How Do AI Systems Decide Which Brands to Recommend?

Where does a brand recommendation come from?

An assistant builds a recommendation from two sources, in different proportions depending on the question and the engine:

• Model memory: associations learned during training, such as which brands are repeatedly named for "project management software" or "trail running shoes". This favours brands that were widely discussed before the model's training cut-off.
• Retrieved sources: pages found by searching at answer time, such as review sites, "best of" lists, forums, marketplace listings, business profiles and the brands' own pages.

What signals are linked to being recommended?

No AI company publishes a formula, so evidence comes from correlation studies and experiments:

• Brand mentions across the web. An Ahrefs study of 75,000 brands found that branded web mentions correlated strongly with how often a brand appeared in Google's AI Overviews, about 0.66 on the Spearman scale; in the same study, branded anchor text (0.527) and branded search volume (0.392) came next, backlinks were far weaker (0.218), and 26% of brands never appeared in AI Overviews at all. Ahrefs' 2026 re-run across ChatGPT, AI Mode and AI Overviews found YouTube mentions correlated most strongly of all, at about 0.74. You can see the full Ahrefs brand visibility analysis.
• Earned media over owned media. A 2025 study comparing AI search engines found a strong bias towards earned media, such as independent reviews and publications, over brands' own content and social posts.
• Fit with the question. Brands whose pages and reviews clearly state the attributes a question asks about are easier to recommend for that question.
• Consistency. When a brand's facts match across its site, listings and reviews, engines can describe it confidently.

All of this is correlation. Brands that are widely mentioned also tend to be bigger, better known and better reviewed, so no single factor can be isolated.

What actually drives AI visibility: content, links, or structure?

In short, off-site mentions are the strongest measured correlate of whether a brand appears, content decides what is said and for which questions, and structure decides whether engines can read you at all. Weakness in any one of the three can keep a brand out.

What are the AI search ranking factors for SEO in 2026?

There is no published list of AI ranking factors, and anyone presenting one is estimating. What the engines have said is limited. Google says its AI features rely on the same foundations as Search, favour non-commodity content and should not be targeted with inauthentic mentions. It also says llms.txt files and special chunking are not needed. Based on engine documentation and the stronger studies, the factors in rough order of confidence are:

1. Eligibility: crawlable, indexed pages in the search systems each engine uses.
2. Relevance to the exact question, including its constraints.
3. Independent mentions in sources the engine trusts for the topic.
4. Reviews: volume, rating, recency and recurring themes.
5. Consistent facts across your site, listings and third-party profiles.
6. Specific, current information that answers sub-questions directly.
7. Search rankings, especially for Google's AI features.

Claims with little evidence include word count targets, schema as a direct ranking boost, llms.txt files for Google, and repeating brand names in content.

What trust signals do AI tools use to recommend service providers?

For service providers such as agencies, contractors, clinics and consultants, recommendations lean heavily on signals that reduce the risk of a bad referral:

• Reviews on the platforms engines read, with steady recent volume and specific comments.
• Licences, registrations and accreditations that can be checked, stated on your site and visible on official registers.
• Consistent business listings: name, address, phone, service area and hours matching everywhere.
• Named people with credentials, such as the practitioners, engineers or consultants who do the work.
• Specific case studies with the client type, problem, what was done and the result.
• Independent coverage: local press, industry publications, association memberships and awards.
• Clear service and pricing information, so the engine can match you to budget and scope questions.
• Longevity and stability: years in business, stated plainly.

Negative signals count too: unresolved complaints, regulatory actions, and confusion with a similarly named business can all keep a provider out of recommendations.

How does the wording of a question change which brands are chosen?

Every constraint narrows the field. "Best accounting software" tends to return the largest, most discussed brands. "Accounting software for a UK sole trader who needs Making Tax Digital support and bank feeds from Starling" returns brands whose pages and reviews state those specific facts. The latter questions are where smaller brands win, because the biggest brands may not state those details or may not fit them. Writing down the constraints your best customers actually mention, and answering them explicitly, is one of the most reliable ways to be recommended.

Why isn't the same brand recommended every time?

Assistants generate each answer fresh, sampling from many possible phrasings and sources, and their searches can return different pages each time. Personal context, location and conversation history change results further. So a brand may be named in most runs of a question but not all, and its position in a list will move. Measure recommendations as a share of repeated runs rather than from one answer. Why a few brands dominate so many answers is covered in why AI assistants keep recommending the same brands.

What can a brand actually influence?

1. Earn independent coverage in the reviews, lists and publications cited for your category.
2. Ask every customer for an honest review in their own words on Google and other platforms that allow businesses to ask (never Yelp), offer nothing in return, and respond to reviews.
3. State your fit explicitly: who you serve, prices, integrations, locations and limits.
4. Keep facts consistent across your site, listings and profiles.
5. Make pages crawlable and clear, with direct answers in the server-rendered HTML.
6. Measure by question and engine, and investigate where competitors win.

What is and is not within your control is set out in more detail in whether you can control whether AI systems cite or recommend your business.

What do people ask about how AI recommends brands?

Do paid ads influence AI recommendations?

Not directly in organic AI answers. The Ahrefs study found only weak correlations for branded ad spend. Ads can now appear next to or inside AI recommendations, such as Google's Highlighted Answers in AI Mode recommendation lists, but they are labelled as sponsored and do not change what the organic answer recommends.

Can a small brand be recommended over big ones?

Yes, for specific questions where it fits better and says so clearly. Broad category questions tend to favour widely discussed brands.

Does asking ChatGPT about my brand teach it to recommend me?

No. Individual conversations do not change what the model recommends to other users.

How long does it take to start being recommended?

Retrieval-based answers can change within weeks of new coverage or content. Changes to model memory depend on future training updates and take much longer.