Fireshift / 449-site study / Corrections
Corrections
When a number we published turns out wrong, we fix the page and log it here: what we said, what was wrong, what's right, and where we changed it. We don't delete the old figure without a note. Three entries so far, all from the first week of scans.
Login conversions: 11.5% used a loose match
What we said. The 449-site study (4 Nov) counted a login conversion on 11.5% of 235 sites with Google Ads conversions.
What was wrong. The login rule matched "sign in" and "log in" inside other words, such as design-inspiration or
blog-insights in a URL. That was fixed in the scanner on 4 Nov.
What's right. Our 5 Nov rescan of 206 different B2B SaaS sites, reading each trigger with the fixed rule, found 7.3% count a login or a page load inside the logged-in app. It is a different sample, so treat the two as separate measurements, not a recount.
What we changed. The login and trial guide (5 Nov) uses 7.3% and explains the gap. The study page carries a correction note (5 Nov).
Duplicate conversions: 63.8% was too high; 30.0% on exact trigger conditions
What we said. Our 449-site study (scanned 4 Nov 2026) reported that 63.8% of the 235 sites with Google Ads conversions have duplicates: the same label on 2+ tags, or one trigger firing 2+ labels.
What was wrong. That count grouped triggers by a short text summary. The summary drops regex URLs and extra conditions, so two different triggers could look the same and count as one trigger firing two labels.
What's right. We rescanned 203 B2B SaaS sites on 4 Nov 2026 and compared the exact compiled trigger conditions, within the same Ads account. 61 sites (30.0%) have the same label on 2+ tags or one trigger firing 2+ labels into one account. The old loose method gives 66.5% on the same 203 sites, so the gap is the method, not the sample.
What we changed. The duplicates guide (4 Nov) and the duplicates check page (5 Nov) use 30.0% and explain the change. The study page carries a correction note next to the old figure (5 Nov). The free scan and the scanner switched to exact trigger conditions on 5 Nov (see the method page).
A video event read as a trial ending
What happened. The scanner's trial-end rule matched any event name ending in "ended". On 4 Nov 2026 it flagged a site
for a conversion "when a trial ends". The event was a GA4 key event named video_view_ended: someone finished a video.
Where it reached. A prospect note on our side. It was caught the same day, before any email went out, and it was never on a public page.
What we changed. On 4 Nov the rule was narrowed: it now needs the word "trial" next to end, ended, expire or over, so
video_view_ended and call_ended no longer match. We added a regression test for it, and rescored every saved
finding that mentioned a trial ending; the flagged site's finding was replaced by what the fixed scanner found. In our 5 Nov
rescan of 206 sites, none counts a trial ending.
Report an error
Email hello@fireshift.ai with the page and the number. If you scanned your own container and see something different from what we reported, send the container ID and we will check it against our rules. How we get numbers in the first place: method and AI use.