Answer Engines Do Not Wait: Why Server Speed Decides Whether AI Quotes You

Answer Engine Optimisation is the newer cousin of SEO. Instead of ranking in a list of blue links, you are trying to be the source an AI assistant quotes when somebody asks it a question.

Almost all the advice about it is about words: structure your headings, answer questions directly, add an llms.txt file. All of that is true. All of it is also useless if the machine gives up before your page arrives.

How an answer engine actually reads you

Somebody asks a question. The assistant decides it needs current information, picks a handful of URLs, and fetches them — while a real person sits watching a cursor blink.

That is the whole budget. These fetchers are not patient the way Googlebot is patient. Googlebot can come back tomorrow; an answer engine has a few seconds and a waiting human. If your server is still thinking, it moves on to a source that replied. You are not penalised, exactly. You are simply not in the answer.

Two more things worth knowing about how they read:

• Most do not run JavaScript. Anything your page assembles in the browser — content loaded after the fact, text injected by a script — may as well not exist. What arrives in the raw HTML is what gets read.
• They usually take the first response. A site that is quick on the second visit because everything is warm in cache, but slow on the first, gets judged on the slow one.

The bottleneck is CPU, not bandwidth

When we measure a slow WordPress site, the delay is almost never the download. It is the gap before the download starts — the server working out what to send. That is Time to First Byte, and it is a CPU number, not a connection-speed number.

Every uncached WordPress request means PHP starting up, plugins loading, a theme initialising, and dozens of database queries running before a single character is sent. On a healthy setup that takes a moment. On a typical small business site it does not, and the usual reasons are boringly consistent:

• No page cache. The site rebuilds every page from scratch, for every visitor, forever. This is the single biggest one and the easiest to fix.
• Plugin sprawl. Thirty plugins means thirty sets of code running on pages that use none of them.
• Oversold shared hosting. Your CPU is not really yours. At the busiest hour you are queuing behind hundreds of other sites.
• Heavy page builders. Some assemble the entire layout in PHP on every request instead of saving the finished result.

Aim for: under 200 ms to first byte. Between 200 and 600 ms is survivable. Past a second, you are gambling with every AI fetch and every impatient human.

What actually fixes it

In rough order of how much they return for the effort:

1. Full page caching. Serve a finished HTML file instead of building one. This alone routinely turns a second and a half into under a tenth.
2. Cache at the edge. With a CDN holding that HTML, the request may never touch your server at all. Now geography stops mattering too.
3. Object caching. For anything that genuinely cannot be cached whole — a logged-in area, a live price — this keeps the database from repeating itself.
4. Remove what you are not using. Every plugin you delete is code that never has to run again.
5. Buy real CPU. Sometimes the honest answer is that the hosting costs about as much per month as a sandwich, and behaves like it.

Then, and only then, the words

Once your server answers quickly, the content advice starts paying for itself. Put one clear question in each heading and answer it in the first sentence underneath. Write the specifics — prices, timescales, areas covered — as plain text rather than hiding them in an image or a PDF. Keep an llms.txt so the crawlers can find your important pages without guessing.

That work is worth doing. It is just second in the queue, and most guides have it first.

The uncomfortable summary

AEO is sold as a content problem. For most small business websites it is an infrastructure problem wearing a content problem’s clothes.

You can write the clearest, most quotable page on the internet about exactly what somebody just asked. If your server takes a second and a half to admit the page exists, the assistant will quote a competitor who answered faster and said it worse.