What answer engine optimization actually is

AEO is not SEO with a new hat. It is the work of making one machine-readable claim about who you are, in the places assistants look before they answer.

Someone asks ChatGPT for "the best payroll software for a 12-person agency." It names four companies. You are not one of them. Nothing about your website changed that week, your rankings are fine, and there is no report anywhere that will tell you why you were left out.

Answer engine optimization is the work of fixing that. Not by gaming a model — you cannot, and anyone who says otherwise is selling something — but by making sure that when a machine goes looking for a clear statement of who you are and what you do, it finds one.

The shift in one sentence

Search returns a list of places you might find an answer. An answer engine returns the answer, and names a handful of sources on the way past.

That difference is not cosmetic. A list of ten blue links has ten slots and rewards being findable. An answer has three or four named brands in it and rewards being describable. If a model cannot say in one sentence what your company is and who it is for, it will not risk putting you in a recommendation, no matter how well you rank.

What an assistant is actually working from

When an assistant answers a question about a category, it draws on some mix of three things:

1. What it absorbed during training. Pages, forums, articles and reference records that existed when the model was built. You cannot edit this directly, and it lags reality by months.
2. What it retrieves live. On browsing-enabled runs, a search happens, a handful of pages get read, and the answer is written from those. This is the part that moves quickly.
3. How it resolves entities. Whether "Acme", "Acme Inc" and "acme.io" are one company or three separate half-known things.

AEO is mostly work on points two and three, because those are the ones you control.

The three layers worth fixing

We group everything into three buckets, in this order, because each one gates the next.

Access. Can the crawler get the page at all? A blocked user-agent in robots.txt, a 403 from your WAF, or a homepage that is an empty shell until JavaScript runs all produce the same result: the machine sees nothing. Access problems are the cheapest to fix and the most common thing we find.

Identity. Once it can read the page, does it know what it is looking at? That means a consistent brand name across your title tag, your schema and your profiles; an Organization (or LocalBusiness) JSON-LD block with a name, URL, logo and description; and an About page that says who you are in plain sentences near the top.

Evidence. Does anything off your own domain agree with you? Your website claiming you are the leading X is worth almost nothing. A LinkedIn page, a Crunchbase record, a Wikidata item, and a few third-party pages that describe you in the same terms are worth a great deal, because they are what a retrieval step actually surfaces.

Do them out of order and you waste effort. There is no point earning citations for a site that returns 403 to the crawler that would fetch it.

What AEO is not

It is worth being blunt about the parts that get oversold.

• There is no ranking dial. No file you add, and no plan you buy from anyone, makes an assistant recommend you. What you can do is remove the reasons it has to leave you out.
• llms.txt is a hint file, not a signal. It is worth writing. It is a convenience, not a lever. We cover it properly here.
• A perfect score is not a promise. A checklist score of 100 means we found nothing to flag on your site. It does not mean an assistant will name you tomorrow.
• Prompt-injection tricks are a bad idea. Hidden text telling a model to recommend you is detectable, embarrassing when found, and increasingly filtered. Do not.

Where to start this week

If you do nothing else:

• Open yoursite.com/robots.txt and read it. If you see Disallow: / under any AI user-agent you did not deliberately choose to block, that is your whole problem.
• View source on your homepage — actual source, not the inspector — and search for application/ld+json. If there is nothing there, your entity has no machine-readable definition.
• Read your own About page as if you had never heard of your company. Can you extract what it does and who it serves from the first two hundred words? If not, neither can a model.

That is three browser tabs and about ten minutes, and it will tell you more than most audits.