At a glance
- AI engines recommend brands the way markets always have, by consensus and proof, which makes AI visibility a brand strategy problem before it is a technical one.
- Brand mentions across the web predict AI citations roughly three times more strongly than backlinks do, according to Ahrefs research covering 75,000 brands.
- Princeton research found that adding quotable statistics, citations and expert quotations to a page lifts its visibility in AI answers by 30 to 40 percent.
- Your own website is necessary but not sufficient. Answer engines trust what others say about you more than what you say about yourself.
- Southeast Asian brands are underrepresented in AI training data, so consultancies and companies that publish structured regional content now can claim citation share that global competitors have not touched.
AI engines recommend brands the way markets always have, by consensus and proof. That makes AI visibility a brand strategy problem before it is a technical SEO problem. This guide explains how generative engines choose which brands to mention, and the order in which to earn their citations.
What does it mean to be mentioned by ChatGPT?
When someone asks ChatGPT, Perplexity or Gemini a buying question, the engine composes an answer rather than returning a list of links. Being mentioned means your brand appears inside that composed answer, either named as a recommendation or cited as the source of a fact the answer relies on. This is different from ranking. A page can rank third on Google and never be quoted by an answer engine, and a page with modest rankings can be quoted verbatim because it states a fact clearly enough to lift.
The discipline of earning these mentions is called generative engine optimisation, or GEO. The term comes from research led by Princeton University and IIT Delhi, presented at KDD 2024, which tested what actually changes citation behaviour across thousands of queries. GEO is not a bag of tricks. It is the practice of making your expertise legible to systems that read the entire web and repeat what they find consistent.
The distinction matters strategically because the buyer who asks an AI engine gets a shortlist, not ten blue links. If your brand is not in the answer, you are not merely below the fold. You do not exist in that conversation.
ChatGPT does not rank pages. It repeats consensus.
Most companies approach AI visibility as a new flavour of SEO, and this is where most of them fail. They optimise their own website harder, add more keywords, publish more posts, and wait. The results disappoint because they are optimising the one source an answer engine inherently discounts: the brand talking about itself.
The evidence on this is unusually clear. Ahrefs analysed 75,000 brands and found that mentions of a brand across the web correlate with AI visibility at 0.664, while backlinks correlate at just 0.218. The strongest predictors of appearing in AI answers are all signals that live outside your own domain. An answer engine behaves less like a librarian ranking documents and more like a diligent analyst phoning around the market. It asks what everyone else says about you, then checks whether your own story matches.
Consider how Michelin built the most trusted restaurant authority in the world. A tyre company earned that position by publishing structured, consistent, independently verifiable judgements year after year, until its word became the consensus. What Michelin stars are to restaurants, AI citations are becoming to brands. They are earned through repetition, specificity and proof, not through a clever homepage.
An AI engine cannot repeat a clear story about your brand if you have never told one consistently.
The reframe, then, is this. Stop asking how to optimise for the algorithm. Start asking whether the web, read as a whole, tells one coherent and well evidenced story about your brand. That is a question brand strategists have been answering for decades.
Exhibit 1: The Consensus Stack
The Consensus Stack describes where answer engines source their trust in a brand, from the ground up. Most companies invest from the top down, polishing owned channels first. Engines read from the bottom up, weighting independent corroboration most heavily. The work is to build all four layers so they agree with each other.
- Owned clarity. Your homepage, service pages and structured data say one consistent thing about who you are, what you do and for whom. Contradictions between pages read as noise, and noise is not quotable.
- Published authority. Articles, guides and FAQ content that answer real buyer questions with definitions, figures and named frameworks. This is the layer engines lift language from directly.
- Corroboration from others. Directories, industry press, reviews, podcasts and listicles that repeat your positioning in words you did not write. This layer carries roughly three times the predictive weight of backlinks, as the Ahrefs data shows.
- Verifiable proof. Named clients, published case studies, credentials and specific results that an engine can quote as evidence rather than assertion. Proof points are the most extractable content of all.
As Exhibit 1 suggests, a brand with strong owned channels but no third layer is invisible to consensus. A brand with all four layers aligned becomes the safest answer for an engine to give.
Your own website is necessary but not sufficient
None of this means neglecting your own site. Owned channels are where you control the canonical version of your story, and answer engines do read them. When we audited our own AI visibility across 22 buyer intent queries in July 2026, the pages quoted most often were our pricing guides, where specific figures and ranges were lifted directly into answer text, and our specialist healthcare pages, where named proof points appeared verbatim. What you publish on your own domain sets the vocabulary everyone else borrows.
But the same audit showed the limit. On broad agency selection queries, answers were assembled almost entirely from directories and independent round ups, not from any agency's homepage. The engine treated self description as a claim and third party listing as evidence. Both layers matter, and they must match.
The practical implication is a division of effort many marketing teams get backwards. Perfect the owned story once, thoroughly. Then spend the recurring effort earning repetition of that story in places you do not control: analyst round ups, industry publications, credible directories, podcasts and awards. One accurate mention in a well read industry publication does more for AI visibility than another ten posts saying the same thing on your own blog.
