How Do AI-Generated Answers Send Visitors to 404 Pages Instead of the Right Content?

How big is the problem?

Ahrefs checked the pages that AI-referred visitors landed on across sites using its analytics tool, looking for titles containing "404" or "not found":

It also checked the URLs AI answers cited, whether clicked or not. There, 2.38% of ChatGPT's cited URLs returned 404, against a baseline of 0.84%. Ahrefs notes the method probably under-counts, since some error pages lack those words in their titles.

A percentage around 1% sounds small. But AI-referred visitors often convert well, and the ones landing on a 404 were sent for a specific page. For a site receiving thousands of AI visits a month, that can mean dozens of visitors a month who arrived wanting a specific page and found nothing. Crawlers see it too. A 2024 study by Vercel and MERJ found about 35% of fetches by ChatGPT's and Claude's crawlers hit 404 pages, against about 8% for Googlebot.

Why are LLMs sending traffic to URLs that no longer exist?

• Stale memory. When an assistant answers from what it learned in training rather than searching, it can only remember URLs as they were. Pages renamed or deleted since then live on in its answers.
• Old third-party sources. Directories, partner pages, old articles and forum posts still link to former URLs, and answers cite what those sources say.
• Site changes without redirects. Migrations, CMS switches, product retirements, changed blog slugs and deleted dated pages all break links when redirects are missed.
• Invented URLs. Ahrefs found URLs that matched the pattern of its blog but had never existed. A model that knows a site has /blog/ posts on a topic can guess a plausible slug.
• Mangled URLs. Truncated paths, missing trailing segments, wrong case and added parameters turn a real address into a broken one.

The pattern may be shifting. Similarweb found that after ChatGPT began showing clickable brand names in answers in May 2026, homepages' share of ChatGPT referrals rose from about 26–29% to about 62–63%. Homepage links rarely break, but deep links from citations and model memory still do.

How do you find the 404s AI visitors hit?

1. Tag your 404 template. Fire an analytics event or set a page-type dimension on the error page, so every 404 visit is countable with its referrer and requested URL.
2. Filter by AI referrers: chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com and others you see. Include sessions tagged utm_source=chatgpt.com.
3. Check server logs for 404 responses with those referrers. Also check 404 responses to ChatGPT-User and Perplexity-User. These agents fetch pages when a user asks about them, so a 404 there means an answer tried to use a page that is not there.
4. Sort by visits. Ahrefs suggests a threshold, such as more than ten visits a month, to focus on URLs worth acting on.
5. Ask assistants directly for links to your key pages, such as "link me to [company]'s pricing page", and test what they give you.

For each URL, note whether it once existed, which you can check in your redirect map, old sitemaps or an archive, or whether it was invented.

How to redirect URLs hallucinated by AI assistants

1. Redirect former real URLs first. Point each with a 301 to its replacement. Google's site move guidance recommends keeping redirects for at least a year and avoiding chains.
2. Redirect invented URLs that get steady traffic to the page that answers what the URL implies. /pricing/enterprise goes to the enterprise section of the pricing page. /docs/sso-setup goes to the single sign-on guide.
3. Consider creating the page when an invented URL gets steady traffic and you have no matching page. If assistants keep sending people to /pricing/nonprofit, buyers are asking about non-profit pricing, and a real page would answer them.
4. Use pattern rules sparingly. Normalising case, trailing slashes or a dropped date folder is reasonable. Broad wildcards that send any unknown URL somewhere are not.
5. Do not send everything to the homepage. Google's guidance says redirecting many old URLs to one irrelevant destination, such as the home page, can confuse users and might be treated as a soft 404. Visitors lose the context they came with.
6. Log each redirect with the date and reason, and review the list every quarter.

What should your 404 page do for AI visitors?

• Return a real 404 status. An error page served with a 200 status is a soft 404, and it can end up indexed as if it were content.
• Say plainly what happened: "The link you followed may be out of date or incorrect."
• Offer a search box, ideally filled in with words from the requested URL. /blog/how-to-choose-a-crm suggests a search for "choose CRM".
• Link to the pages AI visitors usually want: pricing, product overview, documentation, contact, and your most popular guides. Google's crawl error guidance suggests adding links to your most popular content and home page.
• Keep it fast and simple, since the visitor is already one step further from their answer than expected.

How do you reduce broken URLs at the source?

• Keep URLs stable. Avoid years, version numbers and campaign names in slugs for pages meant to last. Google's advice for seasonal pages is a recurring URL such as /sale/black-friday, not /sale/2020/black-friday.
• Maintain a redirect map through every migration, and test it before launch.
• Update third-party links that still point to old URLs: directories, partner pages, integration listings and press kits.
• Make the current pages easy to find, with accurate sitemaps and fast indexing, so search-based answers cite them rather than old copies.
• Name pages the way people ask. A page at /pricing with clear sections is easier to guess correctly than /plans-and-packages-2025.

Migrations can also cause a temporary dip in AI visibility even with redirects in place, as covered in why a website migration with 301 redirects can reduce AI visibility temporarily.

What might an illustrative case look like?

Consider an illustrative project management software company that tagged its 404 template and found about 400 AI-referred visits a month landing on error pages. Most went to three kinds of URL. There were old /features/ pages removed in a redesign, an invented /pricing/education URL, and blog posts whose slugs had lost their date folder. The company restored redirects for the removed feature pages and added a rule for the date folder. It also created an education pricing page, since assistants kept sending teachers there. Two months later, AI-referred 404 visits had fallen by about four fifths. The new education page was converting at about the same rate as the main pricing page.

What do people ask about AI answers and 404 pages?

Will redirecting invented URLs encourage assistants to invent more?

The redirect only catches visitors who were already sent there.

Do these 404s hurt our search rankings?

Google treats 404s as a normal part of the web. Pages that return 404 are not indexed, but they do not harm the ranking of your other pages.

Can we report wrong links to the assistant?

You can use the feedback buttons in answers, but that fixes nothing quickly. Redirects and a useful 404 page work immediately.

How often should we check?

Monthly at first, then quarterly, and after every migration or major URL change.

Should we return 410 instead of 404 for deleted pages?

Either is fine for pages gone for good. Both tell crawlers the page is not there. Prefer a redirect when a close replacement exists.