The short answer
If your menu, price list, or service sheet lives only in a PDF, there's a good chance ChatGPT, Claude, and Perplexity either skip it entirely or guess at what's in it. PDFs — especially scanned ones or ones built as flattened images in a design tool — don't hand over clean, readable text the way an HTML page does. AI assistants either can't extract the words at all, or they extract them in a scrambled order that turns "Grilled Salmon, $24" into gibberish.
The fix isn't complicated: put your menu (or price list, or service catalog) on an actual web page, as real text, in addition to whatever PDF you keep around for printing. This one change closes one of the most common gaps we see when we scan restaurant and service-business websites.
Why PDFs trip up AI assistants
There are three separate failure modes, and most PDF menus hit at least one:
Scanned or image-based PDFs. If the menu was photographed, faxed, or exported from a design tool as a flattened image, there's no text layer at all — just pixels. An AI crawler sees a blob, not words. It can't quote a price it can't read.
Multi-column layouts. Even a "real" text PDF with appetizers in one column and entrees in another often gets extracted in the wrong reading order. A parser might interleave "Caesar Salad" with "Ribeye Steak" and produce a menu that makes no sense, because PDF text extraction reads by position on the page, not by logical structure.
No semantic structure. A web page can mark up "Appetizers" as a heading and "Caesar Salad — $12" as a menu item underneath it. A PDF has no such concept. Everything is just text sitting at x/y coordinates. AI tools that do manage to pull text out still don't know what's a section, what's a price, and what's a description.
The practical result: ask an AI assistant "does this restaurant have a gluten-free option" or "how much is the lunch special," and if the answer only lives in a PDF, the assistant either says it doesn't know or invents a plausible-sounding wrong answer. Neither is good for you.
What "plain text" actually means here
You don't need to abandon your PDF — keep it for people who want to print it or save it to their phone. What you need is a second, parallel version that's a normal web page.
A bakery in this situation might have /menu.pdf linked from their homepage and nothing else. The fix is adding /menu as an actual page: an h2 for each section ("Breads," "Pastries," "Custom Cakes"), each item as a list entry with its name, price, and a short description in plain paragraph text. No canvas rendering, no image of a menu photographed at an angle, no text baked into a JPEG.
For a dental practice, the equivalent is a services and pricing page instead of (or alongside) a "download our brochure" PDF. For a plumber, it's a services list with typical price ranges instead of a scanned rate sheet.
This also plays into structured data. Once your menu or services exist as real HTML, you can mark them up with schema — Menu and MenuItem types for food businesses, or Service and OfferCatalog for service businesses. That's the same underlying idea covered in our guide on schema for local businesses: give AI systems an unambiguous, machine-readable version of the facts instead of making them guess from a layout.
A quick way to check if this is costing you
Ask an AI assistant a specific question only your menu or price sheet would answer — "what's the price of a basic oil change at [your shop]" or "does [your restaurant] have a vegan appetizer." If the assistant hedges, gives an outdated answer, or clearly hallucinates a number, that's a strong signal it never got clean text to work from. This is the same class of problem covered in why AI gets your business info wrong — bad or missing source material, not a smarter competitor, is usually the real cause.
It's also worth pairing this fix with an llms.txt file, which tells AI crawlers directly where your authoritative pages live instead of leaving them to guess between your PDF, your Yelp listing, and a three-year-old cached version of your site. We cover the setup in our llms.txt guide.
Next step
Pull up your site and count how many "facts" — menu items, prices, service lists, hours, policies — exist only inside a PDF or an image. Each one is a place an AI assistant either goes silent or makes something up. Converting even your top three PDFs into real text pages is a half-day project with an outsized payoff. If you want a starting list of what's currently missing or wrong, run a free scan — it flags exactly which facts AI assistants can and can't find on your site today.