Generative Engine Optimisation (GEO)

Generative Engine Optimisation (GEO) is the practice of structuring and positioning content to be retrieved, parsed, and cited by AI-powered answer engines. It overlaps with traditional SEO heavily, but the goals diverge in places: GEO optimises for citation, not click-through.

Google’s stance: GEO is SEO

Google’s May 2026 AI optimisation guide states its own position plainly:¹ “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” From Google’s perspective, there is no separate GEO discipline. Sites that perform well in traditional search, through quality content, clear structure, and authoritative authorship, perform well in Google’s AI-generated answers without special optimisation. The guide also has a mythbusting section listing five things you can ignore for Google Search: llms.txt and other special markup, content chunking, rewriting content for AI systems, seeking inauthentic mentions, and overfocusing on structured data. Other AI search platforms (Bing, Perplexity, ChatGPT Search) have not issued equivalent guidance and may weight signals differently.

How does citation-focused optimisation differ from ranking-focused optimisation?

Traditional SEO optimises a page to rank in a list of results. The user clicks the result, lands on the page, reads the content, and possibly converts. The page is the destination.

GEO optimises a page to be retrieved by a generative system, which then synthesises an answer from one or more sources and presents that answer (with citations) inside its own interface. The page is no longer the destination, it is a source for synthesised answers.

The implications:

• Click-through rate matters less; citation rate matters more. A page cited in AI Overviews 10,000 times that earns 200 clicks is performing differently than one ranked #3 organically with the same impression volume. Both have value, but the value is different in kind.² The compensating factor comes from Knotch’s own client-panel tracking, quoted inside Conductor’s benchmarks report rather than measured by it: LLM-referred visitors converted at twice the rate of other traffic sources, in a third of the number of sessions.³
• Brand visibility extends beyond the click. When a page is cited, the brand is named in the response even if the user never visits. That visibility has compounding effects on entity recognition.
• Passages, not pages, on the one retrieval path that is documented. Microsoft’s Web IQ, a Bing-based grounding API launched in June 2026, returns “passages and structured evidence objects” as well as documents, and Microsoft grades its results with GDSAT (completeness, freshness, authority).⁴ Microsoft says it grounds Copilot directly and supplies ChatGPT with “some of its web answers”.⁵ Where passage selection applies, the model receives a passage rather than the whole article. Google documents no equivalent and says there is no need to break content into pieces for its systems.¹

How does content get into an AI answer?

There are two routes, and every GEO tactic serves one or the other. The first is training: the model learned about a subject, a brand or a page from the documents it was trained on. The second is retrieval: a search step fetched the page at answer time and handed passages to the model, the grounding step in retrieval-augmented generation. How AI search works covers the mechanics; what matters here is what each route asks of you.

Training rewards frequency across many documents. A 2023 study of pre-training data found that a model’s ability to answer a fact-based question “relates to how many documents associated with that question were seen during pre-training”, with “strong correlational and causal relationships between accuracy and relevant document count”, and that retrieval augmentation “can reduce the dependence on relevant pre-training information”.⁶ For a brand, that means being written about often, by many independent sources, over time, and described the same way each time, because what training stores is association rather than fact (how AI search works explains why). Nothing you do on your own pages changes it much.

Retrieval rewards being indexed, fetchable and quotable: a page the search step can reach, with a passage that answers the question. That is what the on-page work on this page is for.

The two interact, and the interaction is inference rather than anything a platform documents: even a grounded answer is steered by what the model already knows, because the model writes the queries the search step runs and recalls which brands to look for. A brand the model has never encountered can still be retrieved, but it is less likely to be searched for by name.

The balance is per product. Perplexity describes its answers as starting with a search that gathers “information from authoritative sources like articles, websites, and journals”.⁷ ChatGPT “may search the web automatically when your question would benefit from current information”,⁸ and otherwise answers from training. Google’s AI Overviews and AI Mode are “rooted in” its core Search ranking systems, and for products Google says Merchant Center feeds “can help” them be visible in its AI responses, a route that is neither training nor page retrieval.¹

On-page GEO serves retrieval. Off-page GEO serves both, which is why the off-page section below carries the argument it does.

What does the evidence say works?

