AEO (answer engine optimization) is the work of getting ChatGPT, Perplexity, Gemini or Google's AI-generated summaries to use your content to build what they answer and to cite your brand as a source. A few weeks ago I saw it show up in the most literal way possible in this blog's Search Console: a 19-word query, "how can I show up in ChatGPT's answers when a buyer is evaluating options in my category?" (written in Spanish), with 13 impressions.
Nobody types like that to find a keyword. That is how you talk to an assistant. And that way of asking is exactly what forces us to rethink how we do content and SEO.
My take, after reading Google's official guide and reviewing which sources AI answers lean on, is fairly simple: getting an AI to cite you takes almost the same things as ranking (being indexed, having the most specific page for the question and having other sites talk about you), and what is really new about AEO is that you work on it and measure it by full questions instead of keywords.
What is AEO?
The term answer engine optimization predates LLMs. It started with Google's featured snippets (that box with the direct answer above the results) and with voice assistants like Siri or Alexa, which read a single answer out loud. If the assistant read only one result, you had to be that result.
Today AEO is used for a larger group of surfaces, the so-called answer engines:
- AI Overviews and AI Mode, the two AI experiences inside Google Search.
- ChatGPT with search, which queries the web and cites its sources with links.
- Perplexity, which cites the sources for each claim.
- Gemini, Google's assistant, and Copilot, Microsoft's.
- Featured snippets and voice search, which are still there.
One distinction that helps me a lot when working on AEO with a content team: some of these surfaces search the web live before answering (AI Overviews, AI Mode, OpenAI's search and Perplexity) and others answer from what the model learned in training. You can work on the first group today, because the citation depends on what they find. With the second group the work is slower: it depends on how much, and how, your brand was talked about online over the years.
How does an answer engine work?
Here comes the slightly more technical part (I promise it's short).
When you ask one of these answer engines a question, the model does not read the whole internet. It takes what you wrote, breaks it into several smaller queries and runs them against an index. Google calls that set of related searches the model generates in parallel query fan-out. Google's AI Mode works this way, and OpenAI's assistant does something similar with its background queries.
With what comes back from those queries, the model picks sources, builds its answer from that content and cites some of them. In other words, before deciding whom to cite, it decides what to read. If your site does not show up in those background queries, it is not in the game.
Two data points that change the diagnosis quite a bit:
- In its official guide, Google says a page must be indexed and eligible to be shown with a snippet. Without indexing there is no AEO in that search engine.
- A Peec AI analysis of more than 10 million prompts sent to OpenAI's assistant found that 66% of prompts written in Spanish (from Spanish IPs) triggered at least one background query in English. If your competitors have English content and you don't, part of the sources were chosen before your site even entered the race (I covered this in more detail in the article on ChatGPT fan-out queries).
How is AEO different from SEO?
The difference is in what counts as winning. In traditional SEO you win when someone clicks your result. In AEO you win when the answer uses your information and names you, even if nobody clicks.
| SEO | AEO | |
|---|---|---|
| Goal | Rank a URL in the results | Get the AI's answer to use your content and cite you |
| Typical query | Two to four keywords ("car insurance") | A question with context ("which car insurance makes sense if I only drive on weekends?") |
| Unit of work | The whole URL | The passage that answers a specific question |
| Main metric | Clicks, impressions and position | Mentions, citations and share of voice against competitors |
| What you get | A visit | Being named in the decision, with or without a visit |
Here is what that difference looks like with an example of my own. In the last 90 days, the article on Google AI Mode appeared 32,277 times for the query "modo ia" and got 7 clicks, at an average position of 8.7. If I only look at clicks, that article does not exist. If I look at where that content appears and for which queries, it tells a different story: Google is showing it as a possible answer thousands of times a month, and that is SEO work too. The click metric is still useful, but it no longer explains all of your visibility.
So, does AEO replace SEO? No. SEO is the foundation everything else runs on: the engine has to find, crawl and index your site before it can cite it, and the authority you built over the years counts here too. What changes is the content strategy, the kind of answer you aim to give and the way you measure.
Are AEO and GEO the same thing?
In practice, yes. GEO stands for generative engine optimization, and you will also read LLMO or "SEO for AI". Some people use AEO for one-off answers and GEO for longer conversations, but the tasks are the same and so are the teams doing them.
