Someone asks an AI assistant whether regular blood tests show drugs, and it sends them to a URL that doesn’t exist. It isn’t one of the usual broken links on your site. It’s a link the AI made up, one that sounds exactly right but leads nowhere.
This happens constantly. Ahrefs studied 16 million URLs and found that AI assistants send visitors to 404 pages 2.87 times more often than Google Search does. ChatGPT was the worst offender.
Why This Matters Now
AI referral traffic is still small next to Google, but it’s some of the highest-intent traffic a treatment center gets. The person clicking a link inside an AI answer has already asked a specific question and chosen your site as the place to get the answer.
When that click lands on a 404, you lose three things at once:
- The inquiry. Someone researching drug testing or detox timelines is often close to picking up the phone. A dead end sends them to a competitor.
- The demand you already earned. The AI recommended you. That’s the hard part, and a broken link throws it away.
- Trust. The user frustration lands on your brand, not on the AI tool that invented the link.
This isn’t a huge traffic channel. But it’s easy to fix, it keeps growing, and it’s worth turning into a simple routine. This guide shows you how.
Why AI Assistants Invent URLs
AI assistants like ChatGPT, Claude, Gemini, Perplexity, and Copilot run on large language models (LLMs). When they write out a link, they often generate it instead of copying it, predicting what a URL should look like from your page title and patterns seen on thousands of other sites.
It’s the same thing behind other AI hallucinations. Language models are built to produce text that sounds right, and a URL is just more text. The AI doesn’t check whether the link works, so users end up on 404 errors. Plausible isn’t the same as correct.
A Treatment Center Example
Here’s what this looked like on one treatment center’s site.
What People Asked
The site’s most popular articles answer questions like whether regular blood tests show drugs, whether magic mushrooms go bad, and what happens if you add water to a urine test. These are exactly the questions people ask AI assistants, so these pages get cited constantly.
What URLs the AI Invented
The real page /do-regular-blood-tests-show-drugs/ picked up more than 30 phantom URLs:
- Typos: /do-regular-blood-tests-show-drygs/ and /do-regular-blood-testsshow-drugs/
- Swapped words: /do-regular-blood-files-show-drugs/ and /do-regular-blood-tests-show-data/
- Rewritten slugs: /do-regular-blood-tests-show-drugs-understanding-what-doctors-see, which reads like a headline, not a URL
- Invented folders: /blog/do-regular-blood-tests-show-drugs and /en/mushrooms/, on a site with neither folder
- Pure noise: a version number like /0.6.10/, and another site’s URL pattern, post ID and .html extension included, grafted onto the domain
Which Ones Signaled Real Intent
The typos, swapped words, rewritten slugs, and invented folders all pointed clearly at one real article. Those were people trying to read about drug testing, often while worried about an upcoming test or a loved one’s use. The version numbers and foreign URL patterns matched nothing and got no real clicks.
Why the Same Pages Keep Getting Hit
Hallucinated URLs cluster around your best-known pages. The more often a page gets cited in AI answers, the more chances a model has to reconstruct its URL slightly wrong.
That repetition tells you two things. There’s real demand for the topic, and the AI is uncertain about where it lives on your site. Both make those clusters the first place to act. New variants of the same handful of pages keep surfacing week after week, so this is rarely a one-time glitch.
Why This Hurts Treatment Centers More Than Most Sites
Most industries lose a bounce when someone hits a 404. Treatment centers lose a moment of readiness. People asking AI assistants about drug testing, detox timelines, or medication interactions are checking a fact before making a decision, and that’s often when someone decides to call, fill out a form, or keep looking elsewhere. For a treatment center, those calls and forms are the conversions that matter.
Each lost visit is a broken journey that repeats, one or two people at a time, across dozens of URLs. It also muddies your data, since referral traffic that lands on a 404 never shows up as engagement on the page it was meant for.
There’s an SEO angle too. Hallucinated links don’t stay inside chat windows. They get copied into scraped articles, forum posts, and AI-written blogs, and those backlinks point at a dead page.
A 5-Step Process to Reclaim the Traffic (No Script Required)
You don’t need a developer or custom code. This process uses tools most SEO teams already have, and it repeats cleanly every month.
Step 1: Find Hallucinated URLs in Search Console Data
Start in Screaming Frog. Connect the Google Search Console API and turn on “Crawl New URLs Discovered in Google Search Console.”
This pulls in URLs that have appeared in Google search results, including ones never linked anywhere on your site. Run the crawl and filter for URLs with no internal links. Those are your phantom URLs, along with the status code each one returns today.
Step 2: Find Them in Your Analytics
Search Console only shows URLs that appeared in Google. Your web analytics shows the ones people actually clicked from AI tools.