Most GEO advice is inference from how retrieval systems appear to behave. The evidence that is not inference is short enough to list in full, in three groups: what the platforms say about their own systems, what has been measured, and what nobody has measured.

What the platforms say

Google says optimising for its generative features “is optimizing for the search experience, and thus still SEO”, names five things you do not need (listed under tactics that do not work), and says structured data “isn’t required for generative AI search”.¹ Its helpful-content questions, which the guide says its generative features are “rooted in”, ask whether a page shows “clear sourcing” and whether it is “self-evident to your visitors who authored your content”.⁹ Those are the only content properties Google itself connects to its AI features.

Microsoft says structured content, with schema-marked product pages, FAQs and comparison tables as its examples, “helps AI systems interpret and summarize your content more effectively” and “increases the chances of your being cited” in its answers.¹⁰ That is a statement about structure, with schema as one form of it, and a claim about Microsoft’s own surfaces.

OpenAI’s publisher FAQ covers crawler access, noindex and referral parameters, and says nothing about content patterns.¹¹ Perplexity and Anthropic document nothing either way.

What has been measured

• Content rewrites: four simulated tests, none live, and they disagree with each other. The 2023 paper that coined the term applied eight automated edits (adding statistics, citations, quotes, an authoritative tone and others) to sources the engine had already been given, and the strongest gained 41% share of the answer text.¹² C-SEO Bench ran the same edits through four frontier models and scored where the edited source landed in the citation order: none reliably moved it, and placing a source higher in the model’s context moved it up to 2.8 places.¹³ Two later benchmarks re-ran the edits on open-weight models and each found a different edit helped most, on its own metric.¹⁴¹⁵ Method and scope for each are in the footnotes. C-SEO Bench, peer-reviewed, remains the only test of citation order across frontier models, and it found the order flat. What survives all four is that being retrieved higher matters more than any rewrite.
• An answer placed late in a passage is retrieved less. Across a retrieval pipeline benchmark, the embedding models that do the initial passage matching lost an average of 15.6% of retrieval performance when the relevant information sat later in the passage; keyword matching and rerankers were largely unaffected.¹⁶ Generic pipelines, not a named engine.
• Brand mentions correlate with AI visibility; backlinks barely do. Ahrefs’ December 2025 study of 75,000 brands found YouTube mentions at 0.735 to 0.740 and branded web mentions at 0.656 to 0.709 against 0.19 to 0.24 for backlinks, Spearman correlations running from 0, no relationship, to 1, identical rankings.¹⁷ Correlation only, and brands that get mentioned are usually brands that were already established. Detail under on-page content.
• Engines draw on different source pools. Semrush found in July 2025 that Google’s AI Mode shared around 54% of its cited domains with Google’s own organic top 10, against over 91% for Perplexity,¹⁸ and its November 2024 comparison had found ChatGPT’s results closer to Bing’s than to Google’s.¹⁹ The window predates Gemini 3 and the August 2026 change to ChatGPT’s fan-out, so the divergence is the durable finding and the figure is of its date.
• Listicles dominate commercial citations, and a brand’s own listicle loses. Evertune counted 63% of roughly 400 million citations across six engines going to listicles, on brand and product prompts, in a sponsored write-up.²⁰ SEO specialist Lily Ray’s June 2026 analysis of 100 B2B “best software” queries found Google’s AI Overviews cited a brand’s own “best [category]” list while recommending competitors around 69% of the time.²¹
• Schema coverage shows no relationship with visibility. Search Atlas compared schema coverage with domain visibility across OpenAI, Gemini and Perplexity in December 2025 and found that “higher schema coverage does not reliably lead to higher visibility scores”.²² It is measuring a mechanism no operator documents, so a null result is the expected one.

What nobody has measured

Question-shaped headings, passages written to stand alone, schema as a citation lever, author markup with sameAs links, and the harm from over-structuring. Each is reasoning about how retrieval matches text, or advice a reader benefits from anyway. Neither is evidence. Microsoft’s October 2025 advice for Copilot recommends question-shaped headings and “self-contained phrasing” without data behind either;²³ no other platform statement covers them. Question headings and stand-alone passages are good for readers and cost nothing; schema is worth keeping accurate for rich results, and Microsoft’s statement is about structure rather than markup. Do them for readers, and expect no measured citation effect.

