Every directory I’ve launched started the same way: a niche that looked promising for about ten minutes, followed by a few hours of research trying to talk myself out of it.
That second part is the important one. Most niche ideas die, or should die, before a single page gets built. The ones that survive aren’t the ones that sounded best on a Tuesday morning. They’re the ones that survive an attempt to kill them with real data.
This is the story of how I picked the niche for my newest directory, and the three other niches I researched, scored, and ultimately passed on to get there. Two of them had real, verified numbers behind them. One of them I’d already bought the domain for. None of that was enough on its own.
I’m not going to name any of the four niches, including the one I picked. In this space, ideas get copied fast, and handing a competitor a validated opportunity, especially one I haven’t launched yet, isn’t a trade I’m willing to make. Everything else here is real: the process, the numbers (rounded where the exact figure isn’t the point), the decisions, and the reasoning.
If you’re trying to figure out what to build, the process matters more than the destination anyway. So here’s the actual process, not the highlight reel, and not a sales pitch dressed up as one.
The trap: niches that sound good
“Directory business” ideas are easy to generate and hard to evaluate honestly. Almost anything sounds plausible if you only think about it for a few minutes:
- People search for it, right?
- There’s clearly a market.
- I don’t see an obvious big competitor.
Every one of those instincts has burned me before. The fix isn’t better instincts. It’s refusing to trust instincts at all until they’ve survived a scoring process that doesn’t care how excited I am.
I use a framework called KQF (Keyword Qualification Framework) for this, now on its second version. The short version: every keyword or niche idea moves through four phases:
Discover → Quick Qualification → Full Evaluation → Weighted Score
Quick Qualification is a five-second filter: business relevance, search intent, minimum volume, keyword difficulty. It’s meant to kill roughly 90% of ideas immediately.
Anything that survives goes into Full Evaluation, where I actually pull the SERP and look at who’s ranking, how many big-authority sites are sitting on page one, what the average referring-domain count looks like, and whether the intent is something a directory can realistically serve. That produces a weighted score out of 100, and the score, not my gut, makes the call:
| Score | Decision |
| 90–100 | Pursue Immediately |
| 75–89 | High Priority |
| 60–74 | Backlog |
| Below 60 | Skip |
Originally I ran this by hand, keyword by keyword. More recently I built an automated pipeline that pulls real search volume, difficulty, and SERP data, and scores an entire niche in one pass, which is what made it possible to actually re-run and compare candidates apples-to-apples, instead of relying on scores I’d calculated by hand months apart under different assumptions. That distinction turned out to matter a lot. More on that below.
Here’s how that process played out across the four niches I evaluated most recently.
The niche I picked
I ran keyword discovery across the specific sub-categories I was actually interested in, plus a general anchor term for the broader field. Close to a thousand keyword candidates came back.
The first thing that jumped out was volume. The broadest term alone was enormous by my usual standards (well into six figures a month), and nothing else in my pipeline was in that range.
But raw volume on a broad, generic term is usually a trap, and the SERP confirmed it. The bare, generic version of the core search term is owned by a handful of major general-information publishers and a couple of national institutions: not a fight a new directory wins. Scored on its own, that generic term came back in the high 60s: Backlog, not Pursue.
The pattern that actually mattered showed up one layer down.
The moment a search got local or provider-seeking (adding “near me,” or a city name, or pairing the core term with a word like “specialist” or “provider”), the SERP flipped almost completely. Instead of the big general-information sites, Google was showing individual local provider websites, often in a local 3-pack, even without the searcher typing “near me” explicitly.
That’s a genuinely different, much more winnable competitive layer sitting directly underneath an unwinnable one.
Scored properly, that layer looked like this, across five sub-categories within the field:
| Sub-category | Search pattern that scored | Volume/mo | Score | Decision |
| General / broadest | “[core term] near me” | Low six figures | Upper 80s | High Priority |
| Sub-category A | “[specific term] near me” | Mid five figures | Low 90s | Pursue Immediately |
| Sub-category B | “[specific term] near me” | Low five figures | Low 90s | Pursue Immediately |
| Sub-category C | “clinic/provider for [term]” | Mid four figures to low five figures | Low 80s | High Priority |
| Sub-category D | “[specific term] providers” | A few thousand | Mid 80s | High Priority |
| Sub-category E | “[specific term] near me” | Low five figures | Low 90s | Pursue Immediately |
| Sub-category F | “[city] + [specific term] provider” | A few thousand | High 80s | High Priority |
All five sub-categories scored High Priority or better, on a market an order of magnitude bigger than anything else on my bench. And critically, nobody had actually claimed the local layer.
The consumer brand most people would assume dominates this space never once showed up across roughly 150 organic results I pulled from sample SERPs.
The only recurring directory-shaped competitor was a locator run by the field’s own professional association, a bare-bones tool with no consumer enrichment. A couple of small existing niche directories were the extent of the competition, not a dominant aggregator: the same shape of gap that’s made my other directories winnable.
The economics backed it up too: sampled ad click costs several times higher than anywhere else in my portfolio, against real ticket sizes running well into five figures for a single transaction, a strong signal that advertisers in this space already pay to be seen.
None of that made it a clean, risk-free pick.
Two things are still genuinely unresolved: this is a higher-trust, higher-liability category than anything else I’ve built, and in some jurisdictions the rules around paying referral fees for this type of provider are stricter than they are for a typical home-services lead, so I’m treating flat advertising/featured-listing fees as the safe default until I’ve confirmed that with real research.
I also don’t have a clean matching domain the way I did with some of my other sites. The obvious naming pattern I use across my portfolio wasn’t cleanly available this time; every literal variant was already registered or parked at a premium price.
That last problem turned into its own small lesson. Rather than break my naming pattern or overpay for a premium domain, I chose a name that leans on the field’s own clinical or technical terminology instead of a generic descriptor.
That reads as more credible in a category where trust is the entire product, even though it’s a little less instantly literal than my usual naming. Sometimes the “obvious” domain isn’t available, and the fallback matters as much as the niche research.
I decided to build it. But that decision only makes sense next to the three I didn’t.
Rejected #1: the closest call
This is the closest of the three, and the most instructive, because it’s the same business model I was actually about to build: a local-provider directory, claimed listings, city pages, just in a different field. This niche had already been manually scored and slotted in as “next in the pipeline” before the niche I picked ever entered the picture.
The problem is that the original score for this niche was calculated by hand, months earlier, under a different process than the automated pipeline I’d since built.
Comparing an old hand-calculated score against a freshly-verified one isn’t a real comparison. It’s comparing a guess to a measurement. So before committing further build effort here, I re-ran it through the exact same pipeline I’d used on the niche I ultimately chose, specifically so the two candidates could be judged on equal footing.
The re-score held up in one place and fell apart in two others. The core “near me” pattern for this field is real. It scored comparably per-keyword to the niche I picked, and the SERP showed the same kind of small local-practice competitors.
But two of the four angles I’d originally planned to build around came back with literally no measurable search volume across dozens of keyword variants combined. And the remaining angles I tried scored Backlog or Skip, either dominated by the field’s own patient-education publishers and major health sites, or reading as general curiosity rather than provider-seeking intent.
Two things sealed it. First, scale: this niche’s best local term topped out under 500 searches a month; the niche I picked ran into the tens of thousands to low six figures on its equivalent terms, an order of magnitude or more larger.
Second, competition: this niche already has two or three real directory-shaped competitors actively ranking, including one run by the field’s own professional association and one multi-location provider group already running its own city pages, a meaningfully less greenfield local layer than what I found in the niche I picked.
Same business model, similar trust bar, similar per-keyword score on paper. Different market, different competitive reality once I actually looked. This niche isn’t dead. It’s back on the bench, waiting for its own dedicated slot, but it lost the seat it was already holding.
Rejected #2: the best score I’d ever verified
This one stings a little, because on paper it was the strongest keyword profile I’d ever measured. The core term for this niche pulled north of 20,000 searches a month at essentially zero keyword difficulty.