In Google Analytics (GA4), filter landing pages by AI referrers (a simple regex like chatgpt|perplexity|gemini|copilot|claude works) and look for pages that returned a 404. Not every AI click passes a referrer, so treat this as a partial view, but it catches fake URLs that never showed up in Google.
Step 3: Monitor for New Ones
New variants keep appearing as AI assistants generate new answers, so you need an alert, not just an audit. A Search Console monitoring tool like SEOTesting can notify you when new URLs on your domain start picking up impressions. You’ll catch fresh hallucinations within days instead of at your next crawl.
Step 4: Triage and Act
Every phantom URL falls into one of three buckets:
| If the URL… | Then… |
|---|---|
| Clearly matches a real page | Set up a 301 redirect to that page |
| Points to a topic you don’t cover, and keeps recurring | Consider creating the page, then redirect to it |
| Matches nothing, like an old image path or random string | Leave it as a 404 |
Then work in priority order:
- High-intent, high-repeat clusters. Variants of your most-visited pages, especially ones that recur.
- URLs with evidence of real visitors. Anything with clicks, AI referrals in GA4, or backlinks.
- Content-gap patterns. Repeated URLs for topics you don’t have a page for yet.
- Noise. One-off, low-frequency hallucinations that match nothing. Leave these alone.
Don’t over-fix it. Every redirect is one more rule to maintain, and chasing every fake path is wasted time. Redirect what has value and let the rest 404. A 404 is the correct response for a page that never existed.
Two notes. WordPress will sometimes redirect a near-miss slug to the closest matching post on its own, which handles simple typos but not rewritten slugs or invented folders. And you may come across developer tools that catch every 404 and suggest the closest page automatically. They can help on very large sites, but for most treatment center websites, a short list of manual redirects does the same job with more control.
Step 5: Mine What’s Left for Content Research
Don’t throw the list away. The fake URLs AI tools invent show you what they expect your site to have. Group them by topic and look for patterns:
- Content gaps. If AI tools keep sending people to URLs like /insurance/aetna/ or /detox-timeline/ and no matching page exists, that’s a question people are asking about your facility that your site doesn’t answer.
- URL structures. If AI keeps inventing a /blog/ or /programs/ folder you don’t use, it’s expecting a structure common on similar sites. That’s a hint about how your information architecture compares, not necessarily a reason to rebuild it.
- Weak internal linking. If variants cluster around one page, the AI knows the topic belongs to you. Make sure that page is easy to find, linked from related pages, and listed in your sitemap.
How to Measure Results
Is this worth the effort? A few simple numbers will tell you:
- Recovered sessions: AI referral traffic landing on the pages you redirected to
- Conversions: calls and form fills from those sessions
- 404s from AI referrers: this number should drop after each round of fixes
- New phantom URLs per month: a rising count usually means more AI citations, which is a good sign
- Top clusters: the pages generating the most variants, which are your most AI-cited content
Set realistic expectations. Most hallucinated URLs pick up only one or two hits each. But a single popular page can spawn dozens of variants, and every one of those hits was a real person with a real question. Recovering even a handful a month is worth the time it takes to run the process.
Hallucinated URLs are one symptom of a bigger question: what are AI tools actually saying about your facility? Our guide to AI visibility audits for rehab centers covers how to check.
Frequently Asked Questions
Should I redirect every hallucinated URL?
No. Redirect the ones that clearly map to a real page, since those are usually people trying to reach your content. Leave junk URLs, like old image paths or random strings, as 404s. They’re typically bots, and there’s nothing to recover.
Should I use a 301 redirect or let it 404?
Use a 301 when a matching page exists. It sends the visitor where they meant to go and passes along any link value. A 404 is the right answer when nothing on your site matches.
Do I need a developer tool to handle this automatically?
For most treatment center sites, no. Automated tools can help on very large sites, but a few dozen manual redirects give you the same result with more control.
How often should I check for new hallucinated URLs?
Set up alerts for new URLs picking up impressions so you hear about them as they appear, then run a full crawl monthly or quarterly to catch anything the alerts missed.
Can I see AI traffic to 404 pages in GA4?
Yes. Filter sessions by AI referrers such as chatgpt.com or perplexity.ai, then look at landing pages that returned a 404. Not every AI click passes a referrer, so treat it as a partial view.
Will hallucinated URLs hurt my rankings?
Not directly. Google doesn’t penalize you for URLs someone else invented. The cost is lost visitors, lost link value, and a worse experience for people trying to reach you, whether they came from ChatGPT, an AI Overview, or another site.
The Takeaway
Hallucinated URLs aren’t going away, and as more people use AI search to research treatment, the volume will grow. The fix is a routine, not a project:
- Redirect the hallucinations that match a real page
- Create content for the gaps that keep recurring
- Ignore the noise
Run it monthly, and those phantom clicks turn back into real visits.