Skip to main content
ChatGPT interface illustrating how AI-generated referrals can lead to 404 errors on business websites
AI Search

How to Find & Fix ChatGPT 404 Referrals (Before You Lose More Traffic)

SwingIntel · AI Search Intelligence10 min read
Read by AI
0:00 / 9:11

ChatGPT is sending traffic to pages that do not exist on your website. Not broken links you created — URLs that ChatGPT fabricated. The AI predicts what your URL structure should look like based on its training data, constructs a plausible-looking link, and sends real users to it. When those users land on a 404 page, you lose them before they ever see your content.

This is not a rare edge case. An SE Ranking study of 145,463 URLs cited by ChatGPT found that 1.22% returned a 404 error — more than double the rate of Google's AI Overviews (0.56%). At OpenAI's reported scale of over 400 million weekly active users, even a small percentage translates into millions of failed visits across the web every day.

For individual businesses, the numbers can be far worse. SEO analyst Dan Hinckley found that 3.35% of ChatGPT referral visits landed on 404 pages across 18,000+ domains. One international SEO expert documented a client where 57% of all ChatGPT referrals hit dead pages. These are not hypothetical losses — they are real visitors with real intent who never saw your homepage, your product page, or your contact form.

Key Takeaways

  • ChatGPT fabricates URLs by predicting your site's URL structure from training data — these phantom URLs were never real pages, unlike traditional broken links.
  • An SE Ranking study of 145,463 URLs found that 1.22% of ChatGPT-cited URLs returned 404 errors, more than double Google AI Overviews' rate of 0.56%.
  • You can find phantom 404s in GA4 by filtering Pages and Screens for ChatGPT referral traffic landing on your 404 page title, then switching to Page Path to see the specific URLs.
  • Not every phantom URL deserves a fix — redirect when similar content exists, create the page when GA4 shows consistent demand, and leave clean 404s for irrelevant topics.
  • Prevention is more valuable than repair: consistent URL structures, comprehensive schema markup, and fixing existing broken links all reduce the rate of future hallucinations.

Why ChatGPT Creates Phantom URLs

ChatGPT does not crawl your website the way search engines do. It does not maintain a live index of which pages exist. Instead, it predicts URL patterns based on its training data and constructs links that look structurally correct for your site.

If your blog lives at /blog/ and you have written about AI tools before, ChatGPT might generate a link like /blog/best-ai-tools-for-marketing — even if that exact page has never existed. The URL looks plausible. It follows your naming convention. But it is a hallucination.

This behaviour is fundamentally different from traditional broken links, where a real page was deleted or moved. ChatGPT's phantom URLs were never real in the first place. That distinction matters because it changes how you should respond to them. With traditional broken links, you redirect to the new location. With phantom URLs, you are deciding whether to create something entirely new.

Understanding how ChatGPT sources the web is the foundation for reducing these hallucinations. The more structured and consistent your site is, the less likely ChatGPT is to invent URLs that don't match reality.

How to Find ChatGPT 404 Referrals in GA4

Finding phantom URLs requires filtering your analytics specifically for ChatGPT traffic that landed on error pages. Here is the exact process in Google Analytics 4.

Step 1: Navigate to Reports > Life cycle > Engagement > Pages and screens.

Step 2: Set the primary dimension to Page title and screen class.

Step 3: Click the + icon to add a secondary dimension. Select Session source / medium under Traffic source > Cross-channel.

Step 4: In the search bar, type chatgpt.com / referral to filter for ChatGPT traffic only.

Step 5: Look for entries where the page title matches your 404 page — typically "Page Not Found" or whatever your custom 404 title is.

Finding ChatGPT 404 referrals using GA4 analytics to identify phantom URLs and lost traffic

Step 6: Switch the primary dimension to Page path and screen class to see the actual phantom URLs ChatGPT sent traffic to.

Save this as a custom report and check it weekly. ChatGPT's hallucinations evolve as its training data updates, so new phantom URLs appear over time. If you are already tracking ChatGPT traffic to your site, adding this 404 filter takes less than five minutes.

