AI SEO: how to get your brand cited in AI answers

Somewhere in the last two years, a client asked you why your competitor gets named when they ask ChatGPT for a recommendation and you do not. That question has spawned an entire industry of acronyms, tools and consultants, most of which are selling the same advice under a new label. This guide cuts through that. It explains how AI answer engines actually select and cite sources, what the evidence genuinely shows about earning those citations, what demonstrably does not work, and what to do about it.

The short version is that this is less exotic than the marketing suggests, and more demanding than it sounds. There is no secret file to upload and no schema to bolt on. The brands that get cited are the ones that rank, and that other people talk about. Those two things have always been the job. What has changed is how much the second one now matters relative to the first.

What AI SEO actually is, and why GEO is a contested word

You will see this work called AI SEO, GEO (generative engine optimisation), AEO (answer engine optimisation), or LLMO. The labels are doing more marketing work than technical work, and it is worth knowing where the real disagreement sits before you spend money on it.

Google’s position is unambiguous. Its official guide to generative AI features, updated in June 2026, states plainly: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” Its AI features documentation goes further, saying there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” At a Search Central event in July 2025, Gary Illyes was reported as telling attendees: “To get your content to appear in AI Overview, simply use normal SEO practices. You don’t need GEO, LLMO or anything else.”

Plenty of practitioners agree. Lily Ray of Amsive, who has done more public research on AI Overviews than almost anyone, wrote in January 2026 that many GEO vendors were “using this opportunity to simply repackage core SEO approaches using a different name.” Her own experience was blunter still: every time she published an article, she saw it cited across LLMs within hours, “without doing anything new or different for AEO/GEO outside of standard SEO best practices. It worked every time because LLMs use search engines.”

The serious counter-argument

Mike King of iPullRank makes the strongest case that something genuinely has changed, and he is not making it lazily. In a Search Engine Land interview he conceded the fundamentals still hold, then drew the line: “the big difference is that what they do with our inputs is dramatically different.” Where a search engine returned your page more or less intact, a model now takes it, recombines it with other sources and produces something you did not write. “The way we get that isn’t through optimization,” he argued. “It’s through engineering.”

Both things are true, and the practical resolution is boring. The tactics that earn AI citations are, overwhelmingly, SEO and digital PR tactics. But it is a distinct surface with its own metrics, and it deserves its own reporting line. Aleyda Solis put this well when she noted that while the core activities overlap, AI visibility “will still require its own dedicated effort, prioritization, and strategy.” Cyrus Shepard, after reviewing 54 separate studies, landed in the same place: “win SEO, win AI citations (most of the time, with extra steps).”

How AI answers actually choose their sources

Understanding the retrieval mechanism explains most of what follows, and it dismantles a lot of the advice you have been given.

Google AI Overviews and AI Mode

Both are grounded in the ordinary Google index. Google describes the process as retrieval-augmented generation: the model relies on “core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index,” then generates an answer from what it retrieves. There is no separate AI index and no separate ranking system. To be eligible for citation, a page needs only to be indexed and eligible to appear with a snippet.

The twist is query fan-out. Rather than answering your literal query, the system issues a set of related sub-queries and pulls sources from those results too. Google’s example: for “how to fix a lawn that’s full of weeds,” the fan-out might include “best herbicides for lawns” and “how to prevent weeds in lawn.” This is why citation and ranking have come apart. When Ahrefs re-ran its citation study in March 2026 across 863,000 SERPs and 4 million cited URLs, only 37.9% of cited pages appeared in the first ten results for the original query, down from roughly 76% in mid-2025. Around 31% did not rank in the top 100 at all.

Read that carefully, because it is widely misreported. It does not mean rankings stopped mattering. It means the page was very likely ranking for one of the invisible fan-out queries instead. As Mike King put it, “that document may have ranked number three for one of these expanded queries, but we just can’t see that.”

