How AI answer engines decide which brands to recommend

“Why does ChatGPT recommend them and not us?” is the question every marketer is starting to ask. The reassuring answer: the mechanism is understandable, and most of it is influenceable.

Step 1: retrieval from trusted sources

Modern answer engines don't recite your marketing copy from memory. When a user asks for a recommendation, they gather relevant material — from their training data and, increasingly, from live search — weighted toward sources they consider authoritative. Perplexity and Gemini even show these sources directly.

So the first filter on whether you get recommended is brutally simple: do trusted sources mention you, in the right context, for the query being asked?

Step 2: synthesis into a consensus

Next, the model summarizes what those sources agree on. If five trusted pages describe a competitor as “the best option for enterprise teams” and none mention you for that use case, the model repeats the consensus. It isn't biased against you — it simply has nothing to pull that puts you there.

To change the answer, change the sources behind it.

What counts as a “trusted source”

It varies by niche, but the usual suspects show up again and again:

• Review and comparison sites (G2, Capterra, and category-specific equivalents)
• Communities where buyers ask for advice (Reddit, niche forums)
• Reference pages (Wikipedia and well-maintained wikis)
• Industry media, news and PR
• YouTube and other creator content
• Structured, well-organized pages that are easy for machines to parse

Notice what's missing from the top of that list: your own homepage. It matters for confirmation, but it rarely creates the recommendation on its own.

What you can actually influence

1. Presence: get mentioned on the sources that matter for your category, in the contexts your buyers ask about.
2. Consistency: make sure those mentions describe you the same way — one clear positioning, repeated, becomes the consensus.
3. Machine-readability: keep your site crawlable by AI bots and provide structured data and an llms.txt so engines can parse you correctly.
4. Freshness: sources age. Keep earning new mentions so the consensus stays current.

Make sure you're even readable

A surprising number of sites quietly block AI crawlers or ship no machine-readable signals — and then wonder why they're invisible. Before anything else, confirm the basics: that GPTBot, PerplexityBot, ClaudeBot and Google-Extended aren't blocked, and that you publish an llms.txt. It's the cheapest GEO win there is.

The honest limit

Answers are probabilistic; the same question won't always produce the same names. That's why serious GEO is measured, not guaranteed — you track visibility across many runs and engines, act on the gaps, and confirm the lift. Anyone selling a guaranteed citation is selling a coin flip.

See where you stand

The fastest way to make this concrete is to look at your own brand. Run a free AI visibility check to see who the engines recommend in your category, and the exact sources shaping those answers.