What is answer engine optimization (AEO)?
How answer engines pick a source and what to check on your pages.
On this page
Short answer
Generative engine optimization (GEO) means getting a business into the replies that AI assistants write: ChatGPT, Perplexity, Gemini, Google’s AI Overviews. A search results page offers ten links to choose from. An assistant offers one composed reply, and a brand either appears in it or does not. GEO is about appearing in it, as a named option or as a quoted source.
The name was coined in a 2023 paper, GEO: Generative Engine Optimization, written by a team from Princeton University, IIT Delhi and two independent authors. They rewrote source texts in controlled ways and measured how visible each source became in AI-written replies. Three edits helped most: pointing to sources, adding quotes, adding numbers. The paper reports up to 40% higher source visibility in their tests.
What happens inside the assistant
With web search switched on, a reply usually goes through three stages:
- The assistant turns your question into one or more searches and opens a small set of pages.
- It pulls facts, lists and recommendations out of them.
- It writes the reply and, in most products, shows which pages it used.
That gives a business two routes in. Either one of its own pages becomes a source, or it is named on the pages the assistant opens.
We watched the second route at work on October 19, 2026. Through a ChatGPT scraping service (web search on, US location) we asked which GEO agencies suit small and mid-size companies. The reply named six agencies and leaned on four “best agencies 2026” round-ups plus the agencies’ own sites. Two agencies appeared with a starting price, and one of the four round-ups had been published by an agency that ranked itself first. A single question proves nothing on its own, yet the lesson is practical: round-up articles carried a lot of weight, and so did a plain service page with a price on it.
Search results versus written replies
| Classic search results | AI-written reply | |
|---|---|---|
| What the user sees | ten links in order | a paragraph or a short list |
| What a business wants | a high spot | a mention or a quote |
| What helps | relevance, links, page quality | the same, plus being named by sources the assistant trusts |
| What to count | rank, clicks, impressions | how often you are named across a fixed list of questions |
| How much it moves | little from day to day | the same question can produce a different reply each time |
There is a lot of common ground. A quick, clear page that already ranks well is often among those the assistant opens. GEO adds two layers on top: work on other people’s sites and a different way of counting results.
Your own site: three checks beyond page layout
Page layout, such as putting the answer first and using tables, is covered in our AEO checklist. For GEO, check three more things:
- One set of facts
Your company name, what you sell, your city and founding year should read the same on your site, in directories and on your social profiles.
Read more
Mixed versions make it harder for an assistant to tie the mentions together.
- Markup
Schema.org data for the organization, its services, articles and breadcrumbs gives search engines a clean outline of each page.
- Bots
Make sure robots.txt lets in OAI-SearchBot, PerplexityBot and Googlebot, the crawlers that feed search-based replies.
Read more
GPTBot, ClaudeBot and Google-Extended relate to model training and some assistant features; whether to allow them is your call. Shutting out Google-Extended will not take you out of AI Overviews, because those come from Googlebot.
Either one of its own pages becomes a source, or it is named on the pages the assistant opens.
Work on other sites
Since replies are built from pages you do not own, part of the job happens off your site:
- Round-ups. Look up the “best X for Y” pieces that appear for your buyers’ questions and invite their authors to look at your company.
- Listings. Keep your entries on industry directories and review platforms complete and in line with your site.
- Comment for journalists. Reply to press requests in your field; trade publications sometimes end up among the sources.
- Communities. On Reddit and similar places, give useful replies where they fit, without pitching.
Counting results
One check means little, because the same question can get a different reply an hour later. A method that holds up:
- List 20 to 30 real questions your customers ask before buying.
- Put every question to each assistant you track three times.
- Note whether your company came up, which competitors came up, and which pages were linked.
- Do it again each month with the identical list and compare.
We tried this on our own website, using Russian-language queries in Almaty, Kazakhstan. On September 22, 2026, Google’s AI Overviews (desktop results) linked to our site for 0 of 17 tracked queries; on October 6 it was 2 of 20 (three queries had been added). The two links on October 6 were for “mobile videographer” and “what is a mobile videographer”. For “what is a mobile videographer” and “local SEO” the link showed up at the next weekly check, 1 and 7 days after we rewrote the page so its opening paragraph answered the question. For “mobile videographer” it took 8 days. By October 6 the “local SEO” link was gone again, a reminder of how unstable these citations are. There was no control group, so we read this as a pattern on one site, not as proof.
Over the same two weeks we saw no growth in mentions by ChatGPT and Perplexity. In September we ran each question once and from October 6 three times, so those counts do not line up exactly.
Mistakes we see
- Reading too much into one reply. Repeat every question.
- Polishing only your own site when the reply is drawn from other people’s pages.
- Writing for machines only. Visitors from Google still need a page that makes sense to them.
- Relying on one score without looking at the actual replies and the pages behind them.
- Wanting a guarantee. No one decides what an assistant will say.
Look up the “best X for Y” pieces that appear for your buyers’ questions and invite their authors to look at your company.
Questions people ask
Are GEO, AEO and AI SEO different things?
People mix the terms freely. AEO, answer engine optimization, tends to mean shaping a page so a short answer can be lifted from it. LLM SEO adds what a model knows from training. GEO reaches further: it covers mentions on other sites and tracking what assistants say over time.
Will GEO replace SEO?
No. Many assistants find pages through a search index, so a page nobody can find in search rarely becomes a source. Think of GEO as a layer that sits on a sound SEO base.
How quickly does it work?
In our own test, AI Overviews started linking to rewritten pages after 1 to 8 days, and one of those links had disappeared a week later. Being named by ChatGPT or Perplexity depends more on other sites, and we have no measured timeline for that yet.
Can a company do this without an agency?
Yes. The counting method above needs a spreadsheet and some hours each month. Outside help makes sense when you want it done monthly across many questions and assistants, together with the off-site work. See our AI search optimization service, or our white label offer if you run an agency.