ChatGPT, Perplexity, Gemini and Claude

The other engines behave differently enough to matter. OpenAI says ChatGPT search draws on “third-party search providers such as Bing,” and it also runs its own crawler, OAI-SearchBot. This is the detail most sites get wrong: blocking GPTBot only opts you out of model training, while blocking OAI-SearchBot removes you from ChatGPT’s search answers entirely. Perplexity runs its own index via PerplexityBot. Anthropic operates Claude-SearchBot for search indexing, separate from ClaudeBot for training. Gemini’s app grounding is governed by the Google-Extended token, which, importantly, has no effect at all on AI Overviews.

The practical upshot is a short technical checklist that takes an afternoon: make sure you are indexed, do not block the search crawlers, and do not suppress snippets. Everything after that is a content and reputation problem.

What the evidence says earns citations

Here the picture is clearer than the noise suggests, because two independent teams found the same thing.

Being mentioned beats being linked

Ahrefs analysed 75,000 brands and correlated various signals against AI Overview visibility. Branded web mentions came out at 0.664. Backlinks came out at 0.218, referring domains at 0.295, and Domain Rating at 0.326. Seer Interactive, working separately across roughly 10,000 questions run through GPT-4o, found backlinks correlating at just 0.10 while Google organic keyword rankings hit 0.65.

Two teams, different methods, same conclusion: how often the web talks about you predicts AI visibility far better than how many links point at you. This is correlation rather than proof, and it is worth noting that Ahrefs sells a brand mention tracker, so the finding is commercially convenient for them. But its replication by an independent agency gives it real weight.

Earned media does the heavy lifting

The most striking number in this field comes from Muck Rack, which analysed more than 25 million links cited by ChatGPT, Claude and Gemini across 17 industries. Earned media accounted for 84% of all AI citations. Owned media, meaning your own website, accounted for 13.7%. Paid and advertorial content accounted for 0.3%. Journalism alone made up 27% of cited links, spanning more than 20,000 outlets.

Muck Rack sells PR software, so treat the framing with appropriate suspicion. But the figure has held between 82% and 89% across three editions of the study since July 2025, and that stability is hard to dismiss. It says something uncomfortable and useful: the models are far more interested in what others publish about you than in what you publish about yourself. This is precisely why digital PR has become the most direct lever on AI visibility, and why editorial coverage now pays twice.

Rankings, and therefore links, still matter

None of this means backlinks stopped counting. It means their effect is transitive. Links drive rankings, rankings drive retrieval, retrieval drives citation. Seer’s finding that Google organic keyword rankings correlate at 0.65 is the strongest single signal in their dataset. Lily Ray, summarising the emerging research, said it plainly: “it’s almost always ranking number one on Google gets you more citations.”

So the honest framing is this. Chasing more links purely to move an AI citation metric is a poor use of budget. Building the kind of relevant, editorially earned links that lift rankings and get your brand named in credible publications is the single most effective thing you can do, and it always was. If you want the underlying principles, our guide to what makes a high quality backlink has not changed.

Format helps, and this is the one causal finding

Almost everything in this field is correlational. One exception is the GEO study from Princeton and Georgia Tech, accepted to KDD 2024, which actually modified content and measured the change across 10,000 queries. Adding statistics, adding quotations and citing sources each lifted visibility in generative responses by roughly 30%, with a headline of up to 40%. Keyword stuffing did nothing.

That is a research harness rather than live ChatGPT, so do not treat 40% as a promise. But the direction is intuitive and it matches what the engines visibly do: they lift specific, attributable, quotable material. Vague thought leadership gives a model nothing to quote.

What does not work, despite what you have been sold

This section will save you more money than the rest of the article combined.

Structured data does not earn citations

The correlation looks compelling. Across 6 million URLs, Ahrefs found AI-cited pages were about three times more likely to carry JSON-LD. Then Ahrefs tried to kill their own finding, and largely succeeded. They took 1,885 pages that added JSON-LD between August 2025 and March 2026, matched them against 4,000 control pages, and ran a difference-in-differences analysis. The result: a 4.6% decline for AI Overviews, and effects statistically indistinguishable from zero for AI Mode and ChatGPT.

