SEO wins a position in a list of links someone clicks. AEO — answer engine optimization — wins a citation inside an answer the machine writes. The mechanisms differ: ranking versus retrieval. The work mostly doesn’t. Google states there are no additional requirements or special optimizations needed to appear in its AI features.
Two things are true at once, and the AEO pitch usually tells you only one of them. The way answers get assembled is genuinely different from the way a results page gets ordered. The inputs — crawlable pages, clear entities, real expertise — are the same inputs that worked in 2019.
Ranking and retrieval are different jobs
Classic search ranks. It takes one query, scores documents against it, and hands back an ordered list. The unit of competition is the page and the prize is a click.
An answer engine retrieves. It gathers candidate material from an index, pulls the parts that look like they answer the question, and writes something new out of them. The unit of competition is a chunk of text and the prize is being one of the sources the model leans on, named when it does.
That distinction explains why a page can sit at position three and never get quoted while a page at position eleven gets cited constantly. Ranking well makes you a candidate. It does not make you quotable.
It also explains the reverse, which nobody selling AEO mentions: retrieval usually starts from a ranked index. Google is explicit that its generative AI features are “rooted in our core Search ranking and quality systems.” OpenAI says ChatGPT search works with third-party search providers, naming Microsoft among them. You are not escaping ranking. You are being ranked, then read.
Query fan-out: one question becomes many
Google documents that AI Overviews and AI Mode use a query fan-out technique that issues multiple related searches across subtopics and data sources while the answer is being generated. OpenAI describes the same shape in different words: ChatGPT search “typically rewrites your query into one or more targeted queries” before sending them out.
So a buyer types one sentence, and the system runs a spread of narrower searches derived from it — rephrasings, sub-questions, implied comparisons, the local variant — then assembles one answer from what came back across all of them.
The consequence: you compete not for a head term but for the sub-questions it decomposes into. A page that answers “what does it cost,” “how long does it take,” and “what goes wrong” has three chances to be pulled in. A page that only asserts you are the best choice has none.
Google names the failure mode directly: do not spin up separate content for every possible variation of how people might search. That is scaled content abuse, a spam policy violation. Cover the sub-questions inside pages that deserve to exist, not across forty pages that don’t.
The passage does the work, not the page
Google says its systems can “understand the nuance of multiple topics on a page and show the relevant piece to users.” That is the documented version. The practitioner version — that content is chunked and selected passage by passage — is inference. It matches observed behavior, but Google has published no specification, and anyone who tells you they know the chunk size is guessing.
What follows is safe enough to act on. Answers get lifted from specific paragraphs, and a paragraph that only makes sense after reading the three above it does not survive being taken out of the page.
Shapes that survive extraction
- A direct answer in the first two sentences under a heading. Heading states the question, paragraph answers it, then the nuance. Not the other way around.
- Self-contained paragraphs. Name the subject instead of relying on “it” and “this.” Pronouns break when the paragraph travels alone.
- Definitions, comparisons, ordered steps, and short lists. These are already shaped like the thing the model is trying to write.
- Concrete specifics. Numbers, ranges, conditions, named tradeoffs. Generic reassurance is not quotable because it is identical to everyone else’s.
None of this is new advice. It is how you write a page that earns a featured snippet.
What carries over unchanged
Crawlability, with more bots to think about
Nothing gets cited that cannot be fetched, and the crawlers are now separate and separately controllable. OpenAI documents OAI-SearchBot for surfacing sites in ChatGPT search, GPTBot for model training, and ChatGPT-User for user-initiated fetches. Block the first and you are out of ChatGPT search answers regardless of the others. Perplexity documents PerplexityBot for search indexing and Perplexity-User for on-demand visits.
Snippet controls matter too. Google’s AI features require a page to be indexed and eligible to be shown with a snippet, and they honor nosnippet, data-nosnippet and max-snippet. A restrictive max-snippet set years ago is now also an opt-out from being quoted.
Entity clarity
The machine has to be sure which organization it is naming before it names one. Consistent business name, address and phone across the site and your profiles; an about page that says who runs the company; author names attached to content. Perplexity publishes source labels and says it evaluates sites on whether they correct mistakes, identify authors, and separate news from advertising. That is an editorial standards checklist, not an algorithm hack.
Actual expertise
Google’s framing: a first-hand review offers a unique perspective from personal experience, while a summary of existing content restates what is already available. Restated content is the least useful thing a retrieval system can find, because it already has ten copies. Original numbers, a process you actually run, a position you will defend — those are the parts that get quoted.
Where AEO is repackaged SEO
Google’s own documentation says there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary, that you don’t need new machine-readable files or AI-specific markup, and that no special schema.org structured data is required. Structured data still earns rich results — as it did before, for the same reasons.
The llms.txt file is the clearest example. Ahrefs checked 137,210 domains and found 97% of the llms.txt files published received zero traffic in May 2026. AI retrieval bots accounted for about 1% of the requests that did arrive. It costs nothing to publish and it is not a strategy.
An AEO retainer consisting of technical cleanup, entity consistency, content restructuring and digital PR is an SEO retainer. That work matters. Paying for it twice under two names does not.
How to audit one page for both
- Confirm it can be fetched and quoted. Indexed, no noindex, no nosnippet, no restrictive max-snippet, AI search crawlers not blocked in robots.txt.
- Read the first 60 words under the title. If they don’t answer the question the page is about, rewrite them so they do.
- List the sub-questions. Cost, timeline, requirements, alternatives, failure modes — the eight to twelve things someone asking this also wants to know.
- Map them to headings, each with a direct answer underneath.
- Cut each paragraph loose. Read it in isolation. If it needs the paragraph above to make sense, name the subject explicitly.
- Find the part only you could have written. If there isn’t one, the page is a summary of other pages and it will lose to them.
- Check entity signals. Named author, consistent business details, a real about page.
- Then look at links and rankings. Nothing gets retrieved from an index it isn’t in.
Steps one through six are a day of work per page and serve both surfaces. Step eight is the SEO program you were already running.
The honest caveat: you mostly can’t measure this
Google now reports generative AI performance in Search Console, but that report gives impressions grouped by page, country, device and date — and no query-level data. ChatGPT and Perplexity give publishers nothing comparable. Third-party AEO trackers work by asking the assistants questions and recording the answers — sampling, not measurement, against outputs that are not deterministic.
Anyone quoting you a citation-share percentage and a plan to raise it is selling more certainty than the data supports. Judge this the way you judge brand search: does the page answer the question better than the alternatives, and are branded queries and assisted conversions moving. Slower signals, and the ones that exist.