Answer Engine Optimization: What Google Has Actually Said, and What Is Left to Do

Answer engine optimization (AEO) is the work of making a page something AI answer engines (ChatGPT, Perplexity, Claude, Gemini, Copilot, Google AI Overviews and Google AI Mode) can find, read and cite inside the answers they write. For Google's AI features, Google's own documentation says there are no extra requirements, no special files and no special schema. The page has to be indexed, eligible for a snippet, open to crawlers and carry its important content as text, and the rest is being a source worth quoting. The stakes are real: Content Science Review reports that a study by Pew Research Center "found participants using Google's AI overviews were much less likely to click on links for the websites referenced." So the answer is often the only part of your page a reader ever sees.

This reference covers:

• what AEO means, and how it relates to SEO and GEO
• the answer engines, and where each one gets its material
• what Google has said is not required, item by item
• the four things that are required, each described the same way
• what outside research adds, and how far it can be trusted
• what no site owner controls
• one product page taken through every requirement
• how AEO gets measured, and short answers to the questions people ask most

What does answer engine optimization mean?

An answer engine takes a question and returns a written answer, not a list of ten links. The answer may cite sources, link to them, or name a product without linking at all. AEO is the name for the practices that make a page more likely to be one of those sources.

The term sits next to two others. They overlap heavily, and the differences below are about what each one is trying to earn, not about separate techniques.

GEO started as an academic term, from a paper of that name presented at KDD 2024. In practice, people use AEO and GEO for the same thing. The fuller comparison of AEO and SEO, including the differences people invented, belongs in its own piece. For this reference, one point matters: SEO earns a ranked link, while AEO earns a place inside the answer. Getting there takes passages that can be lifted out cleanly, structure a machine can read, and content that crawlers can actually reach.

The answer engines, and where each gets its material

Every answer engine that cites the live web needs a copy of your page first. How it gets that copy decides what it can see. The table uses the same fields for each engine and leaves out anything not publicly established.

The rendering column is where most product sites run into trouble. Vercel's published analysis of AI crawler traffic on its network found that the major AI crawlers from OpenAI, Anthropic and Perplexity fetch JavaScript files but do not execute them. Google's Gemini is the measured exception, because it rides on Googlebot. That is one vendor's network over a bounded window, and crawler behaviour can change without notice. Still, it is a direct observation, not a correlation, and it lines up with what Google asks for in its own guidance (covered below). The consequences for a client-rendered product site get their own article in this series.

What Google has said is not required

Most of what circulates as AEO advice is some version of "add this file," "add this markup" or "write this way for the robots." Google's documentation addresses each of these directly. Its page AI features and your website on Google Search Central states:

"There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."

The same page says you don't need "to create new machine readable files, AI text files, or markup to appear in these features," and that "there's also no special schema.org structured data that you need to add."

Google's newer Guide to optimizing for generative AI features on Google Search goes further. It says there is "no requirement to break your content into tiny pieces for AI to better understand it," that "you don't need to write in a specific way just for generative AI search," and that "the best practices for SEO continue to be relevant because our generative AI features on Google Search are rooted in our core Search ranking and quality systems."

Here are the common recommendations, set against that guidance:

Two things this table does not say. First, it covers Google's features only. Other answer engines haven't published equivalent statements, which is not the same as saying they need these things. Second, structured data still has uses. Article, Organization, Product and Software app are in Google's current rich results gallery, and other parsers read them too. The point is narrower: none of it is a condition for appearing in Google's AI answers. The status of llms.txt specifically, and what each engine has said about it, is covered in a separate piece.

What is required: the four prerequisites

What Google does list is short, and none of it is new. Each prerequisite below uses the same fields: what it is, how it fails, how it is checked, and what it looks like on a real page.

1. Indexed and eligible for a snippet

What it is. Google's AI features documentation states that a page "must be indexed and eligible to be shown in Google Search with a snippet." A page outside the index, or one that has opted out of snippets, can't be a source for AI Overviews or AI Mode.

How it fails.

• A noindex robots meta tag or X-Robots-Tag header, often left over from a staging deploy
• A nosnippet directive, a max-snippet value that is too restrictive, or data-nosnippet wrapped around the content that matters
• A canonical tag pointing to a different URL, so the page you care about is never the one indexed
• A non-200 status code on the canonical URL

How it is checked. The URL inspection tool in Google Search Console reports index status and the canonical Google chose. The page source shows robots meta tags, and the response headers show X-Robots-Tag.

Example. A pricing page carries <meta name="robots" content="noindex"> from a template the landing page was cloned from. It ranks for nothing and can't be cited in AI Overviews, however good the copy is.

2. Crawling allowed

What it is. Google's guidance lists "ensuring that crawling is allowed in robots.txt" among the technical basics. For answer engines outside Google, the same logic applies to their own user agents. A crawler that is blocked can't fetch the page.

