AEO vs SEO: What Changes When AI Is the Search Engine

Everyone has an opinion about AEO. Marketing blogs are calling it "the new SEO." Agencies are spinning up AEO services. HubSpot built a free grader for it. G2 already has a category page.

Almost none of them have built an answer engine.

We have. Dewey is an AI answer engine - we build the systems that take a user's question, search a body of expert content, and generate a cited, accurate answer. We've deployed these for health publishers, parenting experts, investigative newsrooms, and foundations. We see, every day, how retrieval-augmented generation systems select which content to cite, which to skip, and why.

That doesn't make us right about everything. AEO is new enough that anyone claiming certainty is selling something. But it gives us a perspective that's missing from the conversation: not how to optimize for an answer engine, but what answer engines actually look for.

Here's what we know, what we don't, and where the honest advice begins.

AEO - Answer Engine Optimization - is the practice of making your content the source that AI systems pull from when generating answers. Whether the AI is Google's AI Overview, ChatGPT, Perplexity, or Claude, the question is the same: when someone asks about your area of expertise, does the AI use your content to answer?

AEO is not replacing SEO. It's the layer on top of it.

SEO determines whether your content gets crawled, indexed, and ranked. AEO determines whether AI systems cite it, synthesize it, and serve it as an answer. You need the first to get the second. But ranking on page one no longer guarantees you'll be the source an AI Overview pulls from - and not ranking on page one no longer means you won't be.

(You may have also seen GEO - Generative Engine Optimization - and LLMO. They describe roughly the same thing. AEO maps most naturally onto existing mental models, so that's the term we'll use.)

Since we build answer engines, we can describe the architecture - not guess at it.

1. The question becomes a meaning, not a keyword.

When a user types "when should I start sleep training?" the system doesn't look for pages containing those exact words. It converts the question into a vector embedding - a mathematical representation of its meaning - and searches for semantically similar content. A page titled "Infant Sleep Methods: What the Research Says" might be a strong match even though it doesn't contain the phrase "sleep training."

At the retrieval stage, the AI matches meaning, not words.

2. Complete answers beat comprehensive pages.

The strongest signal for citation is semantic completeness: does the page contain a full, self-contained answer? Not a teaser. Not a roundup of links. An actual answer. A 100-page CXL study on AI Overview citation sources confirms what we see in our own systems: completeness is the strongest predictor of citation. Thin, vague, or fragmented content gets passed over - regardless of source authority.

3. Position matters less than you'd expect.

Ahrefs studied 863K keywords and 4M URLs and found only 38% of AI Overview citations come from pages ranking in the top 10 - down from 76% just seven months earlier. Nearly a third come from pages that don't rank in the top 100 at all. The AI makes its own editorial judgment about which content best answers the question. A thorough page well outside the top results can beat a thin page at #1. For smaller publishers with deep expertise but limited domain authority, this matters.

4. Information gain is the new differentiator.

Google holds a patent called "Contextual Estimation of Link Information Gain" (filed 2018, granted 2022) that scores how much new information a document adds beyond what's already available for a query. Content that rephrases what fifty other pages say scores low. Content with original data, first-hand expertise, or a unique synthesis scores high.

If you're a medical publisher with a specific clinical perspective on childhood vaccinations, your content structurally outscores a generic health blog summarizing CDC guidelines. Your take is uniquely yours. The generic blog is repeating what everyone else has said. Ahrefs also found that brand web mentions correlate 3x more strongly with AI Overview visibility than backlinks - a signal that the AI is looking for real-world authority, not just link graph position.

5. AEO rewards recency.

AI systems have two kinds of knowledge: parametric (baked into model weights during training, updated on cycles of months or years) and retrieval (looked up in real-time). AEO operates on retrieval knowledge. An article published today can appear in a Google AI Overview this week - and a definitive guide from 2023 may lose to a comprehensive piece from last month because the retrieval system prioritizes recent evidence.

6. "I don't know" gaps are search opportunities.

Google doesn't show an AI Overview for every query. For approximately 40-50% of searches, there's no AI-generated answer at all - a signal that the system couldn't generate a confident response from available sources. If your audience asks questions Google can't answer yet but you have a definitive response for, that's territory you can own.

The AEO conversation tends to focus on what's new. Worth naming what isn't.

Technical SEO is still the foundation. Crawlability, site speed, mobile responsiveness, canonical URLs, clean robots.txt - all still matter. These are prerequisites. If AI crawlers can't access your content, nothing else in this article applies. Check your robots.txt for ClaudeBot, GPTBot, and GoogleOther specifically; many publishers block them without realizing it.

E-E-A-T matters more in an AEO world, not less. 96% of AI Overview citations come from sources with verified Experience, Expertise, Authoritativeness, and Trustworthiness signals. Muck Rack analyzed over one million AI citations and found 82% came from earned media sources - bylined articles, expert commentary, credible press coverage. The AI makes editorial decisions about who to trust. Credentials, author bios, consistent publication history, and demonstrated domain expertise all carry more weight when an AI is deciding whether to stake its answer on your content.