Their explanation is textbook confounding. Sites that add schema also tend to do everything else well. Google’s own documentation agrees there is “no special schema.org structured data that you need to add.” Keep schema for rich results and product data, which is what it is for. It is not an AI citation lever.

llms.txt, chunking and FAQ blocks

John Mueller has stated flatly: “FWIW no AI system currently uses llms.txt,” comparing it to the old keywords meta tag. Google’s documentation confirms it ignores such files entirely.

On chopping your content into machine-friendly fragments, Danny Sullivan was emphatic on Google’s own podcast in January 2026: “turn your content into bite-sized chunks, because LLMs like things that are really bite size… So we don’t want you to do that. I was talking to some engineers about that. We don’t want you to do that. We really don’t.”

As for bolting an FAQ block onto every page, Omniscient Digital analysed 23,387 cited sources and found FAQ pages accounted for 0.41% of citations. It is close to the least cited format there is.

Scaled comparison pages will get you demoted

The one tactic that visibly works in the short term is publishing listicles and “us versus competitor” pages, because models lean heavily on comparison content. The problem is what happens when everyone does it. Lily Ray, discussing a wave of site-level demotions in January 2026, observed that “almost unanimously all the companies had lots of listicles,” and described the programmatic version, where a company spins up a thousand near-identical competitor comparisons, as “scaled content abuse.”

Google says the same in its own guidance, warning that creating content targeting fan-out query variations primarily to influence AI responses violates its scaled content abuse policy. Build a genuinely useful comparison page. Do not build a thousand of them.

The traffic reality, and why this is a brand play

Before you reallocate a budget, understand what this channel is currently worth, because the honest answer is: not much traffic, and a great deal of influence.

AI assistants send roughly 0.4% of referral traffic across the sites Ahrefs tracks, against about 26% from Google. Rand Fishkin’s estimate, derived from OpenAI’s own published prompt volumes, put Google at roughly 210 times the search volume of ChatGPT in September 2025. He is also the most reliable sceptic in this field, and his description of what these systems are is a useful antidote to the hype: they “are statistical lotteries and predictors of next tokens or next words.”

What has changed materially is clicks. The most trustworthy measurement here comes from Pew Research Center, which has nothing to sell. Tracking 68,879 real Google searches from 900 US adults, Pew found that users clicked a result on 8% of visits where an AI summary appeared, against 15% where it did not. Clicks on links inside the AI summary itself happened on just 1% of visits. And 26% of visits ending in an AI summary ended the browsing session entirely, against 16% without.

Seer Interactive’s 2026 analysis, covering 5.4 million queries and 2.43 billion impressions, adds the crucial commercial nuance. Being cited inside an AI Overview delivers around 120% more organic clicks than appearing on the page but not being cited. It still delivers roughly 38% fewer clicks than a query with no AI Overview at all. In other words: if an AI Overview is going to appear, you want to be in it. You just cannot expect it to replace what you had.

The traffic is small, but it converts hard

That small slice of traffic behaves unusually well, and this is where the commercial case sits. Adobe Analytics, drawing on more than a trillion visits to US retail sites, found that AI-referred visitors converted 31% more often than other traffic during the 2025 holiday season, with revenue per visit up 254% year on year. Those visitors also spent 45% longer on site and were 33% less likely to bounce immediately. That is retail rather than B2B, so read it as directional, but the sample is enormous.

Others report the same direction. Semrush puts the average AI search visitor at 4.4 times the value of a traditional organic visitor on conversion rate, though it publishes no methodology behind that figure. Ahrefs, reporting transparently on its own site, found AI search delivered 0.5% of its traffic but 12.1% of its signups, a roughly 23-fold conversion advantage. Ahrefs was also honest about the flip side: users arriving from AI search click links far less often than they do from traditional search.

The pattern across all three is consistent. AI sends very little traffic, and the traffic it sends is unusually close to a buying decision, because the model has already done the shortlisting. That is the whole argument for caring about this now: you are not chasing sessions, you are trying to be in the shortlist.

