What AI Crawlers Actually See on Your Product Page
Open a product page in your browser and you see a sofa, a price, the fabric and the dimensions. Ask ChatGPT about that same page and it may see nothing at all.
12 min read
By Alexander Jordan, Founder
Published: October 28, 2026
Open a product page in your browser and you see a sofa, a price, the fabric and the dimensions. Ask ChatGPT about that same page and it may see nothing at all.
The gap comes down to one technical detail: JavaScript. Most modern stores build at least part of each page in the browser, after the server's first response arrives. Google's crawler waits for that JavaScript to run. The AI assistants your customers now ask mostly don't.
The page that looks fine and says nothing
On 21 October we pulled the raw HTML for one product page at each of five premium furniture brands. Raw HTML is the first thing a server sends back, before any script runs.
Interior Define's Sloan 3-seat sofa was the clearest case. In a browser it's a normal product page. The raw HTML is an empty container and three scripts. No product name, no price, no materials or dimensions, no Product schema, not even a page title.
Everything a shopper sees is assembled in the browser by those scripts. A crawler that reads only the raw HTML gets an empty box.
Alt text: Interior Define's Sloan sofa page: the raw HTML is an empty container and three scripts, while the rendered page shows the name, price and photos.
The other brands show what the fix looks like. Article (Burrard sofa), Joybird (Holt sofa) and Burrow (Nomad sofa) all put the product name, dimensions and price in the raw HTML, with Product schema on top. We couldn't test Room & Board: its bot protection stopped our request before the product page loaded.
Why AI assistants miss it
When someone asks ChatGPT, Claude or Perplexity about a product, the assistant often fetches the page live through its own user agent: ChatGPT-User, Claude-User or Perplexity-User. Independent tests by GSQi in August 2025 and Search Engine World in June 2026 found that none of the three could read content that only appears after JavaScript runs.
The background crawlers that build these companies' indexes look the same so far. Vercel's December 2024 log study found that the major AI crawlers, including OpenAI's and Anthropic's, downloaded JavaScript files but didn't execute them. PerplexityBot behaved the same way. It is still the only large study of those crawlers' logs, so treat it as the best evidence available, not the final word.
Google is the exception. Its crawler renders JavaScript, and Vercel notes that Gemini runs on that same Googlebot infrastructure. That's why a page like Interior Define's can look healthy in Search Console while giving ChatGPT nothing.
There's also a timing point. Fastly's May 2026 network data found that more than half of AI requests had to be pulled from the origin server instead of a cached copy, against under 9% of human requests. AI assistants read what your server returns at that moment, which makes the raw HTML the version that counts.
Schema alone won't save you
The usual advice is to add structured data. It helps, but not the way most people expect. Tests by searchVIU and Otterly found that ChatGPT, Claude and Perplexity ignored JSON-LD on a live fetch, even when it was in the raw HTML. What they read is the visible text.
Schema still matters on Google, where it feeds rich results and AI Overviews, so keep it. Just don't rely on it to carry your product to an AI assistant. The name, price, materials and dimensions need to be on the page itself, in the HTML the server sends.
How to check your own product page
This takes ten minutes and needs nothing but your browser.
Open one of your best-selling product pages.
View the raw source (right-click, then View Page Source), or turn JavaScript off in your browser's developer tools and reload.
Search the source for four things: the product name, the price, the material and the dimensions.
Anything missing there is missing for an AI assistant too.
For the Google side, check that your Product and Organization JSON-LD appear in that same source and aren't added later by a script.
Run it on a product page, a collection page and your homepage. A template problem repeats on every page built from that template.
What fixes it
Send the important content from the server. Most frameworks can do this through server-side rendering or static generation for the product page itself. Keep JavaScript for the extras: review widgets, recommendations, chat.
If your platform makes that hard, start with the fields AI assistants quote most: what the product is, what it's made of, what it costs and how big it is.
Frequently asked questions
Do AI crawlers run JavaScript?
The live fetchers used by ChatGPT, Claude and Perplexity didn't in independent tests from 2025 and 2026. For the background crawlers, Vercel's December 2024 log study found the same. Google's crawler, which Gemini also uses, does render JavaScript.
If my page ranks on Google, is it fine for AI search?
Not necessarily. Google renders JavaScript before indexing, so a page can rank well while its raw HTML is empty. AI assistants that read only the raw HTML see the empty version.
Does schema markup help with ChatGPT?
Tests by searchVIU and Otterly found that ChatGPT, Claude and Perplexity ignored JSON-LD on a live fetch. Schema still helps on Google. Put the same facts in the visible page text.
How do I see what an AI crawler sees?
View your page's raw source, or reload it with JavaScript turned off. Whatever is missing there is missing for every AI assistant that doesn't run scripts.
See how AI describes your brand
Our free AI Perception Report shows which brands AI recommends in your category and why. Get yours at fironmarketing.com/scan.
Alexander Jordan
Founder, Firon Marketing
15+ years scaling brands. A strategic partner for high-growth founders, focused on sustainable revenue models and long-term equity value.
