Here are two facts from the same week, about the same small site.
Fact one: the query "jumpscare link" brought lans.cloud 1,327 impressions and 410 clicks in 28 days, at position 4.4 — a 31% CTR, all landing on our jumpscare prank tool. It's the single biggest thing that has ever happened to the site.
Fact two: for most of that time, Bing's Keyword Research API — the only keyword tool that gives us real numbers instead of a subscription and a vibe — reported the 90-day search volume of "jumpscare link" as exact: 0.
Meanwhile, the same tool happily sized "roman numeral converter" at 1,090 exact (call it 11–22k/month on Google). We have a roman numeral converter. It sits at position 57–85 across ninety phrasings of that intent, with 350 impressions and zero clicks, ever.
The keyword tool and the winnable map aren't just uncorrelated. For a small site, they are close to inverses: the tool is most confident exactly where you will lose, and blind exactly where you can win. I want to argue that this is structural — not a Bing quirk — and then rescue the volume tool by describing the three jobs it's actually good at.
Why the sized terms are the losing terms
A keyword tool reports a volume when a query is searched the same way, by many people, over a long window. Stop and look at what that filter selects for: settled, recurring, canonicalized demand. "Roman numeral converter" has been phrased that way since before some of our users were born. Everyone types it the same way because the phrasing has had twenty years to converge.
But demand that has had twenty years to converge has also had twenty years to be answered. Those SERPs are owned — by exact-match domains, decade-old utility sites, or a Google widget that answers inline. The volume tool's confidence and the incumbent's moat are the same underlying variable: age. A term is measurable because it's mature, and it's unwinnable because it's mature.
So when you sort a keyword export by volume descending — which is what every tutorial tells you to do — you are, as a small site, literally sorting by "least likely to ever send me a click." We proved this to ourselves the expensive way: our roman numeral page is the site's third-biggest page by impressions and has never converted one of them. The tool said 1,090. The tool was right. It didn't matter.
Why your winners are invisible
The inversion has a second half: the queries a small site can win are systematically underreported, for three reasons that compound.
Fresh demand hasn't hit the tool's window yet. "Jumpscare link" is a viral, teen-driven intent that grew fast this summer. Search Console showed it to us live — hundreds of impressions a day, a third of them converting. The keyword tool's 90-day aggregate, thresholds and all, said 0. Then, about three weeks after GSC first showed the surge, the tool finally blinked: this week it reports "jumpscare link" at exact 82. The demand didn't change — the tool caught up. Volume data is a trailing indicator with weeks of lag; your own Search Console is a live one.
Long-tails are individually below threshold by definition. The intent behind our rigged coin flip arrives as "coin flip that always lands on heads", "coin that always lands on heads", "rigged coin toss", "heads or tails rigged" — dozens of phrasings, none of which clears the reporting threshold alone. The tool shows a row of zeros. GSC shows the family: ~40 phrasings, ~650 impressions, position 6–11. If you cluster by intent instead of by string, the demand is obvious; the tool can only see strings.
Some audiences aren't on the tool's index at all. Bing's user base skews older and desktop. Our anime-fandom queries — JoJo stand generators, Dragon Ball power levels — show 0 exact on Bing almost across the board while earning real Google impressions. We've now confirmed this five separate times across five fandoms. The tool isn't lying; its population just doesn't contain the searchers who matter to that page.
Put the two halves together and the practical conclusion is blunt: for a small site, keyword volume must never be a go/no-go signal in either direction. High volume is not a green light (it usually marks an incumbent's moat), and zero is not a red light (it usually marks demand the tool can't see). We wrote this into the project's doctrine as the volume-tool-inversion rule: when the tool and Search Console disagree, Search Console wins, every time.
So what is a volume tool actually for?
Three jobs. They're real, and we'd genuinely miss the tool if it vanished — just not for the job it advertises.
1. Learning the demand's own words
The tool's related-keyword expansion is a thesaurus written by searchers. We built a spelling practice tool and targeted "spelling test generator" — the tool sized it at 47. Buried in the related list: "spelling test maker", exact 317. Same intent, 6.7× the demand, one word different — and our spelling test practice page already was a maker; it just never said the word. Same story in crochet: every "yarn substitution calculator" phrasing we guessed was 0, and the related list surfaced "yarn substitution website" at 830. You don't use these numbers to decide whether to build. You use them to decide what words the page should say.
2. Sizing niches where the tool's population matches yours
The inversion is a statement about mismatched audiences, so when the audiences align, the tool works. Minecraft players are famously all over Bing, and there the tool is genuinely great: this week it sized "villager trading hall" at 1,672 exact, "best villager trades" at 402, and a whole per-profession family (armorer 645, toolsmith 567, fletcher 386…) that told us exactly which sub-intents our villager trading chart should deepen next. Know which of your clusters live on the tool's index, and trust it there.
3. Closing questions with a number instead of a debate
The most underrated job: killing ideas cheaply. We'd half-considered pitching our classroom display page as a "free digital signage" tool. The tool's verdict on "digital notice board": exact 1. One search. In ninety days. That idea is now closed with a number attached, and no one will reopen it next quarter with "but maybe there's demand." A zero from the tool can't green-light or red-light a winnable-SERP question — but it closes a does-anyone-phrase-it-this-way question permanently.
What decides, if volume doesn't?
The screen that actually predicts our outcomes is embarrassingly manual: look at the live SERP and ask who owns it. If page 1 is a recognized brand, a decade-old exact-match domain, or an inline answer widget, we lose regardless of volume. If it's a scatter of small sites, hobby pages, and PDFs, we can win regardless of what the tool says. Every ranking success this site has had — the prank cluster, the fandom calculators, the locker practice tool that Google started surfacing for "practice opening a locker" out of nowhere three weeks before back-to-school (Bing volume: 0, naturally) — came from queries the volume tool couldn't see. Every high-volume head term we shipped anyway sits below position 50 as a monument to the sorted-descending export.
The volume tool didn't earn a place in our build decisions. It earned a different place: phrasing thesaurus, niche-sizer where the audience matches, and the fastest way to end an argument with a number. That's three real jobs — just none of them the one on the label.