DEV Community

Cover image for ChatGPT Stopped Recommending This App Overnight. The Culprit Was a Domain Move (and One Regex)

ChatGPT Stopped Recommending This App Overnight. The Culprit Was a Domain Move (and One Regex)

ArshTechPro on November 14, 2026

For weeks, ChatGPT was quietly acting as an iOS app's best salesperson, sending roughly a third of all its downloads. Then, almost overnight, it we...
Collapse
 
rushdeploy profile image
Rishabh Mishra •

The regex catch is the part I would steal for my own checklist: test the old sitemap URLs from git history, not just the homepage. A single curl loop showing 301 to the exact new URL would have caught the .html pages being proxied as 200s before Bing did.

Collapse
 
arshtechpro profile image
ArshTechPro •

Its always better to check the old sitemap URLs from git history

Collapse
 
rushdeploy profile image
Rishabh Mishra •

Agreed — and I would keep that list in the repo, so the same curl check runs on every deploy. A small change to the asset rule would then show up as a failed 301 instead of a quiet 200.

Thread Thread
 
arshtechpro profile image
ArshTechPro •

that’s correct we can know the changes if have it for every deploy.

Collapse
 
tomerbarm profile image
Tomer Bar-Meir (Tom) •

The detail that stands out is that nothing in the app changed, only the address that assistants had learned to cite. For small sites it is a good reminder to keep a redirect map and to recheck which URLs an assistant actually links before and after any move. Did you see how long it took ChatGPT to start citing the new domain once the redirects were fixed?

Collapse
 
arshtechpro profile image
ArshTechPro •

Keeping a redirect map would help

Collapse
 
paullovelace profile image
Paul Lovelace •

This one hit close to home — the regex bug is brutal. .html pages proxied with a 200 while their canonicals point at the new host is exactly the kind of silent split that takes weeks to notice. The Bing angle is the part most people miss: Google forgiving a move in weeks while Bing treats the new host as untrusted with no history. Your point about verifying the old URLs actually 301 (not 200) is the one I'd tattoo on every migration runbook.

Collapse
 
arshtechpro profile image
ArshTechPro •

the regex was indeed brutal for the downside

Collapse
 
arshtechpro profile image
ArshTechPro •

Curious whether others have seen this. Has anyone tracked ChatGPT or Perplexity referrals before and after a domain or URL change? The before/after was surprisingly sharp here, and it'd be good to know if Bing re-trust usually takes weeks or months.

Collapse
 
posting-dude profile image
Posting Dude •

The .html matching the asset regex is such a clean trap: the old host kept serving 200 copies whose canonicals disowned themselves. Did anything in Bing's tools ever point at that, or did it only show up once you curled the legacy paths from the old sitemap? I'd have trusted that passing Live URL test too.

Collapse
 
arshtechpro profile image
ArshTechPro •

The Bing tool never point to that error

Collapse
 
linkbruno profile image
linkbruno.com •

That isAsset regex is such a believable bug. Anyone reviewing it reads "anything with a file extension" and pictures CSS and images, not the .html URLs that carried all the search history. I like that the test loop runs over the old sitemap from git instead of a hand-picked list. I'd keep that script and rerun it for a few weeks after the move, because a later tweak to the Worker can quietly turn some of those 301s back into 200s. Did Bing's URL Inspection change at all once the proxied copies were gone?

Collapse
 
arshtechpro profile image
ArshTechPro •

Still waiting on the Bing URL Inspection to show green

Collapse
 
yonto profile image
Yonto Player •

Really useful post. Worth adding for anyone reading: ChatGPT's search leans heavily on Bing's index, so after a move, verify the new host in Bing Webmaster Tools too (not only Google Search Console), submit the sitemap there, and ping IndexNow on every deploy. Bing tends to be slower to pick up a move on its own than Google.

Collapse
 
arshtechpro profile image
ArshTechPro •

That's correct. We should definitely cross check for our products

Collapse
 
kursatyz profile image
Kürşat Yılmaz •

The obsession with wrapping remote APIs into multi-layered agent abstractions is creating massive hidden latency and failure cascades in production.

If you profile real-world workloads, bare-metal local inference (via direct C++ / llama.cpp or vLLM bindings) consistently beats fragile cloud wrappers on both predictability and cost. Tooling should eliminate abstraction penalties, not compound them.

Collapse
 
arshtechpro profile image
ArshTechPro •

I will try this local inference and see the performance. Thanks for the tip.

Collapse
 
botsailorofficial profile image
BotSailor •

This is such a great reminder that sometimes the smallest technical details can have the biggest real-world consequences. What I find particularly interesting is how a seemingly harmless domain migration and a tiny regex oversight ended up affecting an entire app's visibility and downloads. The lesson goes beyond SEO — it's about understanding how interconnected our systems really are. Also, appreciate you sharing the actual debugging process rather than just the outcome. Real-world engineering lessons like this are incredibly valuable for developers and indie builders. Great write-up! 👏

Collapse
 
arshtechpro profile image
ArshTechPro •

Thanks for the great words