How it fails.

• A blanket Disallow: / in robots.txt, left from pre-launch
• Rules that block AI crawlers by user agent, added deliberately or inherited from a template or plugin
• A firewall, bot-protection layer or CDN rule that challenges or blocks crawlers before robots.txt is ever read
• Pages reachable only behind a login, or only through on-site search

How it is checked. Read robots.txt directly at the root of the domain and compare each rule against the user agents you care about: Googlebot, and the published crawler names of the other engines. Check server or CDN logs for crawler requests that got a challenge page or an error.

Example. A founder adds a bot-protection setting during a spam wave. Browsers pass the challenge, but crawlers get an interstitial page instead of the product page. robots.txt allows everything, so the problem stays invisible until someone looks at what a crawler actually receives.

3. Important content available as text in the served HTML

What it is. Google asks site owners to make sure "important content is available in textual form." For crawlers that don't execute JavaScript, the practical meaning is stricter: the content has to be in the HTML the server sends, not assembled afterwards in the browser.

How it fails.

• A single-page app that sends an empty <div id="root"> and builds the page with JavaScript
• Key claims (what the product does, who it is for, what it costs) shown only as text inside images, or inside video
• Pricing loaded from an API after the page renders
• Content behind tabs or accordions that only exists in the DOM after a click

How it is checked. Fetch the page without a browser (view the raw source, or request it with a command-line HTTP client) and search the response for the sentences that matter. If they aren't in the response, a non-rendering crawler doesn't see them.

Example. A developer tool's landing page is a client-rendered React app. In a browser it shows a clear headline, three feature blocks and a pricing table. The raw HTML holds a title tag, a script bundle and an empty root element. Googlebot renders it and sees everything. A crawler that doesn't render sees a page with no content.

4. Content worth quoting

What it is. The one prerequisite that is an editorial decision rather than a technical one. Google's guidance asks for "unique, compelling, and useful" content with a unique point of view, and explicitly warns against commodity content that "could easily be produced by a generative AI model." An answer engine chooses between many pages that could answer the same question. A page that only restates what every other page says gives it no reason to prefer that page.

How it fails.

• Pages that describe a category ("the best tools for X") without saying anything only this site knows
• Claims with no source, numbers with no origin, and quotes with no speaker
• Answers buried under introductions, so no single passage states the answer on its own
• Product pages that never say plainly what the product does, for whom, and at what price

How it is checked. Read the page and pick out the one or two passages an engine could quote as a complete answer. If none stand on their own, or if they could appear on any competitor's site unchanged, the page is not yet worth quoting.

Example. A habit-tracking app's homepage says it "helps you build better habits." The same page, rewritten, says in one paragraph who it is for, what it does that a notes app doesn't, which platforms it runs on, and what the paid plan costs. The second version is a passage an engine can lift and cite. The first could belong to any habit app.

What the outside research adds

Beyond Google's guidance, a small body of research exists. Each item is labelled below with the kind of claim it can support.

The GEO paper deserves its scope limits spelled out, because it is the most quoted study in this field and the most often stretched. It was run on GEO-bench, a benchmark the authors built with their own generative-engine setup. By the authors' own finding, the effects varied across domains. It is good evidence that citing sources, quoting real people and including real numbers helps a page stand out in generated answers. It is not evidence of a guaranteed effect in any commercial engine. And it doesn't license inventing statistics to fit the pattern: a fabricated figure fails the trust test that decides whether a page gets used at all.

Note also where these findings agree. Google's quality guidance asks for unique, sourced, useful pages. The GEO experiment found that sourced, specific pages did better on its benchmark. It is the same work, and that agreement is the strongest argument for it.

What no site owner controls

An honest reference to AEO has to say where influence stops.

• Which source gets picked. An engine chooses among eligible pages using systems it doesn't publish. Meeting every prerequisite makes a page eligible. It doesn't make it chosen.
• Whether a mention links. An engine may name a product without linking to it, or summarise a page without naming it.
• Consistency between runs. The same question asked twice can produce different answers and different sources. One screenshot of an answer is an anecdote, not a measurement.
• Third-party coverage. The observation that answer engines lean heavily on platforms such as YouTube and Reddit means some of what gets said about a product is said on pages its maker doesn't own.
• Rule changes. Crawler behaviour, citation formats and Google's guidance all change. The prerequisites above have stayed stable because they are the ordinary conditions of being on the web, but the details beneath them move.

This is why no reputable approach to AEO promises a placement. A site can remove every reason not to be cited. It can't force a citation.

One page through every prerequisite

The same product page is followed through all four prerequisites, so the pieces connect.

