Every number this site published and later changed.
A census is only worth citing if it says when it was wrong. This page lists every figure we published and then corrected. What caused it, how far it moved, and what now stops it happening again. 7 corrections so far. The list is short because the site is young, not because the numbers are safe.
18 September 2026: the guide Bath and windows, side by side
We called the oldest company average the oldest ad. The row said oldest live ad, but the number was the highest median age of any one company's ads. The oldest single ad is older.
| Figure | Published | Corrected |
|---|---|---|
| oldest live bath ad | 712 days | 847 days |
| oldest live window ad | 2,508 days | 2,663 days |
The medians in the same table were right: 19 days for bath companies, 37 for window companies.
The row now reads the census file's own oldest ad, not the company table.
18 September 2026: the home page, the bath and window pages, and the flow calculator
We showed a cabinet franchise's cost per appointment as the bath number. The $399 to $1,370 band came from Art of Drawers' 2026 filing. Art of Drawers sells cabinets. The pages labelled it for bath remodeling and the calculator ranked your cost against it. No bath or window filing publishes a cost per appointment.
| Figure | Published | Corrected |
|---|---|---|
| cost per appointment, bath | $399 to $1,370, filed | no bath filing has one |
| what the home page shows instead | Re-Bath's filed spread: 45.6, 9.8 and 2.7 sold jobs per 100 leads |
Re-Bath's own filing stands: $176 a lead on average across 122 locations.
A filed number now shows only for the trade that filed it, and the window pages show none.
18 September 2026: two guides, 14 and 4 September 2026
We counted "Book now" buttons and called them pages that hold a time. A button is what the ad says. Whether the page behind it can take a booking is a separate read. When our software opened 180 bath pages on 15 September, 3 had a calendar.
| Figure | Published | Corrected |
|---|---|---|
| bath companies that let a homeowner book (14 Sep) | 11 of 216 | 11 have a Book now button; 3 of 180 pages had a calendar |
| bath pages that hold a time (4 Sep) | 7 of 118 | 7 had a Book now button; pages not opened |
The button counts themselves were right.
A button and a calendar are now two different lines everywhere they appear.
18 September 2026: the guides What lasts in bath ads and Bath and windows side by side
We said three in four bath ads were still running two weeks later. Our census page said one in four for the same dates, and when we recounted from the raw rows nothing reproduced three in four. The guide also split that group by price, so that section is gone too.
| Figure | Published | Corrected |
|---|---|---|
| bath ads live on 4 Sep still live ten days later | 1,529 of 2,045, 75% | 808 of 3,199, 25%, on the 122 pages we track in both weeks |
| priced ads against unpriced, still running | 68% against 78% | withdrawn; it rested on the same group |
Across every bath ad we saw from any source, 2,139 of 5,092 were still running, 42%.
Survival is now one query on one named group of pages, published with its base.
18 September 2026: the guides The page your ad sends them to and Ten minutes in your own account, and the home page
We said we opened every bath landing page in America, on a phone, and printed load times. Our software opened the page behind each company we track, once, set up like a phone. That is not every page in America. Its timer also included a 2.5 second pause the software adds, so every load time was too slow by the same amount.
| Figure | Published | Corrected |
|---|---|---|
| pages opened | every bath landing page in America | 180, the companies we track |
| bath pages that can book | 3 of 177, and 4 of 178 in an older run | 3 of 180; 8 we could not read |
| median load time | 6.4 seconds | withdrawn |
44 companies still send the click back to Facebook, and 160 pages promise a free estimate.
No load time is published until the timer stops counting its own pause.
5 September 2026: the window ad census, edition 2026-W36
We counted five advertisers outside the United States as part of the trade. Meta's ad library filters by the country an ad reached, not by where the advertiser is. Five window companies advertise into the United States from Canada. The review that verified them read the ad text for the trade and never for the country.
| Figure | Published | Corrected |
|---|---|---|
| companies | 81 | 76 |
| pages | 85 | 80 |
| live creatives | 3,376 | 3,227 |
| promise a free consultation | 39 of 81 | 35 of 76 |
| non-franchise median creatives per company | 6.5 across 46 | 6 across 41 |
The median age of a live creative stayed at 25 days. Nothing in the bath edition moved.
Every run now re-reads every verified page against the country and trade rules. That check has to pass before we publish an edition.
29 September 2026: the bath ad census, edition 2026-W38
We counted one company as promising a free consultation when it offers nothing free. The count looks for the word "free" near consultation, estimate, quote or design. Bath Hound's ads say "barrier-free walk-in shower design" and offer 50% off. The hyphen let "barrier-free" match. A reading of every company's matching words found it.
| Figure | Published | Corrected |
|---|---|---|
| promise a free consultation | 155 of 216 | 154 of 216 |
| share of bath companies | 72% | 71% |
| pages carrying it | 158 | 157 |
| live creatives carrying it | 1,225 | 1,224 |
Every other bath figure and the whole window edition stayed the same.
The count now skips "barrier-free". Other words joined to "free" by a hyphen, such as "hassle-free design", are being read before they are counted or dropped.
What counts as a correction
A number that was live on this site and then changed. A number we caught before publishing it is not here, because nobody read it.
A correction rewrites the edition it belongs to. The page and the data file then carry one number for that week rather than two. That is why this page exists. It is the only place the original figure stands. Without it, a correction would be a silent edit. The old figures are in the table above, in full.
What is not a correction
- A number that moved because the record grew. The roster gains pages with every count. So you cannot compare a trade's totals between editions, and each page says so. That is the measurement working.
- A method that changed. The edition says what it counts now. The method page says how, so anyone can count it their own way in Meta's public Ad Library.
If you find a number here that is wrong, say so and it joins this page.