How AI assistants decide which brands to recommend

A growing share of buying decisions now begins with a question typed into an AI assistant rather than a search box. Adobe Analytics, drawing on more than a trillion visits to United States retail sites, measured traffic from generative AI sources in February 2025 at 1,200% above its July 2024 level. Through the 2025 holiday season the same measure was up 693% year over year, and that traffic converted 31% better than other sources. When a shopper asks “what’s the best CRM for a small agency?” or “which moisturiser suits sensitive skin?”, a model returns a short, confident list of names. Being on that list is quickly becoming as consequential as a first-page ranking once was.

So the practical question is no longer only “how do we rank?” but “how does an assistant decide whom to name?” The mechanics are knowable, and they differ from classic SEO in ways that matter.

The assistant is assembling an answer, not returning a list

A search engine retrieves and ranks pages. An AI assistant composes an answer in natural language and then, increasingly, supports it with citations. The brands it names are the ones it can describe confidently and, where it cites, evidence cleanly. That distinction matters, because a model will happily omit a brand it cannot characterise clearly, even one with a strong website.

Two inputs drive the output. First, what the model already holds from training — broad, slow-moving, and skewed toward brands the web writes about often. Second, what it retrieves at answer time. Most of the prose in a typical answer comes from the first; the citations come from the second. Confusing the two is the most common analytical error in this field, and it leads directly to the wrong conclusion about where to invest.

Getting cited and getting mentioned are two different things

It is tempting to assume that if you can win the citation, you win the mention. The evidence says these are separate mechanisms with separate drivers.

On the citation side, engines behave nothing alike. One 2026 analysis of 22,295 AI answers and 115,843 citation events found ChatGPT drawing 68.8% of its citations from brand websites and product pages, while Perplexity took 35.4% and Google AI Mode 37.9%, spreading the remainder across comparison lists and editorial coverage. Ahrefs, comparing the top 50 most-mentioned sources across AI Overviews, ChatGPT and Perplexity, found that 86% were not shared across all three — only seven domains appeared in every top 50.

On the mention side, the picture is far more consistent, and it points away from your own website. Ahrefs analysed 75,000 brands to see which factors correspond with appearing in AI Overviews at all. The three strongest correlations were all off-site: brand web mentions (0.664), branded anchor text (0.527) and branded search volume (0.392). Backlinks managed 0.218 — roughly a third the strength of simply being written about. Brands in the top quartile for web mentions averaged 169 AI Overview mentions, against 14 for the quartile below. Twenty-six per cent of the brands studied had none at all.

Correlation is not causation, and the study’s authors say so plainly. But the shape of the finding is hard to miss: your own site is where an engine often goes to confirm what it says about you; the wider web is what determines whether it thinks of you in the first place.

Freshness helps, but less dramatically than the folklore suggests

You will find confident claims circulating that content updated in the last 30 days earns some precise multiple more citations. Those figures rarely name a study, and we could not trace them to one, so we are not repeating them.

What is measurable is more modest. Ahrefs examined 17 million citations across seven platforms and found AI-cited URLs averaging 1,064 days old, against 1,432 days for organic search results — 25.7% fresher. ChatGPT showed the strongest recency preference, citing pages roughly 393 days newer than Google’s organic results. But the average cited page is still nearly three years old. Assistants prefer current material; they do not prefer new material. Long-lived, well-maintained pages still win, which is an argument for keeping facts accurate rather than for churning publication dates.

What actually makes a brand recommendable

Read across these studies and a coherent picture emerges. Assistants name brands that are widely and consistently described. Recognition beats link-building. Presence across independent sources beats publishing volume on your own domain. Currency beats novelty. And because each engine reaches for a different source mix, there is no single channel to optimise — the only strategy that survives contact with all of them is being describable the same way everywhere.

That last point is the one most often missed. A model reconciling a dozen accounts of who you are, what you sell and where you operate will state a claim confidently when those accounts agree, and hedge or skip you when they do not. Inconsistency is not merely untidy; it is a reason to be left out of the answer.

From visibility to governance

This is where the work shifts from marketing tactics to brand governance. If assistants assemble recommendations from whatever facts they can find about you, the underlying asset is a single authoritative, machine-readable definition of your brand — products, claims, voice, and the boundaries of what may be said — that stays consistent everywhere it surfaces. Treating that definition as managed infrastructure rather than a static brand book is the premise behind kbie.ai, which maintains a brand’s knowledge graph so the facts an AI relies on are accurate, current and safe to publish. The aim is not to chase any one engine, but to be legible and trustworthy to all of them at once.

The brands that win the recommendation will be the ones that decide, deliberately, what is true about them — and then make that truth easy for a machine to find, quote and stand behind.

FAQ

Is optimising for AI assistants different from SEO?

It overlaps but is not the same. SEO optimises a page to rank in a list. AI visibility optimises your brand’s facts so a model can describe you confidently inside a generated answer. The measured correlations point at off-site signals — how widely and consistently your brand is discussed — well ahead of backlinks.

Can I control what an AI says about my brand?

You cannot edit a model’s output, but you strongly influence it by controlling the evidence. Consistent facts across your site, reference pages, reviews and press give a model a clear basis to describe you, and reduce the chance it repeats something outdated or wrong.

Why does the same question return different brands on different assistants?

Because each engine draws on a substantially different source mix — Ahrefs found 86% of top-mentioned sources were not shared across ChatGPT, Perplexity and AI Overviews. A brand legible to one can be invisible in another, so the goal is consistent describability rather than tuning for a single model.

What is the single most useful first step?

Audit your own facts. Check whether your website, profiles and the major third-party pages that describe you actually agree on your products, pricing, markets and positioning. Reconciling those contradictions is the highest-leverage move before any more advanced AI-visibility work.