What we ran
- Pages: 259 landing domains from ads we collected from Meta’s public Ad Library on 2026-11-11. We dropped app-store and social links and drew 150 at random. One URL per domain.
- Views: Website cloaking checker, 5 requests per page: desktop Chrome (the baseline), iPhone Safari, Googlebot, the Facebook crawler, and Chrome arriving from a Google ad click.
- Where: our test box, plain fetch, one home IP. Not from a Worker. 150 pages, about 750 requests, 53 seconds.
What we found
| result | pages |
|---|---|
| Chrome refused or failed (19 were HTTP 429, we hit them 5 times at once) | 33 |
| Chrome answered, but another view was refused | 6 |
| Same page in all 5 views | 75 |
| A different page in at least one view | 36 |
| of those: a script reads the ad click ID or referrer, content identical | 23 |
| of those: content differs | 13, one of them noise (same size, random tokens) |
The 12 real differences, by pattern (a page can fit two rows):
| pattern | pages | what it is |
|---|---|---|
| Googlebot or Facebook gets a full page, Chrome gets an empty script shell | 3 | pre-rendering for crawlers. Google calls it a workaround, not cloaking, while the content is similar |
| iPhone is sent to another URL | 4 | app store links, smart links, a mobile site |
| Facebook crawler or Googlebot is sent to another URL | 3 | tracking templates, 2 with an unfilled {{campaign.name}} placeholder in the URL |
| iPhone text or layout differs only | 2 | mobile layout |
A quotable base rate: about 1 in 9 ad landing pages looks different to at least one view. A difference is a lead. Read what changed before you call it cloaking, as in cloaking in advertising.
How you’d use it
- Ad network or reviewer: run every new landing URL the same 5 ways, and queue only pages where a crawler or reviewer view gets a different destination or a different offer. See how ad networks detect cloaking.
- Brand: run the pages of ads that use your name. A reviewer-clean, visitor-dirty split is evidence for a report.
- Regulator: a dated table per target, with the views and the reasons, is what a file needs.
We do not run JavaScript, and we see one address. Pages that cloak by IP range or country need other vantage points.