How do you tell whether an AEO agency has real SEO foundations?

Why the SEO layer is the tell

A language model reaches a page the same way any crawler does. If the page renders client-side and returns an empty document to a non-JavaScript agent, no amount of prompt work will get it cited. If the entity behind the brand is ambiguous in the knowledge graph, the model has nothing stable to attach a claim to. These are old problems wearing new clothes.

This matters commercially because the two failure modes cost very different amounts to fix. A rendering problem is an engineering ticket. A positioning problem is a quarter of content work. A vendor who cannot tell you which one you have is going to sell you the one they happen to do.

Questions that separate the two

• Show me a rendered-DOM diff from a past engagement. What the browser sees against what a plain fetch sees. If they have never produced one, they have never checked.
• Which AI crawlers reached which of my pages last month, and which did they skip? This is a server log question. The answer is in the logs or it is a guess.
• What is your position on attribution? Anyone claiming they can attribute revenue to a model citation is either measuring something narrower than they are describing or has not thought about it hard enough.
• What would make you tell me not to hire you? A practice with a defined method has an answer. A sales process does not.
• Do you resell any tooling, and do you take vendor commissions? Not disqualifying, but it changes how you read every recommendation that follows.

Answers that should end the conversation

• A guaranteed number of citations, or a guaranteed position in AI Overviews. Neither is in anyone's control.
• Live answer-engine testing presented as a measurement baseline without any discussion of run-to-run variance. Ask them to run the same prompt set twice on different days and show you both.
• A recommendation to add llms.txt with no discussion of how it stays current. A static file that never changes is a weak signal.
• Schema proposed for rich snippets. That was the 2019 argument. The current value is entity resolution.

What a defensible engagement looks like

Diagnosis before prescription, in layer order: technical, content, links, then the AEO layer built on what the three below it turned up. Every finding carries the evidence that produced it, so you can hand the report to your own engineers and they can verify it without trusting the author.

If you are mid-selection and want an independent read on a proposal you have already received, that is a service offered here, including when the recommendation is to hire the other vendor.