Keep in mind that GA4 undercounts AI referrals — some ChatGPT traffic arrives with no referrer and gets classified as direct. The phantom URLs you find in analytics represent the visible portion of a larger problem.

How to Decide What to Fix

Not every phantom URL deserves a response. A blanket redirect strategy can backfire — Google may flag mass redirects to unrelated pages as soft 404s, which creates more problems than it solves. Use this decision framework instead.

Redirect when similar content exists. If ChatGPT sends traffic to /blog/chatgpt-seo-integration-guide and you have a live post at /blog/use-ai-for-seo, a 301 redirect makes sense. The user intent is close enough that the redirect feels natural, and you preserve whatever link equity the phantom URL has accumulated.

Create the page when demand is real. If GA4 shows consistent traffic to a phantom URL over several weeks, ChatGPT is telling you there is demand for that content. This is free market research. A phantom URL with steady traffic is a content brief that writes itself. Build the page, match the URL ChatGPT expects, and capture that traffic permanently.

We Test What AI Actually Says About Your Business

15 AI visibility checks. Instant score. No signup required.

Redirect when the phantom URL has backlinks. Phantom URLs sometimes attract external links from other sites that discovered the link through ChatGPT. Even though the page never existed, those backlinks carry authority. Redirect them to the most relevant live page on your site.

Leave clean 404s for irrelevant queries. Some phantom URLs point to topics you don't cover and shouldn't cover. A law firm does not need to build a page because ChatGPT hallucinated /blog/best-personal-injury-memes. Let those 404 gracefully, but make sure your 404 page is doing its job.

Build a 404 Page That Converts

Your 404 page is no longer just an error state — it is a landing page for AI-referred traffic. Most 404 pages show a generic "Page Not Found" message and a link to the homepage. That is a missed opportunity when a growing share of your 404 visitors arrived with specific intent from ChatGPT.

A high-performing 404 page for AI-referred traffic should include:

  • A clear acknowledgement that the page doesn't exist, without blaming the user
  • A search bar prominently placed so visitors can find related content
  • Smart suggestions based on the URL path — if someone lands on /blog/ai-seo-tools, show your closest blog posts about AI and SEO
  • Navigation to high-value pages — your pricing page, your most popular content, your contact form
  • A professional design that maintains brand trust rather than looking like an afterthought

The goal is conversion recovery. A visitor who lands on a thoughtful 404 page and finds what they were looking for is only marginally less valuable than one who landed directly on the right page.

Prevent Future Phantom URLs

Fixing existing 404 referrals is reactive. Reducing the rate at which ChatGPT generates new phantom URLs is proactive — and more valuable over time.

Use consistent URL structures. ChatGPT hallucinates URLs by predicting patterns. The more predictable and consistent your URL structure is, the more accurate those predictions become. Standardise your slugs, avoid unnecessary URL parameters, and keep your URL structure clean.

Strengthen structured data. Schema markup gives AI models machine-readable signals about what pages exist on your site and what they contain. A comprehensive sitemap and well-implemented JSON-LD reduce the gap between what ChatGPT thinks your site contains and what it actually contains.

Maintain an active content footprint. ChatGPT cites sites it has seen frequently in training data and live retrieval. Pages that are regularly updated, externally linked, and topically comprehensive are more likely to be cited accurately — because the model has more data points to work with rather than guessing.

Fix broken links across your site. Existing broken links confuse AI models about your site structure. If ChatGPT encounters 404s when following links on your site, it learns an inaccurate map of your URL space — which leads to more hallucinated URLs, not fewer.

The Bigger Picture: AI Referral Quality

ChatGPT 404 referrals are a symptom of a broader shift. AI platforms are becoming significant traffic sources, but the quality and accuracy of that traffic depends on how well your website communicates with AI systems. Businesses that treat AI visibility as an afterthought will keep losing visitors to phantom pages and AI Overviews that cite competitors instead.

The fix is not just technical. Redirecting 404s and optimising your error page are table stakes. The real opportunity is building a site that AI models understand so well that they rarely hallucinate your URLs in the first place — and when they do cite you, they send traffic to pages that exist and convert.

