Resources · Strategy

The AI SEO checklist

The short answer

An AI SEO audit comes down to five checks. Can crawlers reach you: fast pages, AI bots allowed in robots.txt, clean HTML, a live sitemap. Can a model lift a clean answer from your content: question-first pages, self-contained passages, sourced claims. Does your identity match across your site, G2 and the places buyers talk. Do outside sources back what you say. And are you tracking the same buyer prompts across ChatGPT, Gemini, Claude and Perplexity over time. None of these is a trick. They only work together.
Updated August 20267 min readReviewed by PageLens.ai

Most advice about AI search runs to either panic or hand-waving. This is the plain checklist underneath it. The items below decide whether ChatGPT, Gemini or Perplexity can read your site and name you when a buyer asks for a recommendation. Run each one against your own pages.

They fall into five groups. The first two live on your own site, so you control them outright. The last three depend partly on other people, which is the part most teams skip. Work top to bottom. If the technical checks fail, nothing below matters, because a model can’t cite a page it never fetched. All of this is the groundwork under answer engine optimization.

Technical: can a crawler reach you?

None of the clever writing matters if the page never gets fetched. Start here, because a fail on this list quietly cancels everything else.

  • Pages load fast, and the main text sits in the HTML instead of getting painted in later by JavaScript. A model that hits a blank shell or a slow response gives up and moves on.
  • Open your robots.txt and actually read it. If it blocks GPTBot, OAI-SearchBot or the other AI crawlers, you’ve bolted the door from the inside without meaning to.
  • The HTML is clean: one h1, headings that describe what follows them, and text that lives in the markup rather than a stack of empty divs.
  • Your sitemap is current. It lists the pages you want read, and it doesn’t point at URLs that 404.

Content: is the answer easy to lift?

Models quote passages, not whole pages. Every item here makes a passage easier to pull out and harder to misread once it’s out of context.

  • The first sentence of each important page answers the question that page exists for. Save the brand story for lower down.
  • Write your H2s as real questions.“How much does it cost” pulls better than “Pricing.”
  • Passages stand on their own. Lift any single paragraph off the page and it should still be true and still make sense with nothing around it.
  • Claims carry specifics. A number, a date or a named source does more work than a paragraph of adjectives.
  • Add an FAQ where buyers have obvious follow-up questions, and keep each answer to a sentence or two.
  • Nothing important is out of date. A page that was right last year and wrong now slips out of the answer without telling you.

Entity: does your story match everywhere?

A model assembles a picture of who you are from every place it reads about you. When those places disagree, you turn into a risky thing to recommend, so the model reaches for a brand whose story is cleaner.

  • Your name, category and one-line positioning read the same on your homepage as they do on G2, Capterra and any community profile you keep.
  • Add Organization schema so the machine-readable version of your name, URL and logo lines up with what a person sees on the page.
  • Old product names and retired taglines aren’t still floating around contradicting the current one. Pick the real story and make every source agree with it.

Off-site: who backs up your claims?

This is the group teams control least and neglect most. A model weighs what other people say about you more heavily than what you say about yourself, and you can’t edit those pages directly.

  • You have reviews from recently, not a stack of five-star ratings from three years ago. Recency reads as still in business.
  • You show up where buyers compare notes: a Reddit thread, a forum answer, a category roundup that happens to name you.
  • Third-party mentions agree with your own claims. When you say you’re the affordable pick and outside sources say the same, that agreement is what becomes an AI citation.

Measurement: a trend, or a vibe?

You can’t fix what you’re not watching, and checking once tells you almost nothing. What you want is the direction over weeks, not one lucky answer you screenshotted on a good day.

  • You keep a fixed list of prompts, the real phrasings your buyers would type. Freeze the wording so this month is comparable to last.
  • You run them across ChatGPT, Gemini, Claude and Perplexity, because each engine names a different set of brands.
  • Run each prompt a few times. The answers drift between attempts, so a single reply proves nothing on its own.
  • You track whether you get named more often now than a month ago. That direction is the metric, not any one answer.

Where to start

You don’t need to do all of this at once, and you shouldn’t try. Run the technical checks first, since they gate everything else. Then fix the one content page that matters most to your business and watch whether it moves.

The list compounds. A fast site with clean answers, matching profiles and outside proof behind it adds up to something no single item does alone. So treat this as an opening audit, not a finish line. Once you’ve been through it, the fair next question is what the models actually say about you today, which you can check in about a minute with a free audit.

Common questions

Is AI SEO different from regular SEO?
They share a foundation. A fast, crawlable, well-structured site is required for both. The goal is what differs: classic SEO ranks a page to earn a click, while AI search wants a passage it can quote inside an answer, often with no click at all. You build the second on top of the first.
Do I have to let AI crawlers in?
If you want to be cited, yes. Blocking GPTBot or OAI-SearchBot in robots.txt keeps your pages out of the sources ChatGPT can pull from. Some publishers block them on purpose to protect their content. For a brand that wants to be recommended, it's usually an own goal.
How long before any of this shows up in answers?
It varies. Live search can re-crawl a strong new page within weeks, while a model's training data updates in big, infrequent jumps. That's the reason the last group is measurement: you watch the trend instead of guessing at a single answer.