TL;DR
- LLMs recommend brands from two sources: what they learned during training, and what they retrieve live at answer time. You can influence the second far faster than the first.
- A single user question fans out into many sub-queries behind the scenes ("fan-out"), so you need to win a cluster of related questions, not one exact keyword.
- Entity clarity (being unambiguously and consistently described the same way everywhere) matters more here than in classic SEO.
- Off-site proof, comparison pages, community threads, review platforms, matters as much as your own website, because models weight independent sources as evidence.
- Most invisible brands fail for structural reasons: no clear category definition, no third-party mentions, or content that never states a direct, quotable answer.
Getting named in an AI answer is not the same skill as ranking on Google, and most brands are invisible for reasons that have nothing to do with how good their product actually is. Here is how the recommendation actually happens, and what to change.
How LLMs actually decide what to recommend
An AI model recommends a brand based on two separate mechanisms: what it absorbed during training (a static snapshot of the internet up to some cutoff date) and what it retrieves live when it answers your question, commonly through web search or a retrieval-augmented generation (RAG) system that pulls current pages into the model's context before it writes a response. Most consumer-facing assistants (ChatGPT with browsing, Perplexity, Gemini, and Google's AI Overviews) lean heavily on the retrieval side for anything time-sensitive or comparison-shaped, which is most commercial queries. That means the fastest lever you have is not waiting for the next training run to include you. It's making sure the pages a model retrieves right now say something clear and citable about you.
This is the single most important mental shift for anyone trying to influence this channel: you are not optimising a static index the way you would for classic Google SEO. You are optimising the live evidence a model gathers and synthesises in the seconds it takes to answer. That evidence includes your own site, but also every third-party page that mentions you.
The fan-out concept, and why one keyword isn't enough
When someone asks an AI assistant a question like "best project management tool for a five-person design agency," the model typically doesn't just match that sentence against an index. It reasons about the question, and modern retrieval-augmented systems often decompose or expand a single query into several related sub-queries (by category, by use case, by comparison) before retrieving and synthesising results. This is sometimes called query fan-out. Practically, it means you are not competing to rank for one phrase. You are competing to be a plausible answer across a whole cluster of the ways someone might ask about your category: by use case, by comparison ("X vs Y"), by constraint ("cheapest," "for beginners," "for enterprise"), and by problem ("how do I stop losing track of client feedback").
The practical consequence is that your content strategy needs to cover the cluster, not the single head keyword. If you only have one page targeting "best CRM software," you're set up for classic SEO, not for the dozen different phrasings an LLM might generate on its way to answering a related question.
Entity clarity: being unambiguously, consistently you
Entity clarity means a model can confidently identify what your brand is, what category it belongs to, and what makes it distinct, because that description is consistent everywhere it appears. This matters more for AI recommendation than it ever did for classic SEO, because a language model is, at its core, reasoning over relationships between entities and concepts, not just matching keywords in a string.
Concretely: make sure your homepage, About page, and any structured data (schema.org Organization and Product markup) state plainly what you do, who it's for, and what category you compete in, using consistent language. If your own site describes you five different ways across five pages ("a platform," "a tool," "a solution," "software," "a service"), and your LinkedIn, Crunchbase, and G2 listing each describe you differently again, you are making it harder for a model to form a confident, consistent picture of what you are and when to recommend you. This is a case where being boring and consistent beats being clever and varied.
Owned content: answer-first structure and schema
The single highest-leverage content change is writing the direct answer to a question in the first two or three sentences of the relevant section, before the explanation, background, or caveats. Models retrieving your page to answer a user's question favour content that gives them something quotable near the top, because that is what they can lift into a synthesised answer with confidence. Bury your actual answer under three paragraphs of preamble and you make the model do more work to extract it, which reduces the odds it chooses your page over a competitor's that answered faster.
Pair this with structured data. FAQPage, Article, and Organization schema give models (and the search systems some of them are built on) an explicit, machine-readable version of your content's structure, reducing ambiguity about what's being claimed and by whom. It is not a magic switch, but it removes friction from the retrieval and synthesis process.
You are not optimising a static index anymore. You are optimising the live evidence a model gathers and synthesises in the seconds it takes to answer.
