AEO
Answer engine optimization, without the acronym
Three things about AEO are genuinely new. The rest is old advice with a new label, and one part of it is being oversold. Here is which is which.
The acronym is doing a lot of work in the marketing right now. Under it there are three real changes and a lot of old advice with a new label.
What actually changed
An answer engine does not hand you a list. It reads several sources and writes a new sentence. There is no position 1 to occupy. You are in the answer or you are not, and if you are, you may be cited or you may be absorbed without a link.
The crawlers are simpler than Googlebot. Google renders JavaScript. GPTBot, ClaudeBot, PerplexityBot and CCBot fetch the file and read it. A site assembled in the browser gives them very little to work with, and this is the single most common reason a business is absent from answers about its own category.
Your own site is not the only source. An assistant assembles from directories, review platforms, forum posts, press and whatever it absorbed in training. Some of those you control, some you can correct, and some you can only outweigh.
That is the honest list. Three things.
What did not change
Clear headings, a page that answers the question it is named after, structured data that matches what the page shows, a business listing that agrees with itself across the internet. All of that was already good practice and all of it still is.
If an AEO proposal is mostly this, it is not wrong. It is SEO, and it should be priced as SEO.
The part that is being oversold
Nobody has a clean, published before-and-after showing that a specific change caused a specific assistant to start citing them. Including us. The measurement problem is genuinely hard: the answers are not deterministic, they vary by phrasing, by user, by model version and by the day.
We can measure what a crawler receives. We can measure whether the structured data is valid and whether it matches the page. We can ask the assistants the questions your buyers ask and record what comes back, with dates, and watch it change. That is a log, not an attribution model.
Anybody selling you a guaranteed citation is selling something that cannot be guaranteed, because the answer is regenerated per query by a system nobody outside the lab controls.
Four things worth doing this month
Fetch your own homepage without a browser. curl -s https://yoursite.com/ | wc -c and then look at what came back. If your content is not in there, nothing else on this list matters yet.
Write one flat sentence per page that answers the question the page is named after. Put it near the top, in plain text, with no adjectives. That is the sentence a machine will quote. Resist the urge to make it sound like marketing.
Make your listings agree. Same name, same address, same phone, everywhere. A model reading three versions of your address has to pick one, and it may not pick the current one.
Add an llms.txt. It is a markdown index of your site at /llms.txt, per llmstxt.org. It is cheap, it takes an hour, and when we scanned 13 Connecticut agency sites on 2 September 2026, 1 of them had one.
Where to start if you only do one
The first one. Everything else is decoration on a page a machine cannot read.
Our own site sold AEO and an answer engine could not read it
Our homepage returned 10,866 bytes containing two links, and 22 of 24 pages were invisible to every AI crawler. What was wrong, and what fixed it.
Read it →
Start here
This is the kind of thing the free audit turns up.
Send us the account. In 24 hours you get the wasted spend, the tracking gaps and the recoverable upside, and the findings are yours either way.
- Audit back in 24 hours. Findings are yours either way
- Owner-led. You brief the engineer, not an account manager
- Engineered for revenue, not vanity metrics
- AI woven through, not bolted on