Method
How I measure the pages.
Every ranking number on this site comes from two runs of the same check, three months apart. This page covers what each run did, what counts as a match, what the data can't tell you, and the times the measurement fooled me.
The check
Every page I built targets one search, for example "drywall repair after leak Castaic" or "sablage plancher Blainville". I send each of those searches to Google through the DataForSEO SERP API and read the results that come back. The question is the same for every page: did Google return the page built for this search, and at what position?
The two runs
The pages were built in May 2026. I ran the first check on August 9, 2026 and the second on November 7, 2026, with the same 1,004 searches.
| August 9, 2026 | November 7, 2026 | |
|---|---|---|
| Mode | Live mode | Standard queue |
| Depth | Top 100 | Top 30 (the first three pages) |
| Device | Desktop, not personalised | Desktop |
| Location | Country level: US for the US sites, Canada for Quebec | Country level, same as August |
| Results read | Organic only, no map pack | Organic only, no map pack |
| Client pages kept per search | On the drywall site, only the first one | Every one |
| Searches that returned data | 991 of 1,004 | 1,004 of 1,004 |
The whole November run cost under $2, which is why checking every page every month is realistic.
Why November used the standard queue
In late 2026 Google started wrapping the links in its results. The live mode returned those wrapped links instead of the real addresses, so my first November test matched nothing at all. It looked like every page had vanished. The standard queue returns the real addresses, so the full run used that. Other differences are in the table above.
What counts as a match
A page counts as matched when the page built for that search is the client result Google returned. On the drywall site in November, the town page that replaced a merged page also counts, because that's the page a visitor lands on.
I compare the last part of the URL, the slug, and not the full address. Clients move pages around, and a folder change shouldn't read as a ranking loss.
Why August undercounts
On the drywall site, the August data kept only the first page from the client's site in each search. When a sibling page came first, I can't tell whether the intended page was also there further down. So the August matched and top 3 figures are floors. The November run records every page from the client's site in each search, which closes that gap.
What the numbers don't say
- The median position (#2 in August, #2 in November) covers matched pages only. It says nothing about the searches where the page didn't show up.
- Each run is one snapshot. Positions move every day.
- These are positions, not traffic, calls or revenue. I don't have that data for these sites.
- No map pack. For many local searches the map results sit above the organic ones.
- Without Search Console access, I can't see why Google dropped a page. The cause stays a likely one.
Limits of this data
- On the Quebec build, 14 searches per language have a broken phrase where the town slug leaked into it, for example "sablage plancher lassomption L'Assomption". They're kept in the counts and marked on the client page.
- 13 of the 1,004 August checks failed and returned no data. They're left out of the August "searches checked".
- August was read 100 results deep and November 30. Every count uses the top 30 for both dates. Of the 308 pages found in August, one sat below #30 (at #39). The page tables still show it, but it isn't counted as matched.
- Search volume, competition and page age weren't part of these two runs, so reasons given for a service line's result are likely causes.
Four measurement traps, and how I caught them
Each time, the problem was the measurement and not the pages.
- WordPress renamed pages. When a slug is already taken, WordPress adds "-2" to the end. My records still had the original slug, so the check looked for a page that didn't exist.
- The client moved the pages. Halfway through the drywall project, every page went into a region folder, so
/drywall-repair-castaic/became/santa-clarita-valley/drywall-repair-castaic/. Matching on the full address gave a false zero for 501 searches. Comparing slugs fixed it. - A database that can't see small towns. DataForSEO's Labs database can't return small-town searches that have no ad volume, even when the page is #1 for them. My first pull through it returned 1,097 rows and not one zero-volume keyword. The SERP API fetches the live results for the search you give it, which is the question I'm asking.
- Wrapped links. The November false zero described above.
The rule I work by: a perfect zero is almost always a matching bug, so I check which address actually came back before I believe a result. On the septic site, the November zero held up under that check. Google returned other pages from that site with their real addresses, and the page I spot-checked was no longer indexed.
What your client gets
The same check runs in the audit and in every monthly report: each page against the search it was built for, with the date and the conditions written on it.
Send me one client's website.
I'll check which of its location pages Google still shows and send you a short summary.