The platform statements and the measurements point the same way: being retrieved matters more than any rewrite, and what Google and Microsoft connect to their answers is ordinary content quality, clear sourcing, visible authorship and structure a reader can use. Precise claims and named sources stay on the list because Google asks for them and readers check them, not because a citation effect has been measured. That is the whole of what is known.

Is GEO only about on-page content?

No, and treating it that way means focusing on only half of the work. Everything above concerns the page itself: its structure, its passages, its attribution. None of that decides whether an AI answer names or cites the brand at all.

Ahrefs’ December 2025 analysis of 75,000 brands found the strongest off-site correlates of AI visibility were not links but mentions. YouTube mentions led, at 0.735 on ChatGPT, 0.736 on AI Mode and 0.740 on AI Overviews, with branded web mentions behind them at 0.656, 0.709 and 0.664, and a weak 0.19 to 0.24 for backlinks.¹⁷ On the Spearman scale those run from 1, the two rankings matching exactly, to 0, no relationship at all. The authors are careful to note that correlation is not causation, and the confound is obvious enough that it should temper any strong claim: brands that get written about are usually brands that are already established, and establishment drives both variables. An earlier Ahrefs study, on only the top 50 domains per platform, found the same correlation was strong for AI Overviews but weak for Perplexity and ChatGPT, so the effect is unlikely to be uniform across platforms.²⁴

Set the correlation aside and a structural argument remains. Each engine draws from its own source pool, and those pools diverge: Semrush found in July 2025 that Google’s AI Mode shared only around 54% of its cited domains with Google’s own organic top 10, against over 91% for Perplexity,¹⁸ and its November 2024 comparison had found ChatGPT’s results closer to Bing’s than to Google’s.¹⁹ That window predates both Gemini 3 and the August 2026 change to ChatGPT’s fan-out, so treat the divergence as the durable finding and the specific overlap figure as of its date. A page you own, however well structured, sits in one index. Coverage across third-party publications reaches source pools your domain cannot, which is why digital PR and brand mentions sit inside a GEO programme and not beside it. The on-page work makes a page citable; the off-page work makes the brand a candidate for citation.

What GEO is not

GEO is not “writing for robots” in the keyword-stuffing sense. Google’s position is that optimising for its generative features “is optimizing for the search experience, and thus still SEO”,¹ and the one controlled test of automated rewrites aimed at a generative engine found they did not improve where a source was cited.¹³ Pages that read as machine-targeted (repetitive phrasing, keyword density above natural language norms, content padding) are worse for the reader, and nothing on this page shows a generative engine compensating for that.

GEO is also not a wholesale replacement for traditional SEO. The two share most of their underlying signals, which is Google’s own position on its systems: a site optimised well for traditional SEO is already most of the way to being well-optimised for AI retrieval. The part that does not overlap, retrieval and citation rather than ranking and clicks, is where the discipline lives.

GEO is also not a strategy for social discovery. Social search on platforms like YouTube, Reddit, and Instagram has distinct signals from traditional search. YouTube SEO, Reddit SEO, and Instagram search each require platform-specific approaches, covered under search everywhere optimisation.

GEO, AEO, LLMO: why there are so many names

The practice described on this page goes by several names. GEO is used here because it has the clearest academic provenance: it was coined in a 2023 research paper by Aggarwal and colleagues examining how content performs in generative answer engines.¹² The others are in common use.

AEO (Answer Engine Optimisation) is the oldest term. It predates generative AI, originally describing optimisation for voice assistants and featured snippets: surfaces that return a direct answer rather than a list of links. Since large language models became prominent in search, AEO has been used more broadly and often interchangeably with GEO. Some SEOs draw a narrower distinction: AEO for Google’s own AI features (AI Overviews, AI Mode), GEO for third-party systems such as Perplexity and ChatGPT. Neither usage is wrong; the overlap is near-total.

LLMO (Large Language Model Optimisation) emphasises the model layer: influencing what LLMs incorporate into their training data, beyond what they retrieve in real time. In practice, the content tactics for LLMO are indistinguishable from GEO. The framing differs; the work does not.

