Which Old Pages, PDFs or Docs Make AI Repeat Wrong Details About My Business?

Where should you look?

Also check the sources AI answers cite. Ask assistants the questions where wrong details appear, open every cited link, and note which ones are yours. Then search for the wrong detail itself, such as an old price or a former manager's name, in quotation marks. The pages that still contain it are your list.

How do you prioritise?

Rank each stale item by three things:

1. How often it is cited in AI answers and how much traffic it gets.
2. How harmful the wrong detail is: wrong prices, allergens, legal or safety information come first.
3. How easy it is to fix: a redirect takes minutes, while rewriting a help centre takes weeks.

Fix the most-cited, most harmful items first. A single old PDF price list cited in half your pricing answers matters more than fifty obscure pages.

Which outdated Notion pages are being cited by AI assistants about our services?

Notion pages published to the web can be indexed by search engines when the page's search engine indexing setting is turned on. Teams often publish pages for convenience, such as an onboarding guide, a service overview or a hiring page, and forget them. To find them:

• Search for your company name together with Notion's public site domains.
• Check your workspace's published pages list, or ask workspace admins to review which pages are public.
• Look at AI citations for any Notion URLs.

Then, for each page, decide whether it should be public. If not, unpublish it. If it should stay public, update it, or turn off search engine indexing and link to the official page on your website instead. The same applies to other workspace and document tools that publish pages to the web.

What should I fix first when AI assistants cite an outdated PDF menu instead of our website?

Replace the PDF with an HTML menu page, then deal with the old file.

1. Publish the menu as a text page at a stable address such as /menu, with dishes, prices, dietary and allergen information and the date of the current menu.
2. Redirect the old PDF's URL to the new menu page, so links and citations reach current information.
3. If you keep a downloadable PDF, give it a stable file name, not "menu-2023.pdf", and update it whenever the HTML menu changes.
4. Update your Business Profile's menu link and delivery platform menus.
5. Ask third-party menu sites that copied the old menu to update it.

Assistants find HTML easier to read and quote than PDFs. Menus in images or scanned PDFs are worse still, since the text may not be readable at all. Allergen accuracy deserves special care, as covered in what food businesses should do when AI makes false allergen or gluten-free claims.

What are the options for each stale item?

• Update it if it still has a purpose.
• Redirect it permanently to the current equivalent if one exists.
• Remove it and return a 404 or 410 if it has no successor.
• Keep it but stop indexing if it must stay available, such as an archived report. For PDFs and other non-HTML files, Google's documentation says to use the X-Robots-Tag HTTP header, since a PDF cannot carry a robots meta tag.
• Label it clearly as historical if it must stay public and indexed, with a link to the current version at the top.

A robots.txt block alone is not enough, since blocked files can stay indexed if other pages link to them. Why pages stay cited after a robots.txt change is covered in why AI assistants still cite a page after robots.txt starts disallowing it.

How do you stop stale content building up again?

• Keep an inventory of public files and pages, with an owner and a review date for each.
• Publish important facts in HTML on your site, and use PDFs only as downloads that mirror the page.
• Use stable URLs for things that change, such as menus, price lists and brochures, instead of dated file names.
• Review public workspace pages every quarter.
• Retire campaign pages and microsites when campaigns end, with redirects.

What might an illustrative case look like?

Consider an illustrative bistro whose AI answers quoted prices two years out of date and listed a set lunch it no longer offered. The cited source was "lunch-menu-2024.pdf", still on its site and linked from an old blog post and a local guide. The bistro's current menu existed only as an image on its homepage. The owner published a text menu page at /menu and redirected the old PDF there. They updated the Business Profile menu link, asked the local guide to change its link and replaced the image with text. Within three weeks, answers quoted current prices and cited the menu page.

What do people ask about old files and AI answers?

Do AI assistants read PDFs?

Many can, if the PDF contains real text. Scanned or image-only PDFs are much harder to read, and HTML pages are easier to quote accurately.

Should we delete every old PDF?

No. Keep the ones with lasting value, clearly labelled, and redirect or remove the ones that carry outdated facts.

How do we find files we have forgotten about?

Search your domain for file types, check Search Console's indexed pages and review server logs for requests to old files.

Can a cached copy keep an old PDF alive after we remove it?

For a while, yes. Search indexes and assistants may hold the old version until they recrawl. A redirect to the current page, or a 404 or 410, tells them it has changed. Copies on other sites stay until those sites update them.