How AI Chatbots Decide Which Businesses to Recommend

AI chatbots decide which businesses to recommend based on a combination of what they can find (crawlability and a live web search), what they can extract cleanly (structured data and plain-text facts), and what they can trust (brand mentions and citations from independent sources). No single factor decides it — it’s closer to a chain, where a weak link anywhere in it keeps you out of the answer. Here’s each part of that chain, with sources.

Step one: can the engine find you at all

Most generative engines answer real-time questions by running a live web search first, then writing an answer from what that search returns. If your site isn’t crawlable — blocked in robots.txt for AI crawlers specifically, or so JavaScript-heavy that the actual content never renders for a simple fetch — you can be functionally invisible to this step regardless of how good your business is. This is the most basic gate, and the easiest one to fail without realizing it: a site can look fine in a browser and still be effectively blank to an AI crawler.

Step two: can the engine understand what it found

Once your content is reachable, the next question is whether the specific facts a customer would ask about — service area, pricing range, specialties, hours — are stated plainly as text, or buried in an image, a PDF, or vague marketing copy. Models favor content where the answer to a specific question is stated directly. Structured data helps here too: FAQ schema, LocalBusiness schema and clear headings give both search engines and AI crawlers an explicit, machine-readable version of the same facts, rather than making the model infer them from prose.

Step three: does the engine trust what it found

This is the step most businesses underinvest in, and the data on it is the most concrete part of this whole picture. An Ahrefs analysis of 75,000 brands — extended in a December 2025 follow-up covering ChatGPT, Google AI Mode and AI Overviews — found that unlinked brand mentions correlate with AI citation roughly three times more strongly than backlinks do. The practical read: a claim made only on your own website carries less weight than the same claim echoed across review sites, forums, local press and comparison posts. Generative engines lean on independent corroboration to reduce the risk of repeating pure marketing copy as fact.

Which independent sources matter varies by engine. Perplexity in particular leans heavily on forum content: an analysis by Profound of 10,000 commercial queries found Reddit cited in 46.7% of Perplexity’s responses — far ahead of any other single source. If your category has any presence in relevant subreddits or review threads, that’s a disproportionately influential signal for at least one major engine, even if it feels like a strange place to invest marketing attention.

Step four: consistency across everything

If your business name, category, and location are stated differently across your site, your listings and third-party mentions, that ambiguity makes a model less confident naming you specifically — it may not be sure the “Denver Plumbing Co” on one review site is the same business as the “Denver Plumbing Company LLC” on your website. Keeping facts consistent everywhere you’re mentioned removes an entirely avoidable source of doubt.

Recommendation vs. recognition — the distinction most businesses miss

Being recognized by an engine and being recommended by it are different outcomes. An engine can have enough information to correctly describe your business when asked directly — “how does [business] compare to alternatives?” — without ever volunteering your name to someone who asks an open, unprompted question like “what’s the best plumber in Denver?” The first is recognition; the second is the thing that actually drives new customers. All four factors above — findability, extractability, trust and consistency — matter more for crossing that specific line from recognized to recommended.

A practical checklist

• Confirm AI crawlers can actually reach your content — check robots.txt and whether your key pages render without JavaScript.
• State your core facts in plain text — services, area, pricing range, specialties and hours, not only in an image or PDF.
• Add structured data — FAQ and LocalBusiness schema, clear single-H1 headings.
• Earn mentions where your engines actually look — review platforms and directories broadly, and forums like Reddit specifically if your category has a presence there.
• Keep your name, category and location consistent everywhere you’re mentioned.
• Check what’s actually being said, not just whether you rank — the only way to know is to ask the real questions and read the real answers.

How Surfaced fits

Surfaced runs an AI-readability audit against your site — checking AI crawler access, whether an llms.txt file exists, crawlable content, schema markup, meta description and heading structure — alongside weekly checks of the real questions your customers ask against ChatGPT, Claude, Gemini and Perplexity, with web search on. You see whether you’re recommended or merely recognized, your rank, the sentiment, and who got named instead. If you just need the crawlability piece, our free llms.txt generator is a fast first step, no signup required.

Related reading

For how this rolls up into a single number you can track, see what an AI visibility score is and how it’s measured. For the broader discipline this all sits under, see our guide to Generative Engine Optimization.

Frequently asked questions