AIO appears in two contexts: as shorthand for AI Overviews (the Google feature), and as a generic label for any AI-related optimisation. The ambiguity makes it imprecise as a standalone term.

AI SEO is a catch-all. It covers the application of SEO thinking to AI search surfaces, which is precisely what GEO and AEO already describe.

None of these represent separate disciplines with separate playbooks. The evidence above supports the same practice under every label: precise claims, named sources and direct answers. The label matters less than the work.

Tactics that do not work

Google’s May 2026 guide has a mythbusting section, “what you don’t need to do”, naming five things you can ignore for Google Search.¹ Those five are marked below. The last two items are not Google’s claims: they are consistent with its guidance but stated here on other grounds.

llms.txt files and other special markup. (Google’s guide.) Creating machine-readable files is unnecessary for Google’s AI systems, which parse web pages directly. Google adds that maintaining one “will neither harm nor help” visibility, because Search ignores it. Do not publish expecting SEO or GEO benefits in Google. Anthropic publishes its own llms.txt for documentation purposes and co-developed the llms-full.txt format with Mintlify; whether Claude reads other sites’ llms.txt at inference time is unconfirmed.²⁵

Content chunking. (Google’s guide.) There is no requirement to break content into small pieces for AI to understand it, and Google states there is no ideal page length. Artificially fragmenting paragraphs to trigger retrieval signals poor editorial quality and reads unnaturally.

Rewriting content just for AI systems. (Google’s guide.) Google’s systems understand synonyms and intent, so there is no need to capture every phrasing a user might try. In its words, “you don’t need to write in a specific way just for generative AI search”.¹

Seeking inauthentic mentions. (Google’s guide.) Google’s generative features do surface what is said about a brand across the web, but it names manufactured mentions specifically as less useful than they appear, because core ranking depends on content quality and separate systems block spam. This is the boundary on the off-page argument above: coverage earned because the work is worth covering counts, and coverage manufactured to be counted does not.

Overfocusing on structured data. (Google’s guide.) Structured data is not required for generative AI search and there is no special schema.org markup to add, though Google recommends continuing to use it for rich result eligibility. The separate failure mode is schema that misdescribes the page: marking up content as FAQ or HowTo when there is no genuine FAQ or procedure adds noise rather than eligibility.

Over-structuring content. Not named by Google. Breaking every paragraph into lists or forcing every heading into a question produces poor readability, and Google’s own framing throughout the guide is that structure should serve human readers. Reshape a heading where the section genuinely answers that question, and leave it alone where it does not.

Volume without substance. Google addresses this outside the mythbusting list, in its content guidance: publishing at scale to capture more surface “violates Google’s scaled content abuse spam policy”, and “a high quantity of pages doesn’t make a website higher quality or more relevant to users”.¹ Scale without human review optimises against the signals that earn citations.

The common thread: tactics treating GEO as a separate game to be won independently of content quality will underperform. The sites earning consistent AI citations are those that would have ranked well in traditional search anyway. Understanding how AI search works explains why AI content risks, including hallucination and E-E-A-T erosion, matter for citation as well as for ranking.

How do you start with GEO?

Four steps have evidence behind them. What is left out is named at the end.

1. Open each section with the answer. For your top pages, check whether each major section states its answer in the first sentence. The position-bias measurement is the reason: retrievers that match passages lose ground when the answer sits later.¹⁶
2. State claims precisely, name the source, and sign the page. Specific figures over vague quantities, direct quotations from named sources, and attribution for every checkable claim. Google’s helpful-content questions ask for “clear sourcing” and visible authorship, and its generative features are “rooted in” the systems those questions describe.⁹ The 2023 GEO paper found the same edits raised a source’s share of an answer and the 2025 benchmark found they did not move its citation position,¹²¹³ so do this for the reader and for Google’s stated criteria, not for a measured citation lift. A visible byline and an author page answer the authorship question.
3. Earn coverage off-site. Audit where the brand is mentioned across the publications the engines cite, then close the gaps through digital PR, original research and expert commentary. Mentions are the strongest off-site correlate measured, and each engine’s source pool reaches places your own domain cannot.¹⁷¹⁸ This step is slow and often skipped, because it sits outside the SEO team’s direct control.
4. Track citations alongside rankings. In GSC, the Generative AI performance report gives impressions from AI Overviews and AI Mode by page, country, device and date, available to all properties since 31 August 2026; it carries no clicks and no queries.²⁶ For the click side, monitor CTR against stable impressions in the standard Performance report: a falling CTR is the signal that an AI Overview is absorbing clicks. Monitor AI-driven traffic in GA4’s AI Assistant channel, which captures ChatGPT, Claude, Gemini and Perplexity automatically.²⁷ Supplement with manual sampling across Perplexity, ChatGPT and other surfaces where referrer data is incomplete.

