SEO optimises a website so search engines rank it. AEO (answer engine optimisation) structures content so an AI can lift a clean answer out of it. GEO (generative engine optimisation) is the off-site work that makes a model confident enough to name your brand. In practice the three overlap heavily, and Google's official position is that all of it is simply SEO. We think Google is mostly right, with one important exception, which is the subject of this article.
What do the three terms actually mean?
SEO, search engine optimisation, is the original discipline: make a site crawlable, relevant and authoritative so search engines rank it in a list of results. Roughly thirty years of accumulated practice, well documented, well understood.
AEO, answer engine optimisation, describes structuring content so a machine can extract a direct answer from it. Clear definitions near the top of a page, question-shaped headings, self-contained passages, FAQ markup, tables. The goal shifts from winning a click to being the text that gets quoted.
GEO, generative engine optimisation, describes the work of getting a generative model to cite and recommend your brand. In practice this leans off-site: mentions across the wider web, how third-party sources describe you, presence in the places models draw on.
Taken together, these three are what we mean by AI visibility: whether AI platforms name and cite your brand when a buyer asks. That is the tidy version. The messy version is more useful.
Where the terms came from
The two acronyms have very different pedigrees, and almost nobody mentions this.
GEO comes from academia. It was introduced in a November 2023 paper, GEO: Generative Engine Optimization, by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, researchers from Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi. It was later published at KDD '24. The paper introduced GEO-bench, a benchmark of 10,000 queries across nine datasets, and tested which content modifications changed visibility in generative responses. Whatever you think of how the term is now used commercially, it began as a measured research question.
AEO comes from marketing. There is no founding paper. It emerged as practitioners started competing for the answer itself rather than the click, and it spread because it was a useful label.
This matters because the two terms get treated as equals in most comparison articles, and they are not. One has a citable origin and a benchmark. The other is a convenient description.
The practical comparison
Look at that table honestly and the overlap is obvious. Nearly everything in the AEO column is good content practice. Nearly everything in the GEO column is what PR and brand marketing have always done. Which brings us to the argument.
Our view: most of the difference is vocabulary
We sell this work, so read the following with that in mind. We still think it is true.
The acronyms are mostly marketing, and the industry has oversold them. There is no separate profession called GEO. There is no secret set of levers that only applies to AI answers. Most of what gets sold as AEO is well-structured writing, and most of what gets sold as GEO is public relations with a new name.
You do not have to take our word for it, because Google has said so directly.
What Google actually says
In May 2026 Google published official guidance on optimising for generative AI features. It addresses the acronyms by name:
From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.
The guide includes a section titled "Mythbusting generative AI search: what you don't need to do". Its list is worth reading closely if you are being sold any of these:
- llms.txt and other special markup. Google does not use them. Creating one "will neither harm nor help" visibility in Google Search.
- Chunking content. No requirement to break pages into small pieces. Google's systems handle multiple topics on a page.
- Rewriting content specifically for AI. Not needed. Models understand synonyms and intent.
- Seeking inauthentic mentions. Explicitly called out as less useful than it appears, with spam systems watching.
- Overfocusing on structured data. Not required for generative AI search, though still worth doing for rich results.
Google also points out that its systems run on retrieval-augmented generation grounded in the normal Search index, and that a page must be indexed and eligible for a snippet before it can appear in an AI Overview at all. In other words, the entry ticket is ordinary SEO.
We think this guidance is correct and overdue. If your agency is selling llms.txt files, content chunking, or bought mentions as an AI visibility strategy, Google has now told you in writing that it does not work on their surfaces.
The exception: Google is only talking about Google
Here is where we part company with the "it is all just SEO, relax" reading.
Read that quotation again and note the first four words: from Google Search's perspective. Google is describing how Google's own AI features work. It has no remit over ChatGPT, Perplexity, Claude, Copilot or Grok, and it makes no claims about them.
Those platforms do not run on Google's index and do not follow Google's rules:
- ChatGPT's sourcing is moving toward Google, but nowhere near enough to rely on. Profound tracked the shift: ChatGPT's alignment with Bing's results collapsed from 26% to 8%, while alignment with Google rose from 12% to 33%. That is a real change, and it makes Google optimisation more relevant to ChatGPT than it was in early 2025. It also means roughly two thirds of what ChatGPT cites is not coming from Google's top results. Necessary, nowhere near sufficient.
- Perplexity leans heavily on community content. A Profound study of 10,000 commercial queries found Perplexity cites Reddit in 46.7% of its responses.
- The platforms barely agree with each other. Averi analysed 680 million AI citations in early 2026 and found that only 11% of domains are cited by both ChatGPT and Perplexity. Qwairy independently arrived at the same 11% figure across 118,000 AI responses (reported by Whitehat SEO). Two different datasets, one conclusion: the source pools these engines draw on are almost entirely separate.
So Google's statement is true and also limited. It answers the question "is there a separate discipline for winning AI Overviews?" with a defensible no. It does not answer "is there work involved in being recommended across every AI assistant your buyers use?", because that was never Google's question to answer.
What is genuinely different, then?
Strip out the hype and the acronyms and something real is left. It is not a tactic. It is a measurement problem and a distribution problem.
