The short answer
AEO stands for answer engine optimization. Its useful idea is that being the answer is a different outcome from being a link, and it needs measuring separately. Its checklist is mostly SEO, and two of the most repeated items on it point at Google features that no longer exist.
Nothing below is an argument that AEO is fake. It is an argument that you should be able to check each item against a source, and that most of the checklist does not survive that.
Where the term came from
We could not find out, and we looked properly.
There is no paper, no standards document, and no dated first use that the pages claiming an origin actually link to. Search for the history of AEO and you get a stack of agency posts that each assert an origin story and cite another agency post. Wikipedia's article on generative engine optimization records the same situation from the other direction: it notes there was no consensus definition separating AEO, GEO, LLMO and AI SEO as of early 2026, and that publications use them interchangeably.
The plausible reading is that "answer engine" was in circulation as a description of featured snippets and voice assistants for years, and someone bolted "optimization" onto it. That is a guess and we are labelling it as one. Compare with GEO, which has an actual paper with six named authors and a public benchmark. One of these two acronyms can be checked.
What an answer engine is supposed to be
The working definition is any system that returns a composed answer instead of a list of links. That covers Google AI Overviews and AI Mode, ChatGPT search, Perplexity, and the older featured snippet and knowledge panel surfaces.
The definition is loose enough to be a problem. A featured snippet is an extracted verbatim passage from one page. An AI Overview is generated text synthesised across several sources. Advice tuned for the first does not automatically transfer to the second, and most AEO content treats them as one thing.
The checklist, item by item
| AEO claim | Verdict | Primary source |
|---|---|---|
| Put the answer in the first sentence | Unproven, and often mis-argued | No Google guidance prescribes it |
| Add FAQ schema | Dead | Rich result removed 7 May 2026 |
| Win the featured snippet | Real, and it is plain SEO | Google featured snippets docs |
| Optimise for voice | Nothing published to optimise against | No Google voice optimisation guidance |
| Publish llms.txt | Unread by any documented search product | The proposal itself |
Answer in the first sentence
This is good writing advice. It is not evidence-backed AI visibility advice, and the evidence usually cited for it does not say what people think.
The GEO paper found that fluency and readability edits raised its visibility metric. That metric, Position-Adjusted Word Count, weights a citation by where it lands inside the engine's generated response. It says nothing about where the answer sits on your page. People quote a result about position in the answer as if it were a result about position on the source. Those are different variables.
Write the answer early because readers scan. Do not claim a study supports it unless you have read the study.
FAQ schema
Google's FAQ rich result no longer appears in Search as of 7 May 2026. The narrowing started earlier: in August 2023 Google removed the HowTo rich result outright and restricted FAQ rich results to well-known, authoritative government and health websites.
On top of that, Google's AI features documentation says: "There's also no special schema.org structured data that you need to add." Two separate Google statements, both against the tactic. We covered what to do with the markup you already shipped in FAQ schema is deprecated, what now.
Featured snippets
These still exist and still send clicks. The AEO framing of them is misleading in one specific way. Google's featured snippets documentation answers the question of how to mark up a page as a featured snippet with two words: "You can't." Google's systems decide.
The only levers documented are nosnippet, which blocks all snippets, and max-snippet, which Google suggests experimenting with at low lengths while warning that it "doesn't guarantee that Google will stop showing featured snippets for your page". Both opt you out. There is no opt-in mechanism to sell you.
Voice
There is no Google documentation for voice search optimization to point at. Google's featured snippet documentation, the page voice tactics are usually pinned to, does not mention Assistant, spoken answers or voice anywhere. An entire consulting sub-genre grew on a surface with no published guidance at all.
If you are still being pitched voice optimization as part of an AEO package, ask which product it targets and what documentation describes it.
llms.txt
The proposal is a serious piece of work by Jeremy Howard, first published in September 2024 and revised in August 2026. Its own site notes that OpenAI, Anthropic and Google publish llms.txt files for their developer documentation.
Publishing is not consuming. No major AI search product has documented reading the file to decide what it retrieves or cites. Google states outright that you do not need to create new machine readable or AI text files. We went through the evidence in llms.txt, what it is and who reads it.
What AEO gets right
One thing, and it is worth the acronym existing.
Being named in an answer is an outcome your analytics cannot see. A user who reads "Ahrefs, Semrush and Moz all do X" and never clicks leaves no trace in your referral data, your rank tracker or your logs. Traditional SEO has no instrument for this because until recently there was nothing to instrument.
The Pew Research Center gave the gap a size. Across 68,879 Google searches by 900 tracked US adults in March 2025, 12,593 produced an AI summary. Users clicked a standard result on 8% of those pages against 15% without a summary, and clicked a link inside the summary itself on 1%. Small sample by web-scale standards, single country, seventeen months old. Still the clearest published measurement of the thing AEO is named after.
A three question test for any AEO recommendation
Run every item on a vendor's list through these before you fund it.
- Which surface does it target? Featured snippet, AI Overview, AI Mode, ChatGPT search and Perplexity have different retrieval paths and different documentation. An item that cannot name its surface is not a tactic.
- What is the primary source? The crawler operator's docs, Google's docs, the RFC, or the paper. Not a roundup post. If the trail ends at another vendor blog, treat the claim as unsupported.
- How would you know it worked? If the answer is "traffic", the recommendation is not about answer engines, because being named without a click produces no traffic. If the answer is "our score went up", ask what the score is made of.
Most of the AEO checklist fails at least one of these. The parts that pass tend to be ordinary technical SEO wearing a new name, which is fine, as long as you are not paying twice for it.
The statistics we chose not to repeat
Three numbers circulate as settled fact in AEO content:
- the share of searches now showing an AI Overview
- the share of AI answers that cite the top organic result
- the conversion rate of AI referral traffic against organic
We went looking for a primary source with a stated method behind each and found none. Every trail ended at a vendor tracker, usually with the keyword sample undisclosed.
Presence rates in particular swing wildly with the query mix. Pew's real-browsing figure is about 18%. Our own commercial buyer-intent questions came back at 100%. Both are true and neither is "the" rate. A number without its query set is not a number.
What to do instead
- Write down ten questions a buyer types before they choose your category, in their words.
- Check who Google's AI Overview names for each one, and whether that includes you. The method, and the cost, is in how to measure AI visibility.
- Fix retrieval before wording. If a crawler cannot render your primary content, no answer-first paragraph saves you. Start with make your site readable to AI agents.
- Check what your robots.txt is actually saying, because training and search are separate tokens. See Google-Extended vs Googlebot.
- Delete the deprecated items from your deck.
For the record: SEOBuilder is ours. A default scan now asks seven answer engines, not one: Google AI Overviews, Google AI Mode, ChatGPT, Perplexity, Gemini, Copilot and Claude. We ran it on our own domain across nine buyer questions and were named in none of the 62 answers, which is written up on our answer engine optimization page along with the method.
If you want to poke at the underlying checks first, several are free and need no account in our free SEO tools roundup, including the answer readiness checker.