Most content roadmaps are vibes with a spreadsheet skin. You list what you feel your audience needs, ship it, and six months later Search Console quietly informs you that nobody searched for half of it. When I finished building Ajiez — a collection of 101 free browser tools — I had 37 tutorials to plan and exactly zero interest in guessing again. So I pulled real search data first and let it overrule me. It did, repeatedly. Here's what the numbers said, and the small-site workflow you can copy.
Bing Webmaster Tools exposes an API endpoint that returns weekly impression counts for any keyword, by country and language. It's the quiet cousin of Google's keyword tools: free, boring, and — crucially for a small site — good enough to rank directions even when it can't promise volumes. The usual caveat applies: Bing serves roughly a quarter to a third of Google's search volume in my niches, so I treat every number as a compass reading, not a forecast.
I drafted 108 candidate keywords across my tool categories — finance, PDF, images, dev utilities, text — and queried them all for the US market, English, exact match. Every request succeeded. Seventy-one keywords came back with data. Thirty-seven came back with nothing.
That "nothing" is worth pausing on, because it's the most misunderstood output in keyword research. An empty response doesn't mean zero people search for the term. It means Bing has no measurable ad inventory for it — no impressions recorded. For a tiny site, that's still actionable: these are words where even Bing's demand signal is too thin to plan around. Park them; don't build on them.
The guide I almost didn't write. My assumption was that JSON tooling was my biggest content gap — developers break JSON constantly. The data said otherwise: "diff checker" showed 2,841 weekly exact-match impressions, nearly four times "json validator" at 790. Comparing two versions of a text is a far more universal act than debugging a config file: contract redlines, essay revisions, pull-request descriptions. So the first guide out of the gate was how to diff two texts, pointing at a text compare tool that runs entirely in the browser. The assumption I started with would have spent my best slot on the fourth-best topic.
Same field, different wedge. "Json formatter" pulls 3,946 weekly impressions — and an incumbent site has owned that head term for a decade. Fighting it head-on is a content strategy for sites with domain authority I don't have. But "json validator" at 790 is a different intent: not "make this pretty" but "find what's broken," ideally with a line number. That intent shapes a different page — a "why your JSON breaks and how to read the error" guide that pairs with a formatter and validator that flags the failing line. Volume told me the field was alive; the split between two nearly identical keywords told me where the opening was.
Aggregating families beats staring at single words. "Amortization schedule" alone showed 1,873 weekly impressions. But four related phrasings — the schedule itself, "with extra payments," "extra payment calculator" variants — added another ~230. Families matter because one good guide captures the whole cluster, and the family total told me this was the largest finance content opportunity in my entire dataset. A mortgage amortization guide with worked extra-payment examples went near the top of the queue.
Tool queries are content opportunities too. "Character count" pulls 1,047 weekly impressions, locked up by counters that answer the mechanical question and nothing else. But the interesting question behind that query is platform rules: why does any link on X cost exactly 23 characters, why does a Chinese SMS split at 70 instead of 160, why does Google truncate your 58-character title anyway? That's a cheat sheet, not a tool feature — so I wrote one, next to a plain word counter for the mechanical job. The character limits cheat sheet targets the question the head-term sites were ignoring.
Here's my favorite result, and it was a negative one. My site's whole differentiation is that files never upload — everything runs client-side. Naturally, I queried "no upload," "without uploading," "offline tools" and eight similar phrasings. Eleven seed keywords, zero data. All of them.
The conclusion isn't that privacy doesn't matter. It's that nobody searches for it. People search for the task ("compress PDF"), then choose between results — and "your file never leaves this tab" is what wins them at the moment of choice. That reframed privacy from a keyword strategy (which would have wasted a page on it) into a conversion message that belongs in title tails and trust bars across every page I already have. Arguably the most valuable thing 108 API calls bought me was the death of one bad page idea.
The workflow, distilled List candidate keywords from your own features first — every tool you've built implies five to ten queries it could answer. Pull exact-match weekly impressions for all of them in one batch. Any free source beats none; Bing's API is unglamorous and sufficient. Sort by family, not by word. Aggregate phrasings that one page could satisfy. Read the splits inside a field. Formatter-vs-validator, schedule-vs-extra-payments — near-duplicate words with different intents are where small sites find openings the incumbents left. Honor the empty responses. Zero data means don't build; log it and move on. Separate data from inference. My volume numbers are measurements; my competition calls are me reading result pages with my eyes. I label them differently in my notes, and this post does too.
My tutorial roadmap is now chosen, sequenced, and justified by numbers I can re-run any afternoon. Some of it will still flop — data shrinks the odds of a bad bet; it doesn't erase them. But "I wrote this because 2,841 people a week ask this question and the current answers don't run locally" is a far better reason than a hunch.
If you want to see where the workflow landed, the diff guide and the rest of the stack are live on Ajiez — 101 tools, everything client-side, no accounts. And if you've got your own small-site keyword workflow, I'd genuinely like to hear what your empty responses taught you.