Google said it fairly directly in its guide to generative AI features: it names AEO and GEO and concludes that optimizing for generative AI search "is still SEO". I think that is good news for any marketing team: AEO work fits inside what the team already does well, with one extra way of measuring.
How do you do AEO?
This is what I do when a marketing team asks me where to start with an AEO strategy. I ordered it by impact, and the things that sound most modern ended up at the end.
-
Start with real questions. Before optimizing a single page for AEO, gather what your customers actually ask: long queries in Search Console, what the sales team hears over and over, social media comments and what competitors already answer. Something like "which invoicing software works for a small business with 3 locations?" is worth more than the keyword "invoicing software", because that is how buyers talk to artificial intelligence.
-
Answer at the top and be specific. Put the direct answer in the first lines of the section, with the concrete data point, and explain afterwards. AI, like a person in a hurry, gets more out of two lines with a fact than three paragraphs of context. Specificity matters more than it seems: in a study I ran on 500 prompts (published on Previsible in June 2026), queries that included an industry, a team size, a year or a use case received on average 3.4 times more citations than their generic versions. If your content speaks to "companies" in general, it competes with the whole internet for the same direct answer. If it speaks to "insurers in Chile with an online quote tool", it competes with very few.
-
Get others to talk about you. In that same study, about 67% of citations went to niche domains rather than big media outlets, and when someone asked what options existed in a category, 71% of citations went to comparison lists. Mentions of your brand on industry sites, comparisons, associations and partners give the AI the confirmation your own domain cannot, which is basically what SEO has always called authority.
-
Make sure the engine can read you. Indexing, crawling and a robots.txt that does not block the bots you care about. It sounds boring, and it is the first thing I check in an AEO strategy, because a page that is not in the index cannot answer anything.
-
Update what matters. In the same study, about 78% of the content cited for commercial queries had been published or updated in the previous six months (89% in fast-moving categories). A 2023 pricing page competes poorly against one from this year, even if it is better written.
Do you need schema, llms.txt or chunking?
There is a fair amount of folklore here. You will read guides saying that without schema AI won't understand you, that you have to cut every article into blocks of 100 to 200 words, or that you need an llms.txt file to optimize your site.
Google's guide to generative AI features (updated in July 2026) answers all three:
- Google says llms.txt and other special files or markup are not needed to appear in its search engine, and that creating them neither helps nor harms your visibility there.
- There is also no requirement to split your content into small pieces (the famous chunking) for AI to understand it better.
- Structured data is not a requirement for generative AI search, and there is no special schema.org markup you need to add.
Structured data is still useful for what it was always useful for: rich results. But when someone pitches schema as the fix for an AI visibility problem, I remember that I once said schema is the horoscope of GEO (it works for everything, depending on who reads it). I would first check whether your content can answer better than everyone else's, because that weighs a lot more in AEO.
How do you measure AEO?
It is measured by questions instead of keywords, which makes it look more like a survey than an SEO ranking. The method I use to measure AEO has three parts.
First, a fixed corpus of 100 to 300 prompts about your content's topics that mimic how your buyer asks at each stage (discovery, comparison and decision), run every month on the same engines under the same conditions. Second, five fields for each answer: whether you appear, where, how you are described, which competitors appear and which sources are cited. Third, four indicators: share of voice (in what percentage of answers you appear compared with competitors), description accuracy, source map and referral traffic.
For the Google side, Search Console now has a generative AI performance report with impressions and clicks on those surfaces (I explained it here). And in GA4 you can separate visits coming from chatgpt.com or perplexity.ai. There are paid tools that automate prompt tracking, but you define the corpus and the criteria.
How long does it take to see AEO results?
I would be wary of any exact timeline for AEO, just like in SEO. What I do know is what it depends on: your URL being indexed and crawled, a real question existing for which your content is the best answer, and how much other sites talk about you. On a surface that searches live, a well-indexed new article can start showing up soon. Changing what a model "knows" about your brand takes much longer, and no strategy controls what each engine decides to cite.
What I would do this week
Gather 20 real questions from your customers (Search Console, sales and comments all work). Ask them on three different engines (ChatGPT, Gemini and Google Search) in a clean session and write down who shows up. Then check whether you have a page that answers each one better than what got cited. The ones without an answer on your site are your AEO backlog, and probably your SEO backlog too (doing AEO without a solid SEO foundation is like dressing up the shop window of a store with no door).