Most search markets should never become websites.

Take Home Office Chairs. Amazon US lists 4,722 products in the category, and the listing activity behind them spans about nine years. A 100-product sample from SellerSprite’s 2026.07 report recorded roughly $306,941 in monthly sales and a 56.32% average gross margin — sample figures, not category totals, and gross margin, not net profit. Thirty percent of the sampled products were listed within the last six months. A market this active should make a great website. It doesn’t.

Open the broad chair queries — “home office chair,” “best home office chair” — and the results include several well-documented testing publishers. BTOD has personally tested more than 200 chairs since 2018. Wirecutter logged more than 175 hours of group testing. WIRED has tested 70 chairs, sitting in each for at least two weeks. The demand is real. With the evidence available today, we have not identified a defensible, rankable improvement angle.

The classic operator mistake is reading the demand and stopping there. Search volume looks like a go signal. It isