The Invisible Catalogue: Why ChatGPT Cannot See Your Products

This is the second of three articles on why good Indian business websites go unmentioned. Part 1 covered the faults in how a site describes itself; this one is about whether the assistant can read your products at all.

Why can an AI assistant not read a product page that a customer reads easily?

AI assistants read a web page the way a fast, literal clerk would: as plain text, in the order it appears, in the few seconds the page takes to load. Anything that needs a click, a zoom or a pair of eyes to interpret is skipped, and anything behind a download is read badly if at all. A human visitor does all of those things without noticing. The assistant does none of them and moves on to the next site.

Retrieval at this scale favours plain text, and nothing announced by any of the assistant companies suggests that changing soon. When an assistant answers a buyer's question, it fetches pages from many sites, extracts the text it can lift cleanly, and builds an answer from the extracts. A page that gives it nothing to lift is simply absent from the answer.

The result is that the assistant's picture of your company is built from whatever happened to be in ordinary text on the page. On many Indian B2B sites that is the company slogan, the navigation menu, and a line inviting the reader to download the brochure.

What happens when AI assistants cannot see your product range?

The cost is the enquiry that goes to the competitor whose product pages the assistant could read. When a purchase officer asks "Which companies in Kolkata supply stainless steel 316 fasteners in M12 to M24?" the assistant names the two or three suppliers whose pages carry those words as text. If your range covers exactly that, but the range is inside a PDF, you are not in the answer.

Because the enquiry never arrives, you never learn it existed, and as Part 1 of this series explained, your Google traffic looks normal while it happens.

This affects exactly the businesses that have invested most in their catalogues. A firm with a beautifully printed sixty-page brochure has usually put that brochure on the website as a PDF and considered the job done. A trader with three hundred SKUs has usually photographed the price list. The more effort went into the physical catalogue, the more likely it is to be hidden online.

Hiding place 1: Why is a PDF brochure not enough for AI assistants?

A PDF brochure is designed for printing, not for reading by software. Text is placed by position rather than by meaning, columns run in unpredictable order, product names sit in images, and specification tables lose their structure once extracted. Some assistants fetch PDFs and some do not; the ones that do often extract a jumble. In practice, product information that lives only in a PDF is unreliable at best and absent at worst.

The typical pattern on an Indian manufacturer's site is a "Downloads" or "Catalogue" page carrying five to ten PDFs, each a scan or an export from the designer's layout software, with the actual product pages on the website carrying nothing more than a product name and a "Download Brochure" button. To a buyer with a mouse, this works. To an assistant, the product pages are empty and the PDFs are noise.

A PDF is a fine supplement, but never a substitute for the same facts stated as text on the page where the product lives.

Hiding place 2: Why are prices and specifications inside images invisible to AI?

Text inside an image is not text to a machine, unless the same words appear in the image's alt text or caption, and on a photographed price list they never do. A photograph of a price list, a scanned specification sheet, a designer's JPEG of a comparison table, a banner carrying the product range in stylised lettering: every one of these is invisible to an assistant reading the page as text. The assistant sees an image file with a name like "final-price-list-2026-v3.jpg" and nothing else.

Image-based content is common on Indian trading and distribution sites for a practical reason. The price list changes, the designer is busy, and the quickest way to update the site is to photograph the new sheet and upload it. The practice is efficient and it makes every price and every specification disappear from AI answers.

The same applies to infographics that summarise capabilities, to logos of client companies used as proof, and to product photographs with the specification printed on the label. If the fact matters, and it exists only inside pixels, it does not exist to the assistant.

Hiding place 3: Why do tabs, sliders and "read more" buttons hide your content?

Many product pages show a summary and hide the details in tabs ("Specifications", "Applications", "Downloads"), accordions, sliders, or "Read more" links that load the content only when a person clicks. Some of that content is present on the page from the start and merely hidden from view. Some of it is fetched from the server only after the click. The second kind never reaches an assistant, because an assistant does not click. A third kind is worse: some templates draw the whole product page with a script after the page loads. Google's crawler runs that script. The readers most assistants send do not, so they receive a page with a menu and nothing else, and your Google ranking will not warn you.

