We built an SEO scanner. The honest result: no traffic gain
It is a good defect net. It is not a traffic lever, and we measured both.
We built a scanner that checks our own sites every night for SEO, accessibility and content defects. When we asked what it gained us in search traffic, the honest answer was nothing we can measure. We wrote that down so nobody redoes a flattering version.
The question
"What did the scanner gain us?" We took the framing most generous to the scanner: the day it was first usable, 8 February 2026, with eight months of search data before and seven after.
The numbers
Search clicks averaged about 556 a month before and about 521 after. Impressions were up slightly and click rate was down. Average position improved a lot, from about 33 before to about 17 after.
That would flatter the scanner, except for the step test. Position by era was about 44 in summer 2025, about 22 by the winter, and about 17 after. Most of the gain came four to five months before the scanner existed, and our own independent traffic estimate was already rising before it. No metric shows a jump at the boundary.
The scanner is also read-only. It changes nothing itself, so any effect has to travel through a person acting on a finding.
What it is good for
A nightly defect net across our own domains. Its checks cover sitemaps, redirect chains and loops, links, contrast and our machine-readable site files, and they run without anybody having to remember to run them. That is the value: defects that would otherwise sit silently.
The scanner's own honesty problem
More than half of the last 120 commits to the scanner are fixes. Its commonest historical bug was reporting "could not measure" as "the site is broken". We now require every check to report what it could not see, not only what it found, and a check that cannot answer must be distinguishable from a real negative.
It is also blind to five things: absence (a venue that should send traffic and sends none), third-party profiles, claims stored in the database, render-layer breakage, and how a page looks.
What we say and do not say
- We say: it runs nightly across our own domains unattended and checks things we would otherwise only find by accident.
- We do not say: that it raised traffic, rankings or any coverage figure. Our own data does not show it.
- What to ask any SEO tool vendor: show me a before and after with a step test, and tell me what the tool cannot see.
Want a tester who tells you what a tool did not prove?
We report what we could not measure as clearly as what we found. The first month is a trial.