Put those together and the strategy writes itself. This is not yet a traffic channel. It is a recommendation channel, and it compounds into the same brand equity that our link building statistics have always pointed at.

The playbook: how to actually earn AI citations

Everything above collapses into a short list of things that genuinely move the needle.

Get the technical floor right

• Indexing. Make sure the page is indexed and can show a snippet.
• Search crawlers. Allow OAI-SearchBot, PerplexityBot and Claude-SearchBot in robots.txt if you want to appear in those engines.
• Training versus search. Do not confuse the training crawlers with the search crawlers.

This is an afternoon of work, and it is the only part that resembles traditional technical SEO.

Rank for the question behind the question

Because of query fan-out, you are competing for sub-queries you cannot see. The answer is not to build a page for every possible variation, which Google explicitly calls spam. It is to cover a topic properly and deeply enough that you rank for the neighbouring questions naturally. Comprehensive, genuinely expert coverage is what fan-out rewards.

Get talked about, not just linked to

This is the biggest lever, and the one most companies underinvest in. If earned media drives 84% of AI citations and brand mentions correlate at 0.664, then the work is to be present in the places the models read: industry publications, journalist coverage, credible roundups, expert commentary, community discussion. That is a digital PR outreach programme, and it now pays into rankings, referral traffic and AI recommendation at the same time.

Be the source others cite about you

Lily Ray’s framing is the sharpest advice in this entire space: aim to be “recommended by everybody else without recommending yourself.” A model reading ten reviews that name you is more persuasive than your own landing page claiming you are the best. Third-party listicles and category roundups are where commercial recommendations get formed, and earning a place in them is classic outreach work.

Give the model something to lift

The one causal finding in the field says to add statistics, quotations and cited sources. Original data, a specific number, a quotable line from a named expert, a clear direct answer near the top of a section. Publish things that are worth quoting and you will get quoted. This is also, not coincidentally, how you earn links in the first place.

How to measure it without fooling yourself

Be careful here, because the tooling category has run ahead of the science. Lily Ray’s warning is worth internalising: “it is currently impossible to track exactly what real LLM users are asking in aggregate. Any prompt volume data provided by these new tools is, at best, highly directional and greatly sampled.”

Rand Fishkin demonstrated the underlying problem experimentally. When hundreds of people asked an LLM the same question, “almost no two people got the same list of brands.” These systems are non-deterministic. A tool that tells you that you rank third for a prompt is reporting one roll of the dice.

Measure it anyway, but measure it honestly.

• Share of voice. Track it across a fixed set of prompts that matter commercially, run repeatedly over time, and watch the trend rather than the number.
• AI referral traffic. Track referral traffic from AI sources in analytics.
• Branded search volume. The closest proxy you have for whether the models are learning who you are.
• Rankings. Keep watching them, because rankings remain the mechanism.

If you would rather have this run for you, that is what our AI SEO programme is built to do. Our guide to the metrics worth tracking applies here with very little modification.

Where this leaves your strategy

The uncomfortable truth for the vendors selling AI optimisation is that the answer looks a lot like good marketing. Rank well, because retrieval starts with the index. Earn coverage in publications and communities, because that is what the models overwhelmingly read. Publish material specific enough to be worth quoting. Fix the crawler settings once. Ignore the file formats, the schema promises and the chunking advice.

What has genuinely changed is the balance. For twenty years, the fastest route to visibility was ranking a page. Today, being widely and credibly discussed is at least as powerful, and the evidence suggests it may be more so. That shifts weight from purely technical SEO toward the reputation work that digital PR and quality link building have always done, which is a shift most companies are underprepared for.

It also rewards patience. Brand mentions accumulate slowly, coverage compounds, and there is no file you can upload to shortcut it. The brands being recommended by AI in 2026 are, almost without exception, the brands that have been earning genuine coverage for years. The good news is that the work is not mysterious, and it is not new. It is just harder to fake than the acronym sellers would like you to believe.