and use-case your customers ask about.

2. Independent corroboration

Models lean toward entities that multiple independent sources agree on. Your own website saying you're the best is a weak signal; third-party roundups, comparison articles, directories, and reviews that describe you the same way are a strong one. Recommendations cluster around consensus, so the brands that appear in the “best X for Y” lists other people write tend to be the ones assistants repeat.

3. Retrievability and extractable answers

For the live-search pathway, a page only helps if it can be fetched and the relevant fact pulled out cleanly. Pages that directly answer a specific question — in plain language, with the entity, audience, pricing, and location stated explicitly — get cited more readily than pages that bury the same facts in marketing prose or trap them behind scripts the crawler can't read.

4. Structured data and a clear entity definition

Schema markup, consistent business details, and a crisp “who we are / who we serve” statement help a model resolve what your brand is without guessing. This won't force a recommendation, but it removes ambiguity that otherwise gets you left out or mis-described. (Related: whether you also need an llms.txt file for this.)

5. Consistency across the web

When your name, category, location, and positioning are described the same way everywhere, the model forms a confident, stable picture. When sources disagree — different service descriptions, outdated locations, conflicting claims — confidence drops and the safer move for the assistant is to recommend someone clearer.

6. Reviews and sentiment presence

Visible, specific, credible third-party reviews give the model both evidence you exist and language to describe your strengths. Their absence is a quiet reason a brand gets skipped in favor of a competitor whose reputation is legible on the open web.

What you can influence — and what you can't

It's worth being honest about the boundary, because plenty of “AI SEO” pitches blur it.

You can influence: how consistently and clearly you're described across independent sources; whether you earn third-party mentions and comparisons in your category; whether your pages directly and extractably answer the buying-intent questions customers actually ask; your structured data and entity clarity; and how current and consistent your information is everywhere it appears.

You can't control: what's already baked into a model's training data, the model's internal weighting, or whether any single answer names you. No honest provider can guarantee an AI will recommend you, and there is no paid placement that buys your way into an organic recommendation. Anyone promising guaranteed AI placement is selling something the mechanism doesn't support.

How to find out where you stand today

Before changing anything, it's worth seeing how the assistants currently describe you — whether they mention you at all for your key buying queries, which competitors they name instead, and how accurate their description of you is. That diagnosis is the difference between guessing and knowing which of the six signals above is actually your gap. (For the competitor side specifically, see how to run a competitor AI visibility analysis, and for the tactical follow-through, how to get AI to recommend you over competitors.)

Related reading: How to get AI to recommend you over competitors · How to run a competitor AI visibility analysis · Do you need an llms.txt file?

Run a law firm? See AI visibility for law firms.