GEO for WordPress: how AI search finds your site

This page is the overview. If you want to know whether your site already has a problem, the same ground is covered as eight specific mistakes. If you only want the llms.txt part, that is its own guide.

None of this needs a new platform or a subscription. All five parts can be done on a standard WordPress install, and every one of them can be checked afterwards — that last part matters more than it sounds, because most GEO advice cannot be verified at all.

What changes when the reader is a model

A search engine ranks pages and sends a person to one of them. A generative engine reads several sources and writes an answer, then decides whose sentence to use and whom to credit. So the question stops being "where do I rank" and becomes "is my sentence the one worth quoting, and can the machine tell it came from me".

That shifts weight onto three things a ranking report never showed you: whether your answer stands on its own, whether the facts on the page also exist in a form a machine can read, and whether the crawlers that cite sources are allowed in at all.

Part 1 — Answers that survive being quoted

Write the answer near the top, self-contained, in two or three sentences that still make sense with everything around them removed. No "as mentioned above", no answer that only works after four paragraphs of setup.

• One page, one question. A page that answers five things is quoted for none of them.
• Put the number, the price, the duration or the limit in the sentence, not in an image and not three clicks away.
• Headings that are questions get quoted more often than headings that are labels, because the model is matching a question.

Check: delete the heading and read the first two sentences alone. If they do not answer the title, nothing else on this page will save it.

Part 2 — Structured data that matches the page

Structured data is the machine-readable copy of what a person sees: the product and its price, the article and its date, the questions and their answers. When it is missing or incomplete, nothing looks wrong on screen — which is exactly why it goes unnoticed for months.

• Product, Article, FAQPage, Organization and BreadcrumbList cover most of what a business site needs. More types are not better.
• Every required field, present and true. A missing required property withholds the whole rich result, silently.
• The markup must agree with the page. A price in the markup that is not the price on screen is worse than no markup.

Check: run the page through a structured data test and read the warnings, not just the errors. We learned that one the hard way — our own product pages showed a picture but the markup carried no image property, and the rich result stayed switched off for three weeks while every page looked fine.

Part 3 — Say who you are, the same way, everywhere

A model answering "who makes X" is assembling an entity from whatever it can find. If your name, description, location and links are phrased differently on your site, on wordpress.org, on GitHub and on LinkedIn, each version lowers confidence in the others.

• One Organization definition on the site, generated once and reused, rather than written per page.
• The same short description everywhere you have a profile. Copy it, do not re-improvise it.
• Link out to the profiles you control, so the pieces can be connected to each other.

Check: search your own brand name and read the top five results as if you were a stranger. If they describe four different companies, that is the problem to fix before anything on this page.

Part 4 — llms.txt, an index and nothing more

llms.txt is a plain-text list at the root of your site pointing at the pages worth reading. It is cheap, it does no harm, and it is an index — it cannot make thin pages worth quoting. Generate it from real content rather than writing it by hand, and make sure only one thing on your site is generating it.

The mechanics — which plugins offer it, which one wins when two are switched on, and how to confirm what is actually being served — are in the llms.txt guide. Ours is at /llms.txt, and you are welcome to open it.

Part 5 — Decide which crawlers you let in

AI crawlers are not one group. Some fetch pages so they can cite and link you — OAI-SearchBot, PerplexityBot, Claude-SearchBot, Google-Extended. Others take the content and send nothing back: GPTBot, ClaudeBot, CCBot, Bytespider. Blocking everything with an AI-sounding name costs you the referrals and buys you nothing.

• Look at which crawlers actually visited before deciding anything. Most of this argument happens without that evidence.
• Know how strong each control is: turning a crawler away at the door works; a robots.txt rule is a request that well-behaved operators honour and anything else ignores; the noai header is weak but free.
• Never block Googlebot or Bingbot in the process, and leave the crawlers that build link previews alone.

How to check the whole thing

Give yourself an hour and run these five. Each one either passes or does not — none of them needs a tool you have to pay for.

• Open three of your pages and read only the first two sentences. Do they answer the title?
• Run the same three through a structured data test. Zero errors, and read the warnings.
• Search your brand name. Do the top results describe one company or several?
• Open /llms.txt on your own site. Does it exist, and do three random entries lead somewhere worth reading?
• Look at a month of crawler activity. Are the citing crawlers getting in, and is anything you meant to block still arriving?

What this cannot do

None of it guarantees that a model will quote you, and anyone who promises that is selling something they cannot deliver. What it does is remove the reasons you would be skipped: an answer buried too deep to quote, facts that exist only for humans, an identity that reads as four different companies, and a door closed to the crawlers that would have credited you.

All five parts are things you can do yourself, and this page is written so that you can. If you would rather not, we do the setup at a fixed price — $199, three working days, checklist included.