Left out, deliberately: reshaping headings into questions, adding schema, and author markup. None has a measurement or a platform statement behind it as a citation lever. Keep existing structured data accurate, because Google withholds rich results where markup does not match the page,¹ and treat the rest as reader-facing craft rather than GEO work.

The GEO question for any piece of content

When deciding whether to publish, expand, or rewrite a page, ask: would a retrieval system extracting a passage from this page produce something a user would find useful and trustworthy? If yes, the page is GEO-ready. If the best passage requires several paragraphs of context to make sense, restructure until a stand-alone passage exists.

Frequently asked questions

What is the difference between GEO, AEO, LLMO, and AI SEO?

They largely describe the same practice under different names. AEO (Answer Engine Optimisation) is the oldest, originating in the voice search and featured snippet era. GEO (Generative Engine Optimisation) is more precise, coined in academic research to describe optimisation for systems that synthesise answers from retrieved content. LLMO (Large Language Model Optimisation) emphasises influencing LLM knowledge rather than real-time retrieval; in practice the tactics are nearly identical to GEO. AIO and AI SEO are broader terms with no distinct practice behind them. The tactics are the same whichever label is used.

Is GEO a real discipline or just SEO rebranded?

Both. On Google, the underlying signals are shared with traditional SEO by Google’s own account: its generative features are “rooted in our core Search ranking and quality systems”.¹ Other platforms have not said the same. The framing, measurement, and content patterns are different enough to warrant separate language. Treat GEO as the part of SEO concerned with retrieval and citation in generative systems, not as a replacement.

Can I optimise for one AI engine without affecting others?

The content work transfers; the source pools do not. Writing that is accurate, well structured and clearly attributed helps on every surface, so there is little reason to produce engine-specific versions of a page. What does not transfer is which sources each engine draws on: they overlap only partially, and the off-page correlates differ by platform, as the sections above set out. Expect the on-page work to carry across and the visibility it produces to vary by engine.

Does GEO require sacrificing traditional SEO performance?

Rarely. Most GEO improvements (direct answers, named sources, visible authorship) also help traditional rankings. Cases where the two diverge meaningfully are uncommon.

Footnotes

1. Optimizing your website for generative AI features on Google Search — Google Search Central ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰

2. AI Overviews Reduce Clicks by 58% — Ahrefs ↩

3. 2026 AEO/GEO Benchmarks Report — Conductor, a live page marked “last updated 6 July 2026” when read on 21 September 2026. The conversion figure is quoted in it from Andrew Bolton of Knotch, about Knotch’s own clients. ↩

4. Announcing Microsoft Web IQ — Bing Search Blog ↩

5. Microsoft Web IQ (Microsoft names “AI platforms like Copilot and OpenAI”), with Jordi Ribas, President of Search and AI, quoted in Microsoft releases Web IQ, powered by Bing but designed for how AI-agents search — Search Engine Land ↩

6. Large Language Models Struggle to Learn Long-Tail Knowledge — Kandpal, Deng, Roberts, Wallace and Raffel, arXiv 2211.08411. ICML 2023. Accuracy on fact questions measured against the count of pre-training documents mentioning the question’s entities, across model sizes up to 176B parameters. ↩

7. How does Perplexity work? — Perplexity Help Center ↩

8. ChatGPT search — OpenAI Help Center ↩

9. Creating helpful, reliable, people-first content — Google Search Central. Google’s AI optimisation guide says its generative features “are rooted in our core Search ranking and quality systems”, which is the link between this guidance and AI visibility. ↩ ↩²

10. How AI Search Is Changing the Way Conversions are Measured — Bing Webmaster Blog, Fabrice Canel and Krishna Madhavan, 20 November 2025. ↩

