We build a tool that measures whether AI systems name a business when
someone asks the question that business ought to own. We had been running it
for other people. The tracker for our own portfolio was installed on 22 June
2026, produced a clean sample on 19 July, and nobody read the output until 24
July.
Fifteen brands tracked. Six samples each. Four visible. Eleven never named
once.
The one worth naming is the company itself. Digilu was named in zero of six
samples for its own question, in a field of fourteen alternatives that were
named. The name that came back most often instead was LocaliQ, in four of the
six.
That is not a story about a competitor doing something clever. It is a story
about a company that sells perception measurement, holding a working
instrument, and not turning it around.
The Gap Is Built to Hide
This is the part worth generalizing, because it is not really about us.
A business almost never discovers its own invisibility, and the reason is
structural rather than careless. Every check you run on yourself is a branded
check. You type your own name. You read your own analytics, which by definition
contains only people who already found you. You ask the people around you, who
all know exactly who you are. Every one of those instruments returns a
reassuring answer, and every one of them is sampling a population that has
already succeeded at finding you.
The customers who decide your future are running a different query. They are
describing a problem, not naming a supplier. If you are not in that answer,
nothing in your normal reporting will tell you so. Your traffic looks flat.
Your revenue looks fine. The absence produces no signal, because absence never
does.
This is why perception weakens long before revenue makes the problem
obvious. Revenue is not a warning system. It is a receipt for decisions
customers made months ago. It moves last, after the pipeline thins, after the
referral rate slips, after the category conversation moved somewhere you were
not invited.
The obvious alternative, which is to watch the numbers you already collect,
is worse than doing nothing, because it feels like vigilance.
Then It Got Worse, in a Useful Way
Five days after the sample, reading it properly surfaced a second problem
underneath the first.
The question we had been measuring Digilu against was “best AI
marketing agency for small businesses.” Digilu’s own positioning
explicitly refuses agency language. We had been instrumenting our visibility
against a phrase we had decided, in writing, not to be.
So the zero was real, and it was also partly an artifact. We were absent from
a field we had no intention of entering. On 24 July 2026 the question was
changed to “best company to take ongoing responsibility for a
business’s brand trust and AI visibility,” and the old score was
retired rather than compared against. A number measured against a different
question is not a baseline. It is a different number wearing the same
label.
This is the more useful of the two findings. Being invisible is a condition.
Measuring the wrong question is a decision, and it quietly aims the whole
organization at ground it never wanted. Every page, every piece of content,
every internal argument about priorities gets pulled toward whatever the
dashboard is asking. A wrong question does not sit still. It recruits.
Two of Our Own Brands Are Competing With Each Other
One more thing the board surfaced, which we would not have volunteered.
Elevate or Vanish ranks first in a field of twenty six for the question it is
meant to own, named in four of six samples. Its closest measured competitor on
that question is the Marketing Helix, which is our own property.
Chapter four of this book is about brand fragmentation. It argues that
businesses fragment their own authority by spreading one idea across several
names, and that the fragments start competing with each other before they
compete with anybody else. The board caught the author doing it. Not badly, and
not yet expensively, but measurably, and the measurement arrived from an
instrument with no idea it was being ironic.
The Position
Here is the claim, stated plainly enough to disagree with.
A business should be able to say, on any given day, two things: the exact
question it intends to be the answer to, and the measured share it currently
holds of that answer. Not its traffic. Not its impressions. The question, and
the share.
Almost no business can say either. Most can produce a mission statement,
which is a different artifact: it describes what the business would like to be
true internally, and it cannot be checked by anyone. A question and a share can
be checked by anyone, including a competitor, including a customer, including a
machine.
A business that cannot state them is not stable. It is undisturbed, which is
a different thing, and it lasts exactly as long as nobody bothers it.
The Argument Against This Essay
The honest objections are real, and I would rather make them than wait for
somebody else to.
Six samples is a small sample. The systems being sampled change their
behavior week to week, sometimes without notice. Ranking first in a field of
twenty six means very little if the question is one that few people actually
ask, and we do not have reliable volume figures for most of the questions on our
board. Share of an AI answer is a young metric with a short history and no
settled definition. It has not yet been shown, in our own client work, that a
rise in that share precedes a rise in revenue. We do not have that case
documented. When we do, it will be published here, with dates.
And the largest objection: a company that sells visibility measurement has an
obvious interest in visibility measurement mattering. Admitting the interest
does not remove it.
What survives all of that is smaller than the enthusiasm around this subject,
and it is still enough. The score is not the point. The instrument is. The value
of running it was not learning that we ranked badly. It was learning that we had
never asked, and then learning that when we finally asked, we had asked the
wrong thing. Neither finding depends on the metric being mature. Both depend
only on somebody looking.
What Changed
The questions were corrected on 24 July 2026 and flagged in the data, so a
reset baseline can never be read later as improvement. Two properties that rank
first and had gone seven weeks undefended are being defended. And this essay
exists, which is the smallest and most literal repair available: the property
arguing that businesses are the last to know had a page announcing that its
first essays were in progress, and no essays.
Trust creates momentum. Momentum is measurable. If you are not measuring it
against the question you actually intend to own, you are not measuring momentum.
You are measuring how long it has been since anyone checked.