Off-site proof: the sources engines actually retrieve
Your own website is only half the evidence pool. AI models retrieving information to answer a comparison or recommendation question frequently pull from third-party sources: comparison and "best of" articles, community discussion (Reddit threads, forum posts), review platforms (G2, Trustpilot, Capterra), and independent publications. A model treats a claim you make about yourself on your own site with more scepticism, implicitly, than the same claim appearing on an independent third-party page, because independent corroboration is a stronger signal of accuracy.
This means a meaningful share of the work to get recommended happens off your domain: being named accurately in comparison content (even content you didn't write), having a presence in the communities where your category gets discussed, and maintaining a clean, substantive review profile. If nobody outside your own site has ever described what you do, you are asking the model to take your word for it alone, which is a weaker position than having independent evidence to retrieve.
Measurement: how to know if it's working
Track three things monthly, using a dedicated AI visibility tool such as Peec AI or Profound rather than manually querying assistants yourself, since manual spot-checks are neither repeatable nor comprehensive. Visibility rate across a realistic prompt set covering your category's likely fan-out queries, not just your brand name. Citation source, specifically whether your own domain is what's being cited when you're mentioned, versus a competitor's comparison page mentioning you in passing. And sentiment or framing, since being mentioned negatively or as a "budget option" when you're not positioned that way is a different problem than not being mentioned at all, and requires different fixes. For a practical path from tracking data to citation work, see how to turn Peec or Profound data into citations.
Rerun the same prompt set on a fixed schedule (weekly or monthly) so you can see trend, not just a single snapshot, since model outputs vary run to run and single checks are noisy.
Common reasons a brand is invisible in ChatGPT and Gemini
Most invisibility comes down to a small number of fixable problems, and it's rarely "the product isn't good enough." First, no clear category anchor: if a model can't confidently place you in a category, it won't recommend you for category-level questions, even if you're genuinely excellent within a narrower niche. Second, no independent evidence: if the only place that describes what you do is your own website, you're relying entirely on training-data recall or a single retrieved source, both weaker than corroborated evidence. Third, content that answers indirectly: pages that build up to an answer through narrative or marketing framing rather than stating it plainly near the top. Fourth, being genuinely new or small: models trained on older data and retrieval systems that favour established, frequently-cited domains will structurally under-recommend younger brands regardless of quality, and the fix is patience plus deliberate off-site presence-building, not a technical trick. We break down the structural problem further in why your website never shows up in ChatGPT.
Frequently asked questions
How do I get ChatGPT to recommend my brand? Give the model clear, consistent evidence to work with: an unambiguous description of your category and offering on your own site, answer-first content structured around the real questions buyers ask, and independent third-party mentions (reviews, comparisons, community discussion) it can retrieve as corroboration. There's no submission form or paid placement; it is a function of what's retrievable and how clearly it's written.
What is query fan-out? Query fan-out describes how a retrieval-augmented AI system can expand a single user question into multiple related sub-queries before gathering sources and synthesising an answer. It means you need to be a plausible answer across a cluster of related phrasings, not just one exact keyword.
Does traditional SEO still matter for AI search? Yes, but as a foundation rather than the whole strategy. Crawlability, page speed, and structured data still matter because they affect whether a model's retrieval system can access and parse your content at all. What's different is the content itself needs to be answer-first and the off-site strategy needs to target the sources AI models cite as evidence, not just backlinks for ranking authority.
How long does it take to start showing up in AI answers? Content and technical changes can affect retrieval within weeks once a page is re-crawled, but off-site presence (getting cited on comparison pages, reviews, and communities) typically takes eight to twelve weeks to compound into a visible shift, because it depends on third parties publishing and models re-weighting sources on their own schedule.
Why does a smaller competitor get recommended over us? Usually because they have clearer entity definition, more consistent off-site presence, or content that answers the specific question more directly, not because the underlying product is better. Run a visibility audit across the actual prompts your buyers use to see exactly where the gap is.
Do I need to be mentioned by name, or is being cited as a source enough? Both matter but for different reasons. A brand mention (even without a link) shows up in training-data recall and general reputation. A citation, where your own page is the retrieved source, drives referral traffic and gives you more control over the framing, since the model is drawing directly from your words rather than a third party's summary of you.
