Can AI Search See Your Website? A Five-Minute Check

When someone asks ChatGPT or Perplexity for a good med spa in Boise or the best bakery in Asheville, the assistant names a handful of businesses. It builds that answer from what its crawlers could read. If they could not read your site, you are not in the answer, and no amount of good work changes that.

The useful news: whether they can read you is checkable in about five minutes, with no tools and no subscription.

The five-minute check#

1. Look at what the crawler actually receives. Open your homepage, right-click, View Page Source. This is the raw HTML, which is what AI crawlers fetch. Now search it (Ctrl+F or Cmd+F) for a sentence you know appears on the page: your opening headline, a service description, your opening hours.

Found it? Crawlers can read it. Missing, in a source that is mostly <script> tags? That content is assembled in the visitor's browser after the page loads, and most AI crawlers never execute those scripts, so to them the page is close to empty.

2. Check who is allowed in. Visit yourdomain.com/robots.txt and look for lines naming GPTBot, ClaudeBot, PerplexityBot or Google-Extended followed by Disallow: /. Some platforms and firewall products have blocked these by default at various points, so a block can exist without anyone choosing it.

3. Ask the assistants directly. Put your business name and city into ChatGPT, Claude and Perplexity and ask what they know. Then ask the question your customers would ask, the one that should surface you, and see who gets named instead. That list of competitors is your benchmark, and repeating this check monthly is a perfectly good tracking system that costs nothing.

The four things that hide a site from AI search#

Script-assembled content. Builder platforms and their app widgets often inject sections after the page loads: booking blocks, review carousels, tabbed panels, chat transcripts. A browser runs the scripts and sees everything. A crawler that does not run them sees the gap. Google copes because it renders JavaScript on a second pass; the AI crawlers largely do not.

Blocked crawlers. A Disallow aimed at AI user agents, or a firewall that challenges unknown bots, removes you from consideration entirely. This is sometimes deliberate policy, which is a legitimate choice. It should just never be an accident.

Content locked in images and PDFs. The scanned menu, the price list exported as a JPEG, the services brochure as a PDF download. Visitors read these fine. To a crawler the menu is one <img> tag with no text in it, and everything on it may as well not exist.

An inconsistent identity. Models assemble an answer about you from many sources: your site, directories, reviews, social profiles. If those sources describe you three different ways, with mismatched names, categories or addresses, the model has low confidence in all three versions and tends to recommend a business it is more certain about. One accurate description, used everywhere, beats three clever ones.

Which of these a migration fixes#

The first three are structural, which is why this topic belongs on a migration company's site rather than a marketing blog.

A static-first rebuild serves every word of your content in the raw HTML, because there is no editor runtime assembling the page in the browser. The menu becomes text, the services become text, and the crawler's copy of your site matches the visitor's. The same property is why those sites score well on speed tests: what makes a page fast for a phone on bad signal is what makes it complete for a crawler that reads once.

WordPress lands in nearly the same place, since it renders pages on the server, and this site itself is the static case: every page, including this post, is fully readable in source, with a machine-readable summary at /llms.txt for crawlers that look for one.

The fourth fix, identity consistency, is not a build problem at all. It is an afternoon of tedium: pick one accurate sentence about your business and make your site, your Google Business Profile, your directories and your social bios all say it.

What we deliberately are not telling you#

You may have seen striking statistics about AI search conversion rates and shopping behaviour. Some may even be true, but most trace to vendor decks selling tracking subscriptions, and we do not republish numbers we cannot check. The honest version is smaller and still sufficient: assistants recommend a short list instead of ten links, the short list is drawn from what crawlers can read, and the cost of being readable is low because it overlaps almost entirely with being fast and crawlable for Google.

That overlap is the practical point. You do not need a separate AI strategy. You need a site whose content is actually in the HTML, crawlers that are not blocked, and one consistent identity. The follow-up guide, generative engine optimization minus the hype, walks through the full checklist in order.

If your check just failed, the fix depends on why. Content missing from source on a builder platform is the structural case, and it is the one a rebuild solves. A blocked robots.txt is a one-line edit. A menu trapped in a JPEG is an afternoon of typing. Only one of those needs anyone's help.