Publish content an engine can lift, not content it must interpret
The Princeton GEO research tested nine optimisation methods across 10,000 queries and found a consistent pattern. The three interventions that worked, each lifting visibility in AI answers by 30 to 40 percent, were adding quotable statistics, citing sources, and including expert quotations. Fluent, specific, well evidenced writing wins. Keyword stuffing measurably lost visibility.
This maps to three writing habits worth institutionalising. Lead with the definition: answer the core question in the first hundred words, because engines extract openings far more often than conclusions. State facts as figures: "brand identity projects in Singapore typically run eight to sixteen weeks" is liftable, while "timelines vary" is not. Attribute your claims: a sentence with a named source is safer for an engine to repeat than an unattributed assertion, so it gets repeated.
Structure does the rest. Question and answer sections, comparison tables and named frameworks give an engine clean units to quote. The reason FAQ content is cited so often is not magic. It is that a question paired with a four sentence answer is precisely the shape an answer engine needs.
Positioning discipline is now machine readable
Here is the part of this subject that SEO agencies, who currently dominate the advice on it, are not positioned to say. Every signal described above is downstream of brand strategy. A brand with fuzzy positioning cannot produce owned clarity, because there is nothing clear to state. It cannot earn consistent third party mentions, because journalists and directories describe it differently every time. It cannot offer proof points, because it never decided what it was proving.
Positioning used to be tested slowly, through market perception studies and sales conversations. It is now tested instantly and publicly, every time someone asks an AI engine who does what you do. The engine synthesises everything the web says about your category and reflects back the brands whose story is coherent. Ask it about your own category today. If the answer describes your competitors in their words and omits you, that is not a technical failure. That is a positioning audit, delivered free of charge.
This is also why the fundamentals compound. The work of defining what you stand for, saying it the same way everywhere, and evidencing it with real results was always the job. AI search has simply made the reward for that discipline measurable, and the penalty for skipping it immediate.
Exhibit 2: How ranking worked, and how citation works
| Dimension | Traditional SEO (ranking) | GEO (citation) |
|---|---|---|
| Unit of competition | A page competing for a position | A fact or claim competing for inclusion in an answer |
| What wins | Relevance and authority signals on and off page | Specificity, attributability and consensus across sources |
| Where trust comes from | Links pointing at your domain | Mentions of your brand across independent sources |
| Role of your website | The asset being ranked | The canonical reference other sources corroborate |
| Role of third parties | Sources of link equity | Sources of the consensus the engine repeats |
| How you measure | Rankings, impressions, clicks | Share of citation across a fixed query set, tracked monthly |
Southeast Asian brands are competing for unclaimed ground
The regional picture is unusually favourable, and unusually urgent. AI engines are trained predominantly on Western content, which means Southeast Asian brands are underrepresented in the consensus these systems repeat. Our July 2026 audit found that broad informational branding queries were answered almost entirely by American software companies and global publishers, with no Singapore consultancy visible at all. On Malaysia and Indonesia queries, no Singapore firm of any kind appeared.
Underrepresentation cuts both ways. It means the default answers are thin on regional nuance, and it means the citation space is effectively unclaimed. A Singapore brand that publishes structured, specific content about its market, with local figures, local examples and local pricing, is not fighting a crowded field. It is often the only credible regional source an engine can find, which is the single easiest way to become the cited authority.
The window matters. Consensus, once formed, is self reinforcing, because engines keep citing the sources that earlier answers established as authoritative. The brands that build the regional evidence base in the next year will be disproportionately hard to displace afterwards. This holds for a consultancy in Singapore, a hospital group in Malaysia or a consumer brand in Vietnam. First movers are writing the record that future answers will repeat.
How to measure whether it is working
AI visibility can be tracked without specialist tools, and it should be tracked before anything is optimised. The method is a monthly audit. Fix a set of 15 to 20 questions a real buyer would ask in your category, spanning selection queries, cost queries and informational queries. Run them across ChatGPT, Perplexity, Gemini and Google AI Overviews. Record which brands are named in each answer, which sources are cited, and how you compare against the same list last month.
Three numbers make the exercise strategic rather than anecdotal. Your share of citation: the percentage of queries in which your brand appears. Your source profile: whether citations come from your own pages or from third parties. And your displacement targets: which brands currently own the answers you want. Semrush's 2026 AI Visibility Index, which analysed 126 million AI search prompts, confirms what the category data shows: visitors who arrive from AI answers convert at roughly 4.4 times the rate of traditional organic search on average, so each point of citation share is worth disproportionately more than a ranking position.
Expect the loop to be slow at first. Engines refresh their picture of the web at different speeds, and consensus takes months to register, not days. The brands that win treat the monthly audit like a board metric: reviewed on a fixed cadence, tied to specific content and PR actions, and given at least two quarters before judging the trend.