AEO vs GEO vs SEO: what's fact and what's marketing

From research, not from a marketing department. Generative engine optimisation (GEO) was named in a paper by Aggarwal and co-authors, posted on arXiv in November 2023². The paper framed it as a way for content creators to improve their visibility in answers from generative engines².

Since then the term has spread well beyond that paper and is used loosely. An academic paper naming an idea is not evidence that a commercial service built on the label works. Keep the two apart.

Google has said it plainly, and it is the best primary source available.

AI Overviews and AI Mode are Google's AI features in Search. Google's documentation on them says there are no additional requirements and no special markup to qualify³. Its separate guide on optimising for generative AI search goes further: from Google's point of view, this work is still SEO¹. Google also announced that guide on its Search Central blog⁴.

That is the statement to hold on to: SEO is the base. For Google's AI features, you are not asked to build a second discipline.

Two limits apply. This is Google's position on Google's features. Other AI assistants publish less, and we would not stretch Google's wording to cover them. And the guidance says what is required, not what is wise. Good SEO is still hard work.

Mostly a name, and a different way of looking at the same ground.

AEO usually stands for answer engine optimisation. It has no single agreed definition. Google's own guide mentions AEO and GEO only as terms people use for this work, and still files the work under SEO¹. A vendor's own comparison guide acknowledges that definitions are not standardised⁵. An agency has written a post on how confused the definitions are across agencies⁶. An SEO software vendor publishes its own definition of GEO⁷. Three sources, three framings, and every one of them has something to sell.

In our view, what AEO and GEO add is a change of target. You care less about a position in a list and more about whether your brand is named or cited in an answer. That shift is real and worth tracking. It is not a separate toolkit. For the practical side, see how to get cited in ChatGPT.

Google's optimisation guide says several popular tactics are not needed to appear in its generative AI features: new AI text files such as llms.txt, breaking content into tiny pieces for AI, and writing in a special way just for AI¹. That matters because they are easy to package and invoice.

Take llms.txt. It is a proposed file meant to point AI assistants to your content. Google says creating one is fine, but it will neither help nor harm your visibility in Google Search¹. Whether other assistants read it is a different question, and not one to answer from a sales deck.

A useful test: if a service depends on a tactic Google calls unnecessary, ask what evidence the vendor has for the platform it does apply to. If the answer is a screenshot, that is a story, not evidence.

Less than the dashboards suggest.

You can measure what your own systems record: search impressions and clicks, traffic, and conversions. Visibility in AI answers is harder. In our view, treat every third-party AI visibility figure as an estimate. Ask which prompts were run, when, and how often. A sample is not a count.

Where a platform keeps the data, you are in walled garden territory. You see what it chooses to show. For what you can get without paid tools, read the Search Console AI Overviews report.

And be careful with "AI visibility went up". It is a reading of visibility, not of revenue. Whether it brought incremental sales is a separate question.

• Which part of this is different from SEO, and why?
• What does Google's documentation say about the tactic you are proposing?
• How do you measure AI visibility, and what is your sample?
• What would you stop doing if the evidence did not support it?
• How do you link a visibility gain to sales, not just to mentions?
• Which of these deliverables would I need even if I never bought "GEO"?

For ZFD: In our view, AI visibility is SEO plus a new way of measuring, and a tactic Google calls unnecessary for its own features is hard to justify on an invoice.