Build log
Three things we got wrong, and what changed because of it
A measurement company should be willing to show you a measurement that
went wrong. These are real, they are ours, and each one changed how the tool works.
Nobody else's business is named.
We searched the wrong town for a month
September 2026
A mechanical contractor sat on our list for a month marked no Google Business
Profile found. We had checked four different ways. All four came back clean and
all four found nothing, so the conclusion looked well evidenced — and the opening
line we had drafted for that call was going to be "you don't have a profile at
all."
They had a profile. Six reviews, 3.7 stars. All four of our checks had searched
the wrong town — the business is one town over from where our list said, and
every check inherited that one bad input.
The rule we had was a search that finds nothing is evidence; a search that fails
is not. It needed its other half: a clean search is only evidence about the
thing you actually asked. Four confirmations of one wrong premise is one
observation, not four — and it is more convincing for having been checked
four ways, which is what makes it dangerous.
What changed: a business town now comes from the address on its listing, never
from the page we found it on. And "no profile found" is the one finding the tool
will not put in an email. It goes to a person first.
Two companies, one name, two hundred miles apart
August–September 2026
Two roofing companies in this state share a name. Our first check read one of them
and reported a review count. It was the wrong one — the two are about two hundred
miles apart, and nothing in the result said so.
Quoting a business its own review count and being wrong is the fastest way to end
a conversation, because it is the one error the owner can spot instantly. So the
matcher was rewritten to refuse: if two results could plausibly be the
business, it reports that it could not tell, and a person decides. It would rather
return nothing than return something plausible.
We found two more holes in it since. One matched a listing whose name was
missing a word from the one we searched — and the missing word was the whole
difference between two real companies. The other matched a business in a different
state entirely, because the name agreed and only one result came back. Both are
closed now.
The pattern in all three: a single confident wrong answer is harder to catch
than an obviously ambiguous one, so the checks have to be about more than
whether the tool looked sure.
Why your report might say "Unverified" instead of a number
August 2026
We audited a business with around 350 reviews and a good average. The report scored
its reviews zero out of thirty — the largest single category — and told it to
treat reviews as needing attention.
Nothing was wrong with the business. Google had simply not served us the panel that
carries the review count that day. It does that sometimes, for no reason the
business can influence. Our scoring could not tell the difference between
a business with no reviews and a number we were not shown, so it
treated the second as the first.
That is the worst thing a measurement can do: invent a zero. We changed the
rubric so an unreadable number scores neutrally and the report says, in plain words,
that we could not read it — and shows you the exact search we ran so you can check.
The same audit went from 50 to 64 once it stopped being punished for our blind spot.
A profile we genuinely could not find still scores zero, because that is a real
finding. The distinction between "we looked and it is not there" and "we could
not see" is the whole job, and you will find it printed throughout your report.
These are the short versions. If you are weighing up a bigger piece of
work and want the long one — what broke, what it cost, what we changed — ask. The
reasoning is written down as it happens, which is the only reason these posts can be
specific.