AI Assistants Are Sending Visitors to URLs That Never Existed

AI assistants now send real visitors to websites. When someone asks ChatGPT, Perplexity, Gemini or Copilot for a tool, a guide or a product, the answer often includes a link. The problem is that the link is not always real. Language models can produce URLs that look perfectly plausible for your site — /blog/how-to-fix-x, /pricing/enterprise, /products/blue-widget — even though those pages were never published.

The visitor clicks, lands on a 404, and leaves. You never see a broken link in your crawler because no page on the web actually links there.

Why AI assistants generate broken URLs

There are three common causes:

• Invented paths. The model predicts what a URL on your domain "should" look like based on your naming patterns and other sites in your niche.
• Outdated training data. The URL existed once. You renamed it, migrated the site or deleted the product, but the model still remembers the old address.
• Mixed-up details. A slug from one article, a folder structure from another, or a competitor's URL format combined with your domain name.

None of these show up in Search Console as linked from anywhere, which is why they are easy to miss.

How to identify AI-driven 404 traffic

Look at the 404 hits themselves, not just at crawl reports. Useful signals:

• Query parameters. ChatGPT often appends utm_source=chatgpt.com to outbound links. We covered a surprising side effect of this in How Did utm_source=chatgpt.com End Up in Disney+'s Canonical URL?.
• Referrers. Domains such as chatgpt.com, perplexity.ai, gemini.google.com or copilot.microsoft.com, when the browser passes a referrer.
• URL shape. Clean, descriptive slugs that follow your patterns but were never created are a strong hint.

A 404 log that records the full URL, query string, referrer and hit count makes these patterns visible within days. Server logs work too, but they need filtering.

What to do with each URL

Group the URLs and decide case by case:

1. Clear intent, matching page exists. If /pricing/enterprise gets hits and your enterprise plan lives at /pricing#business, a 301 redirect to that page is a good fix.
2. Old URL that used to exist. Redirect it to the current equivalent, exactly like you would after a migration. See How to Fix 404 Errors After a Website Migration.
3. Repeated demand for content you don't have. If many visitors land on an invented guide URL, that is a content idea. Writing the page and publishing it at that exact URL can turn a mistake into a traffic source.
4. Random, one-off URLs. Leave them as 404. Not every miss deserves a redirect.

Avoid catch-all rules that send every unknown URL to the homepage. As we explained in Why Redirecting Every 404 to the Homepage Is a Mistake, this frustrates users and can be treated as a soft 404.

Reduce future mistakes

You can't fully control what a model writes, but you can make the correct URL easier to find:

• Keep URLs stable and descriptive. Frequent slug changes multiply outdated references.
• Redirect old URLs permanently when you change them, so every past version still resolves.
• Make key pages (pricing, docs, product categories) easy to reach from your homepage and sitemap.
• Keep a helpful 404 page with search and main categories, so a visitor who still lands on a wrong URL has a way forward.

Summary

AI assistants are a new source of 404 traffic that traditional broken-link tools can't detect. Track real 404 hits with their referrers and query strings, redirect the URLs with clear intent, turn repeated demand into content, and let random misses stay 404. no404 records every 404 hit in real time so you can spot these patterns and fix them with a single redirect rule.