SEO (search engine optimisation) is the work of making a site findable and chosen in search results. AEO (answer engine optimisation) is the older name for shaping content so a system can lift a direct answer from it: featured snippets, voice assistants and now AI answers. GEO (generative engine optimisation) is the newest: improving how often and how prominently a source appears inside answers written by generative AI. In practice they overlap almost entirely. Google says optimising for its AI features "is still SEO". The genuine differences are narrower than the marketing suggests, and this article sets them out with sources.
The question matters because the acronyms are now sold as separate services. What a business owner needs is not three definitions but a list of what would actually be done differently, and what is the same work under a new label.
Where each term came from
SEO
Search engine optimisation has been the standard term since the late 1990s. It covers three layers: technical (can a search engine fetch, render and index the page?), content (does the page answer the query better than the alternatives?) and authority (do other credible sources refer to it?).
AEO
"Answer engine optimisation" grew out of the featured-snippet and voice-search period of the late 2010s, when Google began answering questions directly at the top of results and smart speakers read out a single answer. I could not find a reliable source for who first used the term; it appears to be industry vocabulary rather than anyone's invention. Its core idea is concrete, though: write so that a question has a short, self-contained answer a machine can extract.
GEO
GEO has a precise origin. It is the title of a research paper, "GEO: Generative Engine Optimization", by Pranjal Aggarwal (IIT Delhi), Vishvak Murahari, Karthik Narasimhan and Ameet Deshpande (Princeton) with Tanmay Rajpurohit and Ashwin Kalyan. It was first posted in November 2023 and accepted to the KDD 2024 conference. Because most GEO advice traces back to this one paper, it is worth knowing what it actually tested.
What the GEO paper actually found
The researchers built a benchmark of 10,000 queries, GEO-bench, and tested nine ways of rewriting a source page to see which made a generative engine quote it more. They measured visibility as the share of the AI's answer drawn from a source, weighted by position, plus a model-judged "subjective impression" score.
| Method | Effect on visibility (simulated engine) |
|---|---|
| Adding quotations from relevant sources | Among the best, roughly 30–40% gain |
| Adding statistics | Among the best |
| Citing sources | Among the best |
| Fluency and easier wording | Moderate gains |
| Keyword stuffing | Worse than doing nothing |
The headline "up to 40%" figure comes from this. Three details in the paper are almost never repeated, and they change how it should be read:
- The gains went mostly to weaker sources. Citing sources lifted the visibility of sources ranked fifth by 115%, while sources ranked first lost about 30% on average. GEO, as tested, is a redistribution, not a rising tide. It helps the underdog at the expense of whoever was already quoted.
- It measured share of the answer, not traffic. Being quoted more in an AI answer is not the same as being clicked, and the paper does not claim it is.
- It was tested on a simulated engine and on Perplexity in 2023. The authors themselves note that engines change. On Perplexity the effects were smaller but in the same direction: quotations helped, keyword stuffing hurt.
Read that way, the paper's practical advice is unremarkable and sound: put evidence, numbers and attributed sources in your content, and do not stuff keywords. That is also what good editors have always asked for.
What Google, Microsoft and OpenAI now say
Google has been the most explicit. Its Search Central page on AI features says there are no additional requirements to appear in AI Overviews or AI Mode: a page must be indexed and eligible to show with a snippet. Its 2026 guide to generative AI features goes further, saying that optimising for generative AI search is optimising for the search experience, "and thus still SEO", and that no AI-specific files, schema or content chunking are needed. Danny Sullivan and John Mueller of Google repeated the point on the Search Off the Record podcast in late 2025.
Microsoft takes a slightly different tone. A Bing product manager's October 2025 guidance keeps every SEO fundamental (crawlability, metadata, internal links, backlinks) and then recommends structure that helps AI extraction: clear H2 and H3 sections, question-and-answer blocks, lists and tables, schema, and precise, measurable claims.
OpenAI documents only access: a site appears in ChatGPT search if its crawler, OAI-SearchBot, is allowed in robots.txt. Perplexity says the same for PerplexityBot. Neither publishes ranking guidance.
So the official position of every company involved is that the foundation is ordinary SEO. Microsoft adds some formatting advice, which is AEO in all but name.
The real differences
Strip away the vocabulary and there are four things that genuinely differ between optimising for a list of links and optimising for a generated answer.
| Classic search results | AI-generated answers | |
|---|---|---|
| Who reads your page | Googlebot and Bingbot | Also OAI-SearchBot, Claude-SearchBot, PerplexityBot, each with its own robots.txt token |
| Unit that wins | The page | A passage or a fact inside the page |
| Overlap with Google's top 10 | By definition | Partial: about 38% of AI Overview citations came from the top 10 in one 2026 study |
| How you measure it | Rankings, impressions, clicks | Citations and mentions, which most tools estimate by sampling prompts |
1. More crawlers to allow. Google's AI features use Googlebot. Every other AI search product uses its own crawler, and a site that blocks them, often by accident through a security plugin or CDN setting, cannot appear. In my September check, 27 of 34 Nepali websites had no rule for any AI crawler, and one had blocked the wrong ones.
