A beautifully designed PDF menu is invisible to most AI assistants. When someone asks ChatGPT "What's on the menu at Harbor Street Grille?" and that restaurant's menu lives only in a PDF, the AI either guesses from outdated cached snippets or admits it doesn't know — and then recommends somewhere else.
This is one of the most common and most easily fixed AI visibility gaps we see in a free scan. The fix takes an afternoon, not a developer.
Why AI can't read a PDF
AI assistants don't browse your website the way a human does. They work from text that has already been indexed and processed — either by search engines or by the AI company's own crawlers. Both have the same problem with PDFs: the format was designed for printing, not for text extraction.
A PDF menu might look clean on screen, but the underlying content is often a tangle of coordinates, font references, and character streams. When a crawler tries to extract "Spicy Lamb Flatbread — $18" from that structure, it might get back nothing, or garbled fragments. Multi-column layouts are especially bad — columns get merged or scrambled in ways that produce nonsense.
There's also a simpler failure: many websites link to a PDF without embedding any text on the page itself. The link says "Download our menu" — and the AI crawler sees a link to a binary file, skips it entirely, and moves on.
The businesses most affected
Any business that stores key information in a downloadable file is at risk, not just restaurants.
- Restaurants and cafés that upload seasonal menus as PDFs and swap them out quarterly
- Medical and dental practices with fee schedules or new-patient forms as PDFs
- Spas and salons with service menus in a designed PDF brochure
- Law firms with practice area summaries in a PDF capability statement
- Gyms and studios with class schedules as a PDF download
The unifying pattern: the business spent money on a designer, the PDF looks polished, and that polish is exactly what makes it unreadable to AI. Tables, multi-column layouts, decorative fonts, and images embedded in the layout all create noise that text extraction tools struggle with.
What happens when the AI can't read it
There are two failure modes, and both cost you customers.
The first is silence. The AI tells the user "I don't have current menu information for Riverside Bakery — you should check their website." The customer checks, doesn't find an easy answer fast enough, and picks a competitor whose information was already available as plain text.
The second is fabrication. The AI recalls partial information from an old Yelp review, a blog post someone wrote about your restaurant three years ago, or a cached version of a page that no longer exists. It then confidently reports an outdated price, a dish you discontinued, or hours that changed after your renovation. This is the failure mode described in why AI gets your business info wrong — the AI isn't lying, it's working with whatever text it could actually extract.
The fix: put text where crawlers can read it
The answer is not to abandon designed PDFs — they're fine for customers who want to download or print. The answer is to also publish the same content as plain HTML on your website.
For a restaurant, that means a /menu page with dishes and prices written out as real HTML text — not an image, not a PDF embed, not a JavaScript-rendered table that only appears after a button click. Plain, crawlable text, organized with heading tags (<h2> for category names, prices in the body copy).
For a spa or salon, that means a /services page with treatment names, durations, and prices as text paragraphs or a simple HTML table. For a medical practice, a /fees page with your most common procedures and self-pay rate ranges. For a gym, a /schedule page with class names, times, and instructors written out.
None of this needs to be elaborate. A simple, well-organized HTML page outperforms a beautifully designed PDF every time for this specific purpose.
Add schema markup once the text is there
Once your menu or service list exists as HTML, you can layer in structured data to give AI assistants even more confidence in what they're reading. The Menu and Service schema types tell crawlers explicitly what a block of text represents — not just words, but a priced offering from a named business at a specific location. The LocalBusiness schema guide covers the basics of how to mark it up.
Without the HTML page, schema markup is irrelevant — you can't add structured data to a PDF.
Check what AI currently knows about your business
Before rewriting pages, find out what AI assistants are actually returning for queries about your business today. A free scan shows your current A–F grade across the questions customers most often ask, including menu and pricing questions. If those are coming back wrong or blank, that's your starting point.
Once your HTML pages are live, re-run the scan. Most businesses see measurable improvement within two to four weeks as crawlers re-index the new content — and the fix is permanent, unlike chasing search rankings that fluctuate week to week.