I pasted 20 links into ChatGPT and Gemini. Their search crawlers followed none of them

What I expected

I thought there might be a shortcut. When ChatGPT opens a page because a user asked it to, OpenAI now knows that URL. So OAI-SearchBot, the company's own search crawler, might come by afterwards. Same for Gemini and Googlebot. If that held, a few people pasting your link into an app would be enough to put a page on the crawler's map.

I wrote the scenario and the pass/fail thresholds down and froze them before the first run, so I could not move the goalposts after seeing the numbers.

The morning run

I created new paths on pw.migurei.ai. No page linked to them, and they were not in the sitemap. Between about 10:20 and 10:50 Tokyo time, the lab machine ran 24 sessions: 10 paths for ChatGPT, 10 for Gemini and 4 for DeepSeek as a control. A few more paths were never pasted anywhere, also as controls.

Every app really fetched its page. The log has all 24 requests, each carrying the marker of its session.

Then I waited three days.

Nobody came

The window closed on 16 October. ChatGPT pulled OAI-SearchBot to 0 of 10 paths. Gemini pulled Googlebot to 0 of 10. DeepSeek, 0 of 4. The control paths stayed clean. No verified crawler from any company requested any of those paths, apart from ChatGPT's own fetches during the sessions.

The crawlers were not asleep. In the same 72 hours Googlebot sent 110 requests to the site, across 20 URLs. Bingbot sent 41. They simply never went near the new paths.

OAI-SearchBot sent exactly four requests, all for /robots.txt, about 16 hours apart, the same rhythm as before the test. The closest one came 11 hours 33 minutes after the last ChatGPT fetch. It read robots.txt and left.

What this means for you

When a customer pastes your link into ChatGPT or Gemini, the app can read your page and use it in that answer. That does not put the page in front of the search crawler. For that you still need the usual routes:

1. Link to new pages from pages the crawlers already fetch.
2. Put them in your sitemap.
3. For Google, use Search Console.

Do not count on user pastes to get you discovered.

What this does not prove

The claim is refuted by a lab measurement, but the sample is small.

• Zero out of 10 only rules out a follow-up rate above roughly 26% per app. A lower rate is still possible.
• One account per app, one machine, 30 minutes on one morning, one host, the same questions. The 10 sessions are not independent of each other.
• It holds for pw only: a new subdomain where OAI-SearchBot only ever reads robots.txt, and the pasted URLs carried an extra parameter. Crawlers may judge a whole host, and a site they already crawl could behave differently.
• Claude was not measured.
• From 15 October our own tools fetched those paths again, so the later window up to 20 October cannot be read cleanly. The 72-hour result is not affected.
• These are origin logs. A crawler blocked at Cloudflare's edge would not show up.
• A crawler fetching a page would not mean the page gets indexed anyway. Here, no crawler fetched it.

The full data is on the pw research pages.