A handful of independent metro markets each cleared meaningful volume on their own. An unusually high share of the keywords I discovered for this topic passed my initial filter, well above the norm.
The SERP was full of small local businesses, not aggregators. The weighted score came out in the high 80s: at the time, the single highest-scoring candidate in my entire research archive, on verified data, not a guess.
I still didn’t build it, for a reason the score itself can’t capture: total addressable size. This is a single-service niche. There’s no multi-vertical category to expand into the way some of my other niches have natural sub-categories or variants that each carry their own meaningful search volume.
The head term is genuinely excellent. There just isn’t much beyond the head term and its city pages: a strong site, with a ceiling.
I only had one open slot in my active pipeline at the time (I run a hard cap on active directories until an existing one proves out revenue or a repeatable operating system, more on that rule below).
Given a straight choice between the single best per-keyword score in my archive and a comparably-scored opportunity with a materially larger and more multi-vertical market, I gave the slot to the bigger one.
This niche is still sitting on the bench as the strongest candidate there: first in line whenever the next slot opens.
Rejected #3: the one where I’d already bought the domain
This is the one that should have been the easiest yes, and it’s the one I killed outright.
I’d already purchased the domain and pitched it as a full content site (buying guides, product comparisons, troubleshooting) with a local installer/service-provider directory sitting underneath to carry monetization. On the surface, keyword difficulty looked open and the page-one authority-site count wasn’t alarming.
Full Evaluation told a different story once I looked past those two metrics to the thing that actually predicts a multi-year SEO climb: accumulated backlinks.
The core buying-guide terms I’d planned as the bulk of the site’s content were owned by a handful of established niche retailers, each sitting on thousands to tens of thousands of referring domains, plus one major national retail chain.
Difficulty score and authority-site count alone didn’t flag that, because those metrics don’t fully capture backlink depth. Only actually reading the SERP composition did.
The local installer/service-provider directory layer underneath, on its own, actually tested fine: real small local businesses on page one, a winnable competitive band.
But that layer alone would have meant a narrower directory-only business, not the content site I’d pitched and already bought the domain for. Rather than downsize the concept to fit the one piece that worked, I rejected the whole thing and let the domain sit.
The money was already spent. The research hours were already spent. Neither of those is a reason to build something the data says will take years to climb. A purchased domain is a sunk cost, not a commitment.
What actually decided each of these
Line the four up and the pattern is obvious in hindsight, which is exactly why it’s worth writing down before I forget it:
- Bare-term vs. near-me is the recurring split. Almost every niche I evaluate has an unwinnable informational layer sitting directly on top of a winnable local-provider layer. The score on the generic head term alone is close to meaningless; the score on the local variant is the one that matters.
- A high score only means something if it was measured the same way. An old hand-calculated score and a freshly-verified one aren’t on the same scale, and treating them as comparable would have been a real mistake. I don’t trust a comparison between niches until both sides have been run through the identical pipeline.
- Sunk cost isn’t evidence. A purchased domain, a good initial score, or hours already spent researching a niche says nothing about whether it’s still the right one to build today. My third rejection had two of the three and still didn’t survive.
- A great keyword profile isn’t the same question as total addressable size. My second rejection had the best single score I’d ever verified and still lost to a niche with a lower ceiling per-keyword but a much wider one overall.
- A hard cap on active projects forces the comparison to actually happen. I don’t add a new directory just because an idea looks interesting. A new site only gets built when an existing one proves out revenue, a repeatable operating system, or clearly fails and needs replacing. That rule is the only reason two of these ever had to compete for the same slot instead of both just getting built.
The niche I picked is moving into planning and build now, same operating model as the rest of my portfolio, same Find → Validate discipline that killed three other ideas to get here.
The niche is the first domino. Everything downstream (the architecture, the listings, the content, the launch) depends on getting this part right, because no amount of good execution rescues a niche that was never actually winnable.
That’s what the next post is about: what a directory actually needs to have in place before it’s ready to build, not just a good niche, but a real plan underneath it.
