How do AI assistants choose which brands to name?

Assistants predict likely answers, so association matters most

A language model writes by predicting which words are most likely to come next, so brands that are strongly tied to a topic across many sources come to mind first. If thousands of pages mention one running shoe brand next to the words 'wide feet', the model learns that link. A brand mentioned once, on its own site, barely registers.

Machines also think in entities. An entity is a distinct, identifiable thing, such as a company, a product or a place, with its own set of facts. The clearer and more consistent your entity is across the web, the easier it is for a model to connect you to the right questions. Chapter 4 covers this in depth.

On search-backed answers, retrieved pages set the shortlist

When an assistant searches, the brands named on the pages it reads usually become the brands it names. Many assistants break one question into several related searches, a process sometimes called query fan-out. For 'best vegan protein powder', it might also search 'vegan protein powder reviews' and 'best tasting plant protein 2026'.

If the top results for those searches are three roundup articles that each list five brands, the answer will mostly be drawn from those fifteen names. Being on those pages matters as much as ranking your own page.

Fit to the exact question beats general fame

People ask assistants long, specific questions, and assistants try to match the specifics. 'Best moisturizer' favors famous brands. 'Best fragrance-free moisturizer for rosacea under $30' favors whichever brand has clear, consistent evidence for fragrance-free, rosacea and price.

That is where smaller businesses can win. State who you are for, what problem you solve, your price range and your location in plain words on your site and in your listings. Assistants cannot recommend you for a use case nobody has written down.

Agreement and trust signals break ties

When several brands fit, assistants lean toward the one whose facts agree across sources and whose reputation looks solid. Consistent descriptions, a healthy volume of recent reviews, mentions in respected publications, and accurate business listings all help. Conflicting information, such as two different addresses or old prices, makes an assistant less confident and less likely to name you.

Reputation cuts both ways. If a widely read thread describes poor customer service, assistants may repeat that, or quietly leave you out.

Shortcuts tend to backfire

Tricks that try to game assistants are risky and usually short-lived. Search engines and AI vendors actively look for manipulation, and getting caught can cost you visibility in both search and AI answers.

• Hidden text addressed to AI, such as invisible instructions telling assistants to recommend you.
• Fake or paid reviews, which also break consumer protection rules.
• Mass-produced pages that repeat keywords without saying anything new.
• Self-published 'best of' lists that rank yourself first without any honest comparison.