Owners rarely know which kind they have. The designer chose a template, the template used a slider, and the specifications ended up loading on demand because that is how the slider works. The page looks tidy. The most valuable text on it is unreachable.

Product comparison tools, configurators and "select your variant" dropdowns fall into the same category. A page that offers a buyer forty variants through a menu, with the details appearing after selection, offers an assistant only what the page shows before anyone selects anything, which is usually one default variant.

Hiding place 4: Is your website blocking AI crawlers without your knowledge?

Some websites are configured to refuse AI readers entirely, and the owner usually does not know. Since July 2025 Cloudflare, the most widely used website protection service, has blocked AI crawlers by default on newly added domains, and some hosting companies and WordPress security plugins do the same. A line in the site's robots file, added by an SEO plugin, has the same effect. A setting switched on to stop content theft or reduce server load also stops the assistants that would recommend you.

There is a legitimate debate about which AI systems should be allowed to read a site. Some businesses choose to block the crawlers that collect training data while allowing the ones that fetch pages to answer a live question. That is a reasonable choice, but it should be a choice you made, not a default your hosting provider applied while you were not looking.

The sign that this has happened is a site that is otherwise well built, with product information in ordinary text, that still never appears in assistant answers. Nothing on the site explains it, because the site is fine. The readers were never let in.

How can you check in ten minutes whether your products are visible to AI?

You can find out roughly where you stand without any tools. Pick your single most important product, the one that brings the best enquiries, and run these three checks.

1. Ask ChatGPT, Perplexity and Gemini "Which companies supply [that product] in [your city or state]?" If you do not appear and a competitor does, open the competitor's product page. Look at whether its details are in ordinary text. In most cases they are.
2. Open your own product page, right-click and choose "View page source", then press Ctrl+F and search for one specification you know is there: a size, a grade, a capacity. If it is not in the source, an assistant that fetches the page will not find it either. It was inside an image, a PDF, or a script.
3. Ask Perplexity "Summarise the products listed at [your product page URL]". If it says it cannot access the page, or summarises only the navigation menu, the page is either empty to a machine or closed to it.

What these checks cannot tell you is which of the four hiding places is responsible for most of the loss, or whether a block sits at the hosting level, the plugin level or the page level. That needs a proper look at how the site serves its pages to different readers, which is one of the things the audit does.

Why is moving the catalogue into text not just a copywriting task?

Putting product facts into ordinary text on the page sounds like a job for a content writer. Two of the four hiding places are indeed fixed by writing: stating the facts on the page rather than pointing at a PDF, and replacing images of tables with real tables. The other two are technical. Content behind clicks needs the template changed so that the details are on the page from the start. A block needs the protection settings reviewed reader by reader, and that review is easy to get wrong in both directions.

There is also a sequencing question. A site with three hundred products cannot rewrite every page at once. The audit takes your list of the products that bring the enquiries that matter, checks how each of those pages is served, and sets an order that puts the highest-value pages in front of the assistants first.

Questions people ask about product pages and AI search

Can ChatGPT read PDF files on my website?

Sometimes, and unreliably. Some assistants fetch PDFs and extract text, but layout-heavy brochures come out as a jumble and image-based PDFs come out empty. Treat PDFs as a supplement for humans, not as the place where product facts live.

Do I have to remove the PDF brochure?

No. Keep the brochure for buyers who want to print or forward it. State the same product facts as text on the page as well, so that both the buyer and the assistant can read them.

Is it safe to allow AI crawlers on my website?

Allowing the readers that fetch pages to answer live questions carries little risk and is a precondition for being recommended. Whether to allow the crawlers that collect training data is a separate decision. The problem is not either choice; it is a block applied by default that you never made.

My product pages are on IndiaMART and TradeIndia. Is that enough?

Marketplace listings help assistants confirm you exist and what you sell, but the assistant will often recommend the marketplace page, not you, and it will list your competitors on the same page. Your own site needs to carry the product facts as well.

Will fixing this help my Google rankings too?

Usually, yes. Google's rankings and its AI Overviews are built from the same index, and text on the page is what that index reads most reliably. Product facts moved from images and PDFs into page text tend to improve ordinary rankings for those product terms as a side effect.