The page. A solo founder ships a web app for freelance designers to send invoices. The marketing site is the app's own front end: a client-rendered single-page app with a homepage, a pricing page and a short docs section. Nobody finds it through search, and when the founder asks ChatGPT and Perplexity for invoicing tools for freelance designers, it never comes up.

Prerequisite 1, indexed and snippet-eligible. The homepage is indexed. The pricing page isn't: it was cloned from a staging template that still carries noindex. For Google's AI features, the page that answers "how much does it cost" isn't a candidate at all. Removing the tag makes it eligible again. It doesn't make it cited.

Prerequisite 2, crawling allowed. robots.txt allows all user agents. The hosting provider's bot-protection mode, turned on during a sign-up spam wave, shows a challenge page to unrecognised clients. Browsers pass it, and some crawlers get the challenge instead of the page. From the outside, the site looks open.

Prerequisite 3, content in the served HTML. A raw fetch of the homepage returns a title, a meta description, a script bundle and an empty root element. The headline, the feature list and the pricing table exist only after JavaScript runs. Google's AI Overviews and AI Mode can use the rendered page, because Googlebot renders. ChatGPT, Claude and Perplexity, going by Vercel's observations of their crawlers, get the empty shell. The fix is server rendering or static generation for the marketing routes, so the same sentences appear in the raw response.

Prerequisite 4, worth quoting. With the content now in the HTML, the homepage reads "Invoicing, reimagined." Nothing on it says who the product is for, what it does that a spreadsheet template doesn't, or what it costs. A rewrite puts one self-contained paragraph near the top: invoicing for freelance designers, what it handles (for example, recurring retainers and client approval of line items), what platforms it runs on, the free tier's limit and the paid plan's price as the pricing page states them. The docs gain a page answering a question designers actually ask, written from the founder's own experience of building the feature.

What changed, and what didn't. After all four, the site is indexable, reachable, readable without JavaScript, and says something specific enough to quote. None of this guarantees that ChatGPT names it next week. What it removes is every technical reason for the product not to be named, which was the part in the founder's control.

How AEO is measured

There is no official AEO metric from any engine. The measurement that holds up is simple to describe and tedious to do: choose a fixed set of questions a buyer would really ask, put them to each answer engine, record whether and how the product appears, and repeat on a schedule. The share of answers that mention the product, across the whole set, is more meaningful than any single answer, because single answers vary between runs.

Alongside that sits a second layer: checking the prerequisites themselves, because a regression there (a redesign that drops server rendering, a deploy that reintroduces noindex) quietly undoes the rest. Sites drift. A new page ships without the tags its siblings have, or a firewall rule changes. A check that passed once says nothing about next month.

The prerequisites can be checked by hand, using the methods listed under each one above. They can also be checked in one pass: LaunchScaler's free readiness scan reads a site the way search engines and answer engines read it and covers both SEO and AEO in the same run. Either way, the order matters. Prerequisites first, because share of the answer can't improve for a page an engine can't read.

Where to start

Take the one page that matters most for your product, usually the homepage or the pricing page, and fetch its raw HTML without a browser. Search the response for the sentence that says what the product does and who it is for. If that sentence isn't there, work on prerequisite 3 before anything else. If it is, check the page for noindex and nosnippet, read your robots.txt against the crawler names you care about, then reread the page as an engine choosing between it and every competitor. Ask whether any single passage on it deserves to be quoted.

Short answers

How do you do answer engine optimization?

Meet the four prerequisites in order: the page is indexed and eligible for a snippet, crawling is allowed in robots.txt and at the firewall, the important content is in the served HTML as text, and the page says something specific and sourced enough to be worth quoting. Google says its AI features need nothing beyond this. The remaining work is editorial, and it is the same work good SEO asks for.

What is AEO vs SEO?

SEO aims for a ranked link that a searcher clicks. AEO aims for a mention or citation inside a generated answer, where fewer readers click through. For Google, the underlying requirements are the same, because Google states its AI features are rooted in its core ranking and quality systems. The practical difference shows up with non-Google crawlers, which don't execute JavaScript, and in how success is measured.

What's the best answer engine optimization tool?

No single tool settles this, and none can promise a citation. It helps to split the job in two. One kind of tool checks the prerequisites: index status, snippet directives, crawler access and whether content exists in the raw HTML. Another kind measures share of the answer by putting a fixed set of questions to several engines over time. Judge a tool by whether it shows the evidence behind each verdict and whether its questions resemble what your buyers actually ask.

Is AEO real?

The goal is real: answer engines cite sources, and being one matters more when readers click less. As a separate technical discipline with its own files, markup and writing rules, it is mostly not real for Google, whose documentation says no special optimizations, AI text files or schema are needed. What remains is ordinary web hygiene, server-rendered content for crawlers that don't run JavaScript, and pages worth quoting.