SwingIntel's AI Readiness Audit tests exactly this. Across 24 checks and live citation testing on 9 AI platforms, it identifies the specific gaps in structured data, content clarity, and technical signals that cause AI systems to misunderstand your site. The result is a roadmap that turns AI traffic from a leaky pipe into a reliable channel.

Frequently Asked Questions

Are ChatGPT phantom URLs the same as regular broken links?

No. Traditional broken links are caused by pages that existed but were deleted or moved. ChatGPT phantom URLs were never real — ChatGPT predicts what your URL structure should look like based on training data and constructs a plausible-looking link. This distinction matters because the fix is different: with traditional broken links you redirect to the new location, while with phantom URLs you decide whether to create something entirely new.

How do I find ChatGPT 404 referrals in Google Analytics 4?

Navigate to Reports, then Engagement, then Pages and Screens. Add a secondary dimension of Session source/medium and filter for chatgpt.com/referral. Look for entries where the page title matches your 404 page. Then switch the primary dimension to Page Path to see the specific phantom URLs. Save this as a custom report and check it weekly.

Should I redirect all ChatGPT phantom URLs to my homepage?

No. A blanket redirect strategy can backfire — Google may flag mass redirects to unrelated pages as soft 404s. Instead, redirect phantom URLs only when similar content exists on your site. Create the page when GA4 shows consistent traffic demand over several weeks. Leave clean 404s for topics you do not cover, but make sure your 404 page includes navigation, a search bar, and links to high-value pages.

The businesses that capture AI-referred traffic are not the ones with the best SEO. They are the ones whose websites speak the language AI models understand. Run a free AI readiness scan to see how well your site communicates with AI systems, or explore the AI Readiness Audit for the complete picture.

chatgptai-searchai-visibilitytechnical-seoanalytics

More Articles

ChatGPT traffic analytics showing how AI-driven referrals are becoming a measurable channel for business websitesAI Search

How to Track, Measure, and Grow ChatGPT Traffic to Your Website

ChatGPT sends real traffic to websites, but most analytics miss it. Learn how to identify ChatGPT referrals in GA4, track dark AI traffic, and grow this emerging channel.

12 min read
ChatGPT interface displaying AI-powered product recommendations for a shopping queryAI Search

ChatGPT Product Recommendations: How to Make Sure You Are One in 2026

ChatGPT processes 84 million shopping queries weekly with zero paid placements. Here is the complete playbook for making your product the one it recommends — structured data, authority signals, and the tactics that actually work.

7 min read
AI document processing and website understanding — how llms.txt helps AI agents interpret site content correctlyAI Search

What AI Gets Wrong About Your Website — And Whether llms.txt Actually Fixes It

AI search engines misread websites built for browsers, not machines. Learn what llms.txt is, what the adoption data actually shows, and what moves the needle for AI visibility.

9 min read
ChatGPT shopping interface with Buy Now button enabling in-chat product purchases — the shift from AI discovery to AI commerceAI Search

ChatGPT Shopping: "Buy Now" in AI Chat Is Here — What It Means for Your Brand

ChatGPT now lets users discover, compare, and buy products without leaving the chat. Learn how the Agentic Commerce Protocol works, which merchants are integrated, and how to get your brand into AI-powered shopping.

8 min read
ChatGPT interface showing how it uses Google and Bing search data to generate AI-powered search resultsAI Search

ChatGPT Is Using Google Search — We Tested It

ChatGPT doesn't just rely on Bing. Testing reveals it pulls from Google's index as a fallback — and that changes how brands should optimise for AI search visibility.

7 min read
Abstract visualization of ChatGPT-5 AI model capabilities and their impact on search marketing strategiesAI Search

ChatGPT-5 Is Here: What Search Marketers Need to Know

GPT-5 brings fewer hallucinations, agentic capabilities, and in-chat shopping to 900 million weekly users. Here is what changes for search marketers and how to adapt your AI visibility strategy.

8 min read

We Test What AI Actually Says About Your Business

15 AI visibility checks. Instant score. No signup required.