Good content still wins. "Good" has been redefined - it now means complete, accurate, uniquely valuable, and structured for semantic understanding - but the underlying logic of SEO and AEO is the same: build content that earns trust, and the systems built to serve users will find it.

This is the part most AEO guides skip.

Nobody has a proven playbook. The field is months old. The data - AIO citation rates, ranking factor correlations, information gain scoring - comes from early studies with small sample sizes. Anyone selling a definitive AEO strategy is extrapolating from incomplete data. Including us.

AIO behavior is volatile. Google expanded AI Overviews from roughly 15% of queries to nearly 50% in the past year. The rules change quarter to quarter. Optimizing for today's AIO behavior is worthwhile, but building your entire content strategy around it is risky.

The publisher relationship is being rewritten. On March 18, 2026, Google announced that publishers will be able to opt out of AI Overviews. Opt-out implies a future opt-in with explicit terms - potentially including better data about when and how your content gets cited. Whatever AEO best practices look like today may look different in twelve months.

We don't know if AEO becomes its own discipline or gets absorbed back into SEO. The market may decide these are the same thing - that AEO is just what SEO becomes as search engines become answer engines. Or it may split into a distinct specialization with its own tools, agencies, and budgets. Both are plausible. The answer probably depends on how different the optimization work actually turns out to be in practice, and we're too early to know.

Measurement is rough. There's no equivalent of Google Search Console for AEO. You can't reliably track which pages are cited, across which AI systems, for which queries. A handful of startups - Graphite, aeochecker.ai, SE Ranking - are building citation trackers, but you can't optimize what you can't measure, and right now, measurement is rough.

Organized by confidence level, not priority - because honesty about uncertainty is more useful than false precision.

AI Overviews now appear in approximately 48-60% of all US searches. For informational queries - the kind your content probably targets - the rate is closer to 99%. When users get their answer in the AIO, many never click through. Dataslayer found organic CTR drops 61% for queries with AI Overviews. Being cited still drives traffic and brand awareness, but the value equation is shifting. The question isn't just "did my content get cited?" It's "did I keep the relationship with my audience, or did I give it to Google?"

That question points to something the AEO conversation rarely asks directly.

Everyone is trying to solve the same problem: how do I get Google's AI - or ChatGPT's, or Perplexity's - to cite my content? Fair question. Worth pursuing. But it accepts a premise worth examining: that the right response to AI search is to optimize for someone else's system.

Consider the alternative. Instead of making your content the source for someone else's answer engine, you make it the source for your own. An AI answer engine on your site, grounded exclusively in your content, delivering cited answers in your voice, under your brand. Your audience gets the same experience they'd get from an AI Overview - natural-language questions, direct answers with sources - but they get it from you. The relationship stays on your site. The trust belongs to your brand, not Google's.

This isn't either/or. The same content that makes Google's AI cite you makes your own answer engine accurate. But it's worth asking which relationship you're optimizing for - and what kind of hospitality you want to offer your audience.

AEO is real and growing. Search volume for "answer engine optimization" is rising sharply quarter over quarter, and enterprises are building it into their marketing budgets.

The durable strategy hasn't changed: create content so good and so uniquely yours that AI systems can't ignore it. Content grounded in real expertise. Content that says something no one else is saying. Content that answers the question completely, accurately, and with citations.

The search engine asked "what pages match this query?" The answer engine asks "what content can I trust?" If your values are embedded in your content, and your content is built on genuine expertise, you'll be cited - regardless of what we end up calling the optimization.

Dewey is an AI answer engine built for content businesses. We turn expert archives into interactive, cited, brand-safe experiences. Book a consultation to see how it works with your content.

FAQ

What is AEO?
AEO stands for Answer Engine Optimization - the practice of making your content the source AI systems cite when generating answers. It applies to Google AI Overviews, ChatGPT, Perplexity, Claude, and other AI-powered search experiences. AEO builds on traditional SEO rather than replacing it: SEO gets your content indexed, AEO gets it cited.

Is AEO replacing SEO?
No. AEO is a superset of SEO, not a replacement. You still need technical SEO fundamentals - crawlability, site speed, structured data, indexation. AEO adds a layer on top: making your content semantically complete, uniquely valuable, and structured so AI systems can retrieve and cite it. Think of SEO as getting into the library. AEO is being the book the librarian recommends.

What is parametric knowledge in AI?
Parametric knowledge is information an AI model "remembers" from its training data - stored in the neural network's weights. It's distinct from retrieval knowledge, which is information the AI looks up in real-time when answering a question. SEO influences parametric knowledge over long timescales (months to years, as models retrain). AEO influences retrieval knowledge immediately - an article published today can be cited in an AI Overview this week.