Every keyword tool ranks candidates by volume, difficulty and cost per click, then lets fit be one factor among them. That single design choice is why so much content ranks and sells nothing.
A brand sells dog shampoo. A keyword tool suggests gentle clean shampoo: strong volume, real cost per click, moderate difficulty. It passes every filter in every tool on the market and scores well in all of them.
It is worth nothing. The person typing it is buying shampoo for themselves.
Almost every keyword tool computes a score. Volume contributes, difficulty contributes, cost per click contributes, and relevance contributes. Then the list is sorted.
The moment fit is one term among several, volume can outvote it. Not always, not obviously, but reliably at the edges, which is exactly where a large content programme lives. A term with ten times the volume and slightly worse fit will outrank a term with perfect fit, and it will do so while displaying a respectable number next to it.
The failure is silent, and that is the expensive part. You do not get an error. You get five hundred green generations, five hundred published pieces, traffic that goes up and to the right, and no buyers. Every dashboard says the programme is working. The only number that disagrees is revenue, and it disagrees two quarters later.
Treat fit as a gate evaluated before any ordering exists.
A candidate that fails fit never enters the ranking, at any strength of any other signal. It is not a weighted term. It is not a tiebreak. It does not contribute to a score, because a score is precisely the structure that lets a big enough number elsewhere buy its way past.
This is a smaller change than it sounds and a much larger one in effect. The output of a scored system is "here are your keywords, ranked". The output of a gated system is "here are the four that qualify, and here are the three hundred and thirty six we refused, each with the reason".
The second output is less satisfying and considerably more useful.
A gate is only as good as its inputs, which creates a rule that matters more than the gate itself: every unmeasured input has to make a candidate less likely to pass, never more.
If we cannot read the results page for a term, we do not know whether its searchers are buyers. The tempting behaviour is to let it through and let the writer judge. The correct behaviour is to refuse it, because a gate that waves through what it cannot see is not a gate, it is a formality.
That produces a specific and slightly uncomfortable outcome. When the data source degrades, the system delivers fewer keywords rather than lower quality ones, and it says so. A month where the gate returns four instead of twelve is a month where it worked.
Three practical consequences.
Under-delivery is a feature and has to be treated as one. If your engine or your agency promises twelve pieces a month and the qualified set is four, the right answer is four pieces and an explanation. Publishing the other eight is volume with no buyer behind it, and it costs you twice: once to produce, and again in a corpus that dilutes the pages that do qualify.
Refusals are the most informative output you get. Three hundred rejections sorted by reason tell you something no ranked list can: whether your problem is the market, the product category, or the way you described who you sell to. A cluster of "searcher is not your buyer" against terms you were sure about is a positioning finding, not a keyword finding.
Ask any tool what it refused. If it cannot tell you, it is scoring rather than gating, and somewhere in your last five hundred pieces is a gentle clean shampoo.
We hold ourselves to this and it bites. Running our own selection against our own domain this week, the gate read a live results page, produced thirteen candidates, and passed none of them, because the data source we were reading from could not surface the commercial signals the gate requires. Thirteen refusals, one reason, zero recommendations.
The correct response to that is to fix the measurement, not to lower the gate. A system that recommends something because it could not check is worse than one that recommends nothing, because the first kind is trusted.
Persistent brand memory. Self-correcting generation. Built for DTC operators who can't afford generic copy.
Book a discovery call →Three ways to run marketing at a 5 to 50 person company, priced against what each one actually delivers in a month, and the question that decides between them.
Nine out of ten business domains sit behind a mail provider that refuses to confirm whether a mailbox exists. Here is what the verification industry is measuring instead, and what the badge on your list means.