11. Publishers and developers FAQ — OpenAI Help Center. As of 21 September 2026 the page covers OAI-SearchBot access, noindex and referral parameters, and does not mention content patterns or structured data. ↩

12. GEO: Generative Engine Optimization — Aggarwal et al., arXiv 2311.09735, KDD 2024; v3, June 2024. ↩ ↩² ↩³

13. C-SEO Bench: Does Conversational SEO Work? — Puerto, Gubri, Green, Oh and Yun, arXiv 2506.11097. NeurIPS Datasets & Benchmarks 2025, NAVER-funded. A simulated engine: gpt-4o-mini, Claude 3.5 Haiku, o3 and o4-mini answering from a fixed document list, scored on citation order. ↩ ↩² ↩³

14. CC-GSEO-Bench: A Content-Centric Benchmark for Measuring Source Influence in Generative Search Engines — Chen et al., arXiv 2509.05607. Technical report, v2 December 2025; an offline simulator with cached retrieval and gpt-oss-120b and gpt-5-nano as generators, 5,353 test instances, nine rewrite strategies mirroring the GEO paper’s, scored 0 to 10 by an LLM judge. ↩

15. GEO-Bench: Benchmarking Ranking Manipulation in Generative Engine Optimization — Nimase, Chen, Qi, Zhao and Hu, arXiv 2605.29107. Preprint, v2 May 2026; all ten C-SEO Bench strategies rewritten with GPT-4o-mini and scored for rank promotion against Llama-3.1-8B-Instruct on five datasets; authoritative rewriting reached the highest normalised rank gain, 0.83 on the C-SEO Bench data and 0.00 on another. ↩

16. An Empirical Study of Position Bias in Modern Information Retrieval — Zeng et al., arXiv 2505.13950 (EMNLP 2025 Findings), EMNLP 2025 Findings. Two constructed benchmarks, not a live engine; the 15.6% average drop covers dense embedding and ColBERT-style late-interaction models. ↩ ↩²

17. Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews (75k Brands Studied) — Ahrefs. Spearman correlations, December 2025. The authors explicitly caution that correlation is not causation. ↩ ↩² ↩³

18. How Google’s AI Mode Compares to Traditional Search and Other LLMs — Semrush. 5,000 keywords, 150,000+ unique citations, July 2025. ↩ ↩² ↩³

19. SearchGPT vs. Google and Bing: A Quick Comparison of Search Engines — Semrush, 13 December 2024: 5,000 US keywords queried in November 2024; ChatGPT Search shared an average of 3.5 top-10 domains with Bing and 2.5 with Google. ↩ ↩²

20. AI search loves listicles: What 25,000 URLs reveal about citations — Evertune, via Search Engine Land. Sponsored content; the 6,000 most-cited URLs per model, March and April 2026, from Evertune’s own brand-tracking panel. ↩

21. Why Calling Yourself the ‘Best’ Could Be Helping Your Competitors Win in AI Search — Lily Ray, 17 June 2026: Ray’s own analysis of 100 B2B “best [category] software” queries, with AI Overview answers and cited sources pulled through Ahrefs Brand Radar on three dates between 15 April and 8 June 2026. ↩

22. The Limits of Schema Markup for AI Search — Search Atlas, December 2025. Schema coverage from extracted HTML against a separate domain-visibility dataset across OpenAI, Gemini and Perplexity. ↩

23. Optimizing Your Content for Inclusion in AI Search Answers — Microsoft Advertising, Krishna Madhavan, Principal Product Manager, Microsoft Bing, 8 October 2025. ↩

24. Google Seems More Biased Towards Big Brands Than ChatGPT and Perplexity — Ahrefs, Patrick Stox, 7 July 2025. Spearman correlations across the top 50 domains in each system, based on ~76.7M AI Overviews and roughly 950,000 prompts each for ChatGPT and Perplexity, June 2025. ↩

25. Mintlify: The value of llms.txt ↩

26. Introducing Search Generative AI performance reports in Search Console — Google Search Central. Carries Google’s note: “As of August 31, 2026, we’ve rolled out these insights to all websites worldwide.” ↩

27. Google Analytics AI Assistant traffic measurement roll-out — Google Analytics LinkedIn ↩