Direct answer: Keep the current, decision-critical facts in a usable HTML product page and offer the PDF for the complete document or offline handover. Google can index PDFs. The reason for an HTML counterpart is to give buyers a clear, navigable answer path and a place to maintain the current key facts, while the PDF itself still needs accessibility and version checks.
A common site pattern puts “Download specifications” beside a product photo and leaves dimensions, compatibility, installation conditions, and exclusions inside a 28-page PDF. A buyer on a phone may want one fact now; a procurement team may need the full sheet later. Those are different reading tasks.
Google's indexable-file list explicitly includes PDF. That rules out the easy claim that a PDF is invisible to Google or to every AI system. It does not establish that every page, file, table, or image inside a PDF will be retrieved, understood, or cited. Write the web page for the user's decision rather than for a presumed parser preference.
Divide the information by task
On the product page, put the facts a buyer checks before requesting a quote or purchase: model identifier, dimensions and units, material, compatibility, available options, what is included, and the date or version when those facts can change. Use real text and headings, not a screenshot of a table. Link the current full PDF with a descriptive label such as “Installation specification, revision 4, PDF,” and show its date if relevant. If a previous edition must remain available, label it as an archive and explain which product generation it covers.
The PDF still deserves its own quality check. Adobe's accessibility guidance explains that tags and logical reading order help screen readers interpret a PDF, and that automatic tagging can misread complex layouts and tables. Open the delivered file, check whether text can be selected, inspect headings and table order with an accessibility tool, and ask someone to follow the document without relying on its visual layout. A beautiful export can fail that test.
Keep the two versions honest
Assign a single owner for specification changes. When a value changes, update the HTML and PDF together, record the revision, and check that old links do not present an obsolete file as current. If the HTML page and PDF are truly duplicate versions and you want one preferred search URL, Google supports a rel="canonical" HTTP response header for non-HTML files. An HTML <link> tag cannot be inserted into a PDF. Canonicalization is a hint that Google may or may not select; do not point a full technical PDF at a short summary as if they were duplicates.
For a scoped ARCHV website, the useful deliverable is a content model: which facts are visible in HTML, which document is authoritative, who updates both, and where the buyer can act. This is different from the JavaScript configurator search-visibility article, which covers a dynamic interface. Here the risk is a static document that becomes the only or outdated source of a key claim. Bring a current and an older specification sheet to a website estimate to identify the maintenance work.