Answer engine optimization is the practice of writing and structuring a page so that search engines and AI assistants use it as the answer, rather than listing it as one of ten options. That is the definition everybody agrees on. The disagreement starts one sentence later, over what you have to do about it, and the cheapest way to settle that is to read what Google publishes rather than what a tool vendor emails you.
Google’s documentation on AI features says this, in its own words: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” And more bluntly: “You don’t need to create new machine readable files, AI text files, or markup to appear in these features. There’s also no special schema.org structured data that you need to add.”
Eligibility is the ordinary bar. A page has to be indexed and eligible to appear in Search with a snippet. That is it. If a vendor’s pitch requires you to believe there is a secret AEO markup standard, the people who build the feature disagree with them in writing.
So what is actually left to do? Quite a lot, but none of it is new technology.
What changes is where the answer sits on the page
An answer engine has to lift a self-contained statement out of your page and stand behind it. If your answer to “how long does a site migration take” arrives in paragraph nine, after a preamble about how important migrations are, there is nothing clean to lift.
This is the whole craft, and it is older than AI. Featured snippets rewarded the same behaviour for a decade.
Take a page targeting “how much does a technical SEO audit cost”. The version that loses:
Pricing for technical audits varies considerably depending on a number of factors. Every site is different, and agencies take different approaches to scoping work, which means there is no single answer to this question.
Four sentences in and the reader still has no number. The version that gets quoted:
A technical SEO audit costs between $1,500 and $8,000 for most sites. Agencies price by crawl size and template count, so a 500-page brochure site sits at the bottom of that range and a 200,000-URL ecommerce catalogue sits above it. Hourly rates run $100 to $200.
Same information. The second one can be quoted in two sentences without the engine inventing anything, it names a figure, and it survives being pulled out of context. Write that paragraph, then put the caveats underneath it where they belong.
The rest of the content work is unglamorous. One question per heading, phrased the way a person would ask it, which is what header tags have always been for. Tables for anything comparative, because tabular data gets lifted intact. Dates and figures stated in the text and not only in an image. Definitions that stand alone, so a model quoting one sentence of yours does not misrepresent you.
Schema markup still earns its place, just not for the reason it gets sold. Google says plainly that no special structured data is required for AI features. What markup does is remove ambiguity about what your entities are, which matters for the knowledge-graph layer underneath all of this, and it drives rich results in ordinary search. Mark up what your page genuinely is — Product, Organization, Article, LocalBusiness — and stop there. Our schema markup guide covers the types that still do something.
The uncomfortable number
Here is the part most AEO articles leave out. Pew Research tracked the browsing of 900 US adults across 68,879 Google searches in March 2025, 12,593 of which returned an AI summary. People who saw an AI summary clicked a traditional result on 8% of visits. People who did not clicked on 15%, nearly twice as often. And clicks on the sources cited inside the summary happened on 1% of visits.
One per cent.
Ahrefs, looking at 300,000 keywords, found that the presence of an AI Overview correlated with a 34.5% lower clickthrough rate for the page ranking first.
Read those two findings together and the business case for AEO is not the one in the vendor decks. Being cited is a brand impression, not a traffic source. It behaves like a billboard: broad reach, trivial click rate, real effect on whether your name is familiar when somebody later searches for you by hand. Price it as awareness, measure it as awareness, and stop forecasting sessions from it. Anyone selling AEO on a traffic model is selling against the only public clickstream data we have.
That does not make it worthless. Appearing in the answer to “best white label SEO reporting tools” when a buyer asks an assistant is worth something. It is just worth something different from a number-one ranking, and the 1% figure is the honest ceiling on how much of it arrives as a visit.
AEO, GEO, AIO, LLMO
Four acronyms, roughly one job. AEO came out of the featured-snippet era and points at direct answers. GEO was coined for generative engines specifically. AIO and LLMO are later arrivals, mostly vendor branding. If anyone tells you these are separate disciplines with separate budgets, ask them what work item differs between them, and watch the answer collapse into “write clear answers and get cited by credible sources”. Where the two sets of work genuinely do diverge, and how to split a budget across them, is GEO vs SEO.
The one distinction worth holding on to is where the volume actually is. Ahrefs’ prompt index runs to 449 million modelled prompts, of which 315 million are Google AI Overviews. Gemini, ChatGPT, Perplexity and Copilot sit at roughly 30 million each, and Google’s AI Mode at 12.7 million. The AI answer surface is, overwhelmingly, still Google. Optimising for ChatGPT as though it were the main event gets the proportions backwards for most businesses.
How to tell whether it is working
Start by ruling out the thing everyone gets wrong. Open Search Console, segment by query, and compare impressions against clicks across the period you think you lost. Impressions holding while clicks fall is consistent with an answer absorbing the click. Both falling means you lost rankings, which is an ordinary problem with an ordinary diagnostic, and no amount of answer-shaping will fix it.
Then build a baseline you can repeat. Write 20 to 50 prompts your buyers would actually type, including the ugly ones with your competitor’s name in them. Run them against the engines that matter, record whether you are named, whether you are linked, and which domains got cited instead. Run the identical set monthly. The whole value is in the comparison, so the prompts must not change, which is the discipline most teams skip.
You can do that by hand in a spreadsheet for free, or pay for it. If you want the pricing reality on the tools that automate it, including which ones are already bundled into the SEO suite you probably pay for, we put every vendor’s numbers side by side in what the LLM SEO tools actually cost. The deeper measurement logic, and which parts of your existing SEO carry over, is in getting cited in AI answers.
Two things not to measure: a single spot check (“I asked ChatGPT and it mentioned us”) and anything a vendor calls a visibility score without publishing the formula. Model answers are non-deterministic, so one run is an anecdote, and a proprietary index with an unpublished method cannot be audited or carried to another tool.
What I would do first
If you have never done any of this, the ordering is boring and it matters. Fix whether you are crawlable and indexable, because the eligibility rule is still just Search eligibility. Then rewrite the opening 60 words of your 20 highest-intent pages so each one answers its own question in a liftable sentence. Then set the prompt baseline so you can tell, in three months, whether any of it moved.
What I would not do is buy a new content system, add markup Google has said it does not use, or let anyone charge you a retainer for AEO that looks suspiciously like the on-page work you were already paying for. If you want somebody to establish the baseline against a fixed prompt set first, that is our AI visibility audit, and the honest version of that engagement ends with us telling you whether you have a problem at all.
