What AI Actually Sees When It Visits Your Website

Last updated: 2026-05-27

When ChatGPT, Claude, Perplexity, or Google’s AI describes your business, it works from a parsed copy of your website. Not a screenshot. It strips out the design, ignores most images, and reads what’s left as plain text. If that plain-text version is thin, contradictory, or hidden behind JavaScript, AI will either get the wrong picture or skip you entirely. The gap between what your homepage looks like in a browser and what it looks like to a crawler is wider than most owners expect.

This matters for solo practitioners and small Canadian practices because most independent professional sites were built for human visitors. They look fine in Chrome. They look terrible to a crawler. Here is what AI actually sees on a typical small-business site in 2026, and what changes that.

Which AI crawlers visit a small business website?

Four crawlers do most of the work for AI today, and any of them can reach a small practice site within hours of publishing. They identify themselves in your server logs by user-agent string, so you can confirm which ones are reading your pages. The four names to know: GPTBot from OpenAI, ClaudeBot from Anthropic, PerplexityBot from Perplexity, and Google-Extended from Google. Each feeds a different AI surface.

GPTBot. OpenAI’s crawler for ChatGPT and SearchGPT.
ClaudeBot. Anthropic’s crawler for Claude.
PerplexityBot. Perplexity’s live retrieval crawler.
Google-Extended. Google’s signal for Gemini and AI Overviews.

These bots fetch your pages the same way a basic web reader would. They request the HTML, follow internal links, and pull out readable text. A few attempt to render JavaScript. Most do not, or do it poorly. If your homepage builds its main content with JavaScript after the page loads, a portion of these crawlers will see an empty shell where your services should be.

What does an AI crawler actually grab from a page?

A crawler grabs the parts of your page that look like a printable document. Title tag and meta description first, then heading tags, then body text inside paragraphs and lists, then anchor text on internal links, then any structured data you have published. Image alt text is read. The image itself is not. The text in the first few hundred words of your main pages carries most of the weight when AI builds its picture of who you are.

What it ignores or mishandles: background images, decorative SVGs, animations, carousel slides that load after the page renders, content tucked behind tabs or accordions that need a click, video transcripts not published as text, and anything inside a chat widget.

So if the most compelling thing on your front page is a rotating office photo with a tagline overlaid, that is invisible. If your service descriptions live in expandable accordions, half the crawlers will miss them. If your About page is a single beautifully designed image, AI sees nothing at all.

Why does AI sometimes get my business wrong?

AI gets a small practice wrong for three reasons that show up in nearly every audit. The site fails to describe what the business actually does in plain text, the service pages are named for marketing instead of for queries, and the contact details on the website disagree with what Google and the provincial regulators publish. Any one of these is enough to push the model toward a competitor.

Your homepage doesn’t say what you do. Many small-practice front pages open with a tagline like “Care that puts you first” or “Trusted advice since 1998.” A human reads that and looks for context in the photos or the navigation. A crawler reads it and moves on. Without a plain sentence like “Independent personal injury law firm in Hamilton, Ontario,” the model has to guess.

Your services live on separate pages with vague titles. A page called “Solutions” or “Approach” tells AI nothing. A page titled “Estate Planning for Ontario Families” gets categorized correctly the first time a crawler visits.

Your contact details disagree with the rest of the web. AI cross-checks the site against your Google Business Profile, your provincial regulator, directory listings, and review platforms. If the address on the website is the old one, or the hours haven’t been updated, the model often takes the third-party source as truth and treats your own site as outdated.

What are the fastest fixes for what AI sees?

Most independent practices can close the visibility gap in a single afternoon, with no redesign required. The fixes below are the ones that come up most often in audits for owner-operated firms. None of them change how the site looks to a human visitor. They change how it reads to a machine.

Put a one-sentence description of the business in the first 100 words of the front page. Include city, profession, and one specialty.
Use clear H1 and H2 headings on every page. “Family Law Services” beats “What We Do.”
Move accordion and tab content into visible body text, or duplicate it on a clean services page.
Add LocalBusiness or relevant Organization schema. It is a small JSON block that tells crawlers your name, address, phone, hours, and service area in a format they cannot misread.
Make sure the main content appears in the HTML, not after a JavaScript load. If you are not sure, right-click the page and pick “View Page Source.” If you can find the main copy there, you are fine.
Confirm name, address, and phone number match across the website, Google Business Profile, and the regulator’s directory.
Publish a short FAQ page with five to seven questions and direct answers. FAQs are the most extractable format for AI.

How do I check what AI sees on my site right now?

Open the home page, right-click, and pick “View Page Source.” Search the raw text for the business name, the city, and the main service. If those words appear in clear sentences, the site is in good shape. If they only appear inside script tags or image filenames, AI is missing them. This 30-second test catches a large share of the basic visibility problems we see in audits, before any paid tool is needed.

For a sharper picture, fetch the site as a crawler would. A free tool like cURL, or a browser extension that disables JavaScript, will show you what GPTBot sees on its first pass. The difference between the rendered page and the raw text is usually startling.

Frequently Asked Questions

Do AI crawlers obey robots.txt?

Yes, the major ones do. GPTBot, ClaudeBot, PerplexityBot, and Google-Extended all honour robots.txt directives as of 2026. If you block them, they stop crawling within a few days. Most small practices should not block them. Blocking removes the firm from the AI training and retrieval pipeline entirely, which means competitors get cited instead.

Will adding schema markup make my site look different?

No. Schema is invisible JSON-LD sitting in the page source. Human visitors never see it, and it does not change the layout, fonts, or images. It is read only by crawlers and search engines. Google’s own documentation lists structured data as one of the strongest signals for accurate entity recognition.

How long does it take for AI to notice changes to my site?

For retrieval-based AI like Perplexity, ChatGPT with browsing, and Google AI Overviews, updates often appear within a few days of being crawled. For training-data-based answers in the base ChatGPT and Claude models, it can take months until the next model refresh. This is why clean live content and consistent third-party citations both matter at the same time.

Is a one-page website enough for AI?

It can be, if the page covers services, location, credentials, and contact details in clear text. A thin one-pager with mostly images usually loses to a competitor with five well-written pages. Independent practices in regulated industries also benefit from a dedicated About or Credentials page, which gives crawlers a clean place to grab licence numbers and association memberships.

Does it matter if my site is on Wix, Squarespace, or WordPress?

Not directly. What matters is whether the content renders in the HTML and whether you can add schema markup. All three platforms can produce crawler-friendly sites, and all three can produce invisible ones. The build choice matters less than what ends up in the HTML.

Want to see exactly what AI sees on your website? Run a free AI visibility audit at manta-aeo.com. We will show you which crawlers reach your pages, what they extract, and where you are being missed by ChatGPT, Claude, Perplexity, and Google.