2. Passages matter more than pages. An AI answer quotes a sentence or a figure, not your whole article. A page that states its key fact plainly, near a heading that names the question, is easier to use. This is the AEO idea, and it is the main thing that is genuinely new for most sites.
3. Ranking first is not enough, and not required. Ahrefs found in August 2025 that only about 12% of URLs cited by ChatGPT, Gemini, Copilot and Perplexity for 15,000 prompts ranked in Google's top 10 for the same prompt. A March 2026 update put the share of AI Overview citations from the top 10 at about 38%. These are vendor studies with their own methods. The direction is consistent with the GEO paper, though: AI systems draw from deeper in the results, which is an opening for smaller sites with specific, well-evidenced pages.
4. Measurement is weaker. Search Console reports AI Overviews and AI Mode traffic inside the ordinary "Web" search figures. For ChatGPT, Claude and Perplexity there is no equivalent console; you see referral visits in analytics and little else. Any "GEO dashboard" is estimating from a sample of prompts it chose. That is useful for trends, but it is not a ranking report.
What does not change
Everything that makes a page eligible in the first place. A generated answer cannot cite a page that was never indexed, cannot quote a fact the page does not state, and has little reason to trust a site that nobody else refers to. The failures I find most often on Nepali websites are all at this level: 41% of the 34 sites in my August audit had no H1, nine had neither an H1 nor any structured data, and most stated nothing in Nepali at all. None of that is fixed by a new acronym.
Why this matters more in Nepal
For many local questions, such as permits, hospital departments, admission dates or bank charges, there are few strong sources online, in English or in Nepali. An AI system answering those questions is choosing from a shallow pool. The GEO paper's finding that lower-ranked sources gain the most from clear evidence is, in that setting, good news: a well-structured page from a small Nepali business can be the best available source more easily than it could in a crowded market.
It also means the price of the basics is higher. A site that is unreachable from abroad, written only in images, or typed in a legacy font cannot be that source however it is "optimised".
What to actually do
- Fix eligibility first. Indexable pages, one clear H1, real text rather than images, structured data that matches the page.
- Decide your AI crawler policy in robots.txt, token by token, and check your CDN is not overriding it.
- Answer early. Put the direct answer to the page's main question in the first paragraph, then explain.
- Write question-shaped headings where the reader really has a question, and answer each in the first sentence below it.
- State facts with numbers, dates and sources. This is the one GEO tactic with research behind it, and it also earns links from people.
- Earn mentions. Original data, as on this site's research page, gets cited by journalists and other writers, and those citations are what AI systems see too.
- Measure honestly. Use Search Console for Google, referral data in analytics for the rest, and treat AI visibility tools as samples.
If you are offered GEO as a separate service, ask which of these seven they would do that your SEO provider would not. The honest answer is usually item 2, and perhaps some rewriting for item 3. Both are worth doing; neither is a new discipline.
Questions people ask
Is GEO replacing SEO?
No. Every company that runs an AI search product either says GEO is SEO (Google) or builds its advice on SEO fundamentals (Microsoft). GEO describes a new place where results appear, not a new set of rules.
What is the difference between AEO and GEO?
AEO is about being the extracted answer: one passage chosen and shown. GEO is about being one of several sources woven into a generated answer. In practice the same writing serves both.
Do I need special schema or files for AI search?
Google says no. Structured data still helps systems understand what a page is about, as long as it matches the visible content. Files such as llms.txt have no documented effect on Google, ChatGPT, Claude or Perplexity.
How do I know if AI search cites my site?
For Google, AI Overviews and AI Mode traffic is counted in Search Console's Performance report. For other assistants, look for referrals from chatgpt.com, perplexity.ai, claude.ai and copilot.microsoft.com in your analytics, and test the questions your customers ask.
Does keyword stuffing help in AI answers?
The GEO paper found it made visibility worse than doing nothing, on both the simulated engine and Perplexity.
Sources
- Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, GEO: Generative Engine Optimization, arXiv 2311.09735, November 2023, revised June 2024; KDD 2024.
- Google Search Central, AI features and your website, accessed 5 November 2026.
- Google Search Central, Optimizing your website for generative AI features on Google Search, accessed 5 November 2026.
- John Mueller, Top ways to ensure your content performs well in Google's AI experiences on Search, Google Search Central Blog, 21 May 2025.
- Krishna Madhavan, Optimizing your content for inclusion in AI search answers, Microsoft Advertising blog, 8 October 2025.
- OpenAI, Overview of OpenAI crawlers; Perplexity, Perplexity crawlers, both accessed 5 November 2026.
- Ahrefs, Only 12% of AI-cited URLs rank in Google's top 10, 11 August 2025; 38% of AI Overview citations pull from the top 10, 2 March 2026 (vendor research).
Featured photo: Dr Duncan Pepper, Wikimedia Commons, CC BY-SA 2.0. Cropped to 16:9.
