Our specifications are in PDF datasheets. Do they need to be on the product page too?

The short answer

The question isn't whether Google can read a PDF. It's whether the right value for the right model can be found. On the 248 product pages we checked, 70% linked to a PDF, but only 26% showed a specification table of their own. The page was usually a short introduction and a download button.

What we found

PDFs can be indexed. That's not the question.

Google lists PDF among the file types it can index. So "nobody can find our specs because they're in a PDF" is the wrong starting point. Most of the datasheets we read, 90%, had a proper text layer that software can extract. A file being indexable doesn't mean a particular datasheet is indexed, shows for the right search, or is read correctly, but it does mean the format alone isn't the barrier.

The real issues are the ones a format test can't see: which values are on the page, which model they belong to, which revision is current, and whether the person who lands on the PDF from search can get back to the product and ask for a quote.

Where the specification lives

The median datasheet carried 13.5 specification values. The median product page carried 6. We counted numbers with units, such as pressures, voltages, dimensions and temperatures, plus IP ratings, and checked which of the datasheet's values also appeared on its product page.

• A median of 20% of the datasheet's values were on the page.
• 57% of datasheets had under a quarter of their values on the page.
• 24% had none at all.
• 15% had three-quarters or more, so it can be done.

Put plainly: on most of these sites, the answer to "what pressure is this rated to?" or "will it run on 24 V?" exists only inside a download. A buyer comparing three suppliers on their phone, or an AI tool summarising a product, works from the page in front of it.

What the datasheets themselves get wrong

Three things came up again and again, and none needs new content to fix.

• No revision or date. Only 23% of datasheets showed a revision number or date on their first three pages. Without one, a buyer, a distributor or an AI tool can't tell whether the file is current or two catalogues old.
• Scanned images. 10% had no text layer at all. They look fine to a person and carry no extractable text. Text that does exist was usually clean: only 3 of 121 text PDFs had even 1% scrambled characters.
• No way back. A third didn't mention a web address. Someone who lands on the PDF from search has no link to the product page, the range or the enquiry form.

The file's own title was also usually missing or left over from the design software: only 25% had a usable one, and the rest were empty or read like "Microsoft Word - Brochure.docx". Google's documentation doesn't say how it titles PDFs, so we treat this as housekeeping rather than a ranking fix, but it is what some viewers show as the document's name.

Three traps when moving specs onto the page

A separate hand check of six product families on 24 October showed why copying values across needs care.

• The page and the PDF can disagree. One heating manufacturer's product page and its linked specification sheet gave packaged weights 1 kg apart for the same model. Neither was obviously wrong, and only the manufacturer can say which is current. Agree a single source of truth before publishing values in two places.
• One datasheet often covers several models. A component maker's family datasheet listed a higher supply voltage for its premium variants than for the base part. A summary that drops the model suffix turns two correct values into one wrong one. Put the exact model next to every value.
• The file date isn't the revision. Several PDFs carried a 2026 modification date in their file properties while the printed revision was from 2024 or 2025. Use the revision printed on the document, not the date the file was last saved.

What AI tools said when we asked

We asked Perplexity and Gemini our question word for word on 24 October. Both recommended the same hybrid: key specifications in the page, detailed PDF kept for download. Gemini added that HTML is indexed much more effectively than PDF, a claim nothing it cited supports. On the first run it cited no sources at all; on a repeat it cited two commercial blogs. ChatGPT blocked our logged-out session, so it wasn't tested.

The advice was reasonable. What it lacked was evidence: neither answer could say how often the specification is missing from the page, which is the number a manufacturer needs before deciding how much work this is. How we run these checks is in our prompt-set guide.

What to do

• Pick one product family and list the five to ten values a buyer uses to choose. An engineer should decide which, not a marketer.
• Put those values on each product page in a plain table, with the exact model, units and test conditions next to each.
• Keep the PDF for the full specification, and print the revision and date on its first page.
• Make sure each datasheet has a text layer, a real document title, and the product page's address on it.
• Keep one source of truth for each value, so the page, the PDF, your feeds and your distributors don't drift apart.
• If a product isn't sold in a market, say so on that market's page rather than letting it redirect to a generic range.

If your catalogue is being rebuilt, this belongs in the plan from the start. Our redesign checklist covers who owns each step, and our note on schema explains why markup alone won't make a value visible if it isn't on the page.

How we checked

One checker, the same rules on every site. We sampled 64 manufacturer and supplier websites: Australian suppliers listed in a pump industry directory, excluding global groups, and UK and US manufacturers returned by searches for product datasheets across 14 product categories. 56 were reachable. For each, we sampled up to six product pages, took the datasheet that matched the product, and read it in memory. We recorded the response headers, the file's title, its text layer, any printed revision or date, and the specification values it shared with the page. Nothing was downloaded to keep, and no business is named.

Limits

• A search-based sample of 46 sites with usable product pages, mixing manufacturers and some distributors. It shows a pattern, not an industry average.
• We matched values mechanically, so differently written values count as misses.
• We checked what is on the page and in the file, not what Google has indexed or how either ranks.
• The three traps come from a small hand check of six product families. They illustrate the risk; they don't measure how common it is.