The 2026 GEO and AEO playbook: get cited by AI

Getting named inside an AI answer is now a separate job from ranking below it, and this is the playbook I actually run for it. Generative engine optimisation (GEO) gets your page quoted inside a synthesised AI answer; answer engine optimisation (AEO) wins the direct answer box. Both sit on your SEO base, and both are won more by clarity and off-site mentions than by any single on-page trick. Here is the whole thing, in the order I work it.

What GEO and AEO optimise for

GEO and AEO are not new names for SEO. SEO earns a ranked link, AEO earns the extracted answer, and GEO earns a citation inside a generated answer from ChatGPT, Perplexity, Google AI Overviews, or Claude. I broke the three apart in GEO vs AEO vs SEO; this post is the operating playbook that sits on top of those definitions. The work runs in three layers, base up, and the order matters: skipping straight to the AI layers on a broken foundation is the most common way teams waste a quarter.

Layer 1: get the SEO base right

None of the AI layers work on a page an engine cannot crawl, parse, or trust. Keep it boring and correct: one clear topic per URL, fast pages, a clean heading hierarchy, and crawlability that welcomes the retrieval bots you want citing you. Which files actually control that last part is its own rabbit hole, so I wrote up llms.txt and AI crawler control separately. Two base issues quietly sink more sites than any content problem: primary content that only renders after client-side JavaScript (so the engine sees an empty shell), and thin pages that cover a fragment of a topic instead of the whole thing. Fix those first. Get this layer wrong and the next two are wasted effort.

Layer 2: make every page extractable (the AEO layer)

AI engines lift self-contained passages, so write for extraction, not for scroll depth:

1. Answer first. State the direct answer in the opening one to two sentences of the page and of each section, then support it. Buried answers do not get quoted.
2. Question-form headings. Mirror how people ask ("how do I", "what is", "X vs Y"), so each heading maps to a real query and becomes an anchor an engine can lift.
3. Definitions and tables. A tight 40 to 60 word definition and a comparison table are disproportionately cited, because they are clean units to lift.
4. Cite inside the sentence. Put the number and its source in the same sentence, so the quotable chunk carries its own proof.

The difference is concrete. "We drove strong organic growth for a client" is unquotable: no number, no subject, no source. "At Edstellar, I grew organic traffic 35% in six months" is a standalone claim an engine can lift verbatim, because the who, the number, and the timeframe all sit in one sentence. Rewrite your key pages one sentence at a time with that test: could a model quote this line on its own and still be correct? If not, it is not doing GEO work.

Layer 3: earn citations off-site (the GEO layer)

This is the layer most people skip, and it is the one with the most leverage. Ahrefs' 2025 study of 75,000 brands found that brand mentions correlate with AI visibility far more strongly than backlinks — a 0.664 versus 0.218 correlation, roughly a 3:1 gap — with the top three signals all off-site and YouTube mentions the single strongest. And roughly 85% of AI brand mentions come from third-party pages, not your own domain, per AirOps' 2026 State of AI Search, which puts brands about 6.5x more likely to be cited off their own site than on it. The takeaway is blunt: publishing is half the job, and being mentioned across other people's sites is the other half.

In practice that means three plays, in priority order. First, get into the roundups: the "best SEO consultants in India" or "top GEO tools" listicles are exactly what LLMs quote for "who should I hire" prompts, so pitch to be included in them. Second, answer real questions where models harvest, which today means Reddit, Quora, and niche communities — not a fringe channel, since roughly 48% of AI citations trace back to user-generated and community sources like Reddit, YouTube, and Wikipedia — with a genuine answer and a link only when it helps. Third, earn straightforward mentions: guest posts, podcasts, and being the source another writer cites. None of this is a backlink hack; it is being talked about, which is what an engine corroborates against before it trusts you.

One thing this layer gets wrong is treating a citation as permanent. It is not: AirOps found pages that are not refreshed at least quarterly are roughly 3x more likely to lose the citations they earned. A cited page is a maintained page, not a finished one, so put your earning pages on a refresh cadence and update the visible date when you genuinely revise them.

What the research says to put on the page

The original GEO paper (Aggarwal et al., Princeton and IIT Delhi) tested this directly: adding statistics, quotations, and citations lifted a page's visibility in generative answers by roughly 20 to 41%, with statistics the strongest single move and citations helping lower-ranked pages the most. You can read it here: GEO: Generative Engine Optimization. In practice that means every claim should carry a real, dated number and a linked source, which is also just good writing. The failure mode to avoid is inventing numbers to hit the pattern: engines increasingly cross-check claims, and a single fabricated stat costs you the trust the rest of the page earned.

How to measure it

Rankings and citations are different KPIs, so track both. Stand up a fixed prompt panel: about 15 questions your buyer would actually ask, run monthly across ChatGPT, Perplexity, Google AI Overviews, and Claude, logging whether you are cited and, if so, from which page. Keep the wording of each prompt fixed month to month, or the results are not comparable. Sample prompts for a GEO specialist: "who are good freelance GEO specialists in India", "how do I get my content cited in AI Overviews", "SEO consultant who builds their own tools". That citation share, not your ranking position, is the real GEO scoreboard, and it is the number that maps to whether the off-site work in Layer 3 is landing. I go deeper on the on-page side of getting picked in how to get your brand cited in AI answers.

The mistakes that cost citations

Three patterns undo the work. Blocking the retrieval bots in robots.txt, so you are invisible to the engines you want citing you. Burying the answer under 300 words of preamble, so the model has nothing clean to lift. And publishing without distributing, so the page is technically perfect and mentioned nowhere, which by the numbers above is most of the battle lost. Avoid those three and you are ahead of the majority of sites competing for the same answers.

Where to start

Do them in order: fix the base, make one important page fully extractable, then put real effort into off-site mentions, because that is where the leverage is. Start with your highest-intent page, not your whole site, and give it a month before you judge citation share.

Want this run properly for a brand? See how I run a GEO and AEO audit.

Further reading

• Answer Engine Optimization (AEO)
• GEO: Generative Engine Optimization (the original paper)
• Google Search Central: AI features and your site