Ranking and citation have come apart, measurably, inside Google itself. In March 2026 Ahrefs analysed 863,000 SERPs and 4 million AI Overview URLs and found just 38% of cited URLs ranked in the top ten for the same query. Eight months earlier that figure was 76%. Ahrefs attributes the change to AI Overviews moving to Gemini 3 in January 2026, plus query fan-out. You can accept every word of Google's guidance and still have to explain why your rankings no longer predict your citations.
The strongest signals are off-site, and they are not links. Ahrefs studied 75,000 brands and found brand web mentions, linked or unlinked, correlated with AI Overview visibility at 0.664, against 0.218 for backlinks. A follow-up study found YouTube mentions correlated at roughly 0.737, the strongest factor measured. Note that this does not contradict Google, which warns against inauthentic mentions. Earning genuine ones is the oldest idea in marketing.
Measurement is the actual gap. Rankings do not tell you whether ChatGPT recommends you.
Google announced a Generative AI performance report in Search Console on 3 June 2026, and it is worth understanding precisely, because it is easy to over-read. It is not something you switch on. It is rolling out to a subset of properties, beginning in the UK, so many readers will not see it at all yet. Data begins on 18 May 2026 with no historical backfill. It reports impressions, pages, countries and devices. It does not report clicks, click-through rate, average position or queries.
There is a related control worth knowing about: inclusion in Google's generative AI features is the default, and the opt-out is a separate setting that is also still rolling out. If you manage several properties, check whether an exclusion has been set at domain level, because child properties inherit from their parent. An opt-out applied once, somewhere above you, can quietly cover properties you administer separately.
Even fully deployed, all of this covers Google's surfaces only. Nothing in your existing stack tells you what Claude says about your brand.
That is the honest case for treating this as its own workstream. Not because there are secret tactics, but because there is a large and growing surface where your existing reporting is blind.
What this means for how you spend
Do not buy hacks. llms.txt, chunking, AI-specific rewriting and purchased mentions are either useless or actively risky. Google has now said so in writing, and we agree.
Do fund the fundamentals harder than before. Google's own guidance points at unique, first-hand, non-commodity content as the biggest lever. Their example is instructive: a first-hand review beats a summary of other people's content. That is not a new rule. It is the old rule with more at stake.
Do measure across platforms. Pick twenty five buying questions, run them monthly across ChatGPT, Perplexity, Gemini and Claude, and record whether you appear and how prominently. This costs an afternoon and tells you more than any tool.
Do treat off-site presence as a marketing budget line, not an SEO afterthought. If mentions correlate three times more strongly than links, resourcing should reflect that.
Where we changed our mind
We added an llms.txt file to our site in July 2026, before Google's guidance was on our radar. Google has since confirmed it does nothing for Google Search, in either direction.
We are leaving it up, for one narrow reason: Google's wording allows that other systems may use these files, and we would rather test that than assume. But we have moved it from the "things that help" column to the "cheap experiment with no evidence behind it" column, and we will publish what we find either way. If you were sold an llms.txt file as a meaningful part of an AI visibility strategy, you were oversold.
We would rather write that paragraph than quietly delete the file.
Frequently asked questions
Want to know where you actually stand across every AI platform? Get a free AI Visibility Audit.
Sources
- Google Search Central, "Optimizing your website for generative AI features on Google Search" (published May 2026, last updated 10 July 2026) — https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2024). "GEO: Generative Engine Optimization." KDD '24. arXiv:2311.09735 — https://arxiv.org/pdf/2311.09735
- Ahrefs, "Update: 38% of AI Overview Citations Pull From The Top 10" (863K SERPs, 4M AI Overview URLs, March 2026) — https://ahrefs.com/blog/ai-overview-citations-top-10/
- Ahrefs, "76% of AI Overview Citations Pull From the Top 10" (July 2025 baseline) — https://ahrefs.com/blog/search-rankings-ai-citations/
- Ahrefs, "An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied)" — https://ahrefs.com/blog/ai-overview-brand-correlation/
- Ahrefs, "Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews" (May 2026) — https://ahrefs.com/blog/ai-brand-visibility-correlations/
- Profound, "AI Search Shift: ChatGPT's growing alignment with Google's index" (Bing alignment 26% to 8%, Google 12% to 33%) — https://www.tryprofound.com/blog/ai-search-shift
- Profound, "AI Platform Citation Patterns" (10,000 commercial queries; Perplexity cites Reddit in 46.7% of responses) — https://www.tryprofound.com/blog/ai-platform-citation-patterns
- Averi, cited analysis of 680 million AI citations, early 2026 (primary publication not located; figure widely reported) — https://www.averi.ai/
- Qwairy, AI Citation Analysis (118,000 AI responses, Jan–Mar 2026), as reported by Whitehat SEO, "Perplexity vs ChatGPT vs Gemini: AI Citations" — https://whitehat-seo.co.uk/blog/ai-engines-comparison-citations
- Google Search Central Blog, "Introducing Search Generative AI performance reports in Search Console" (3 June 2026) — https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports
- Google Search Console Help, "Generative AI performance report (Search)" — https://support.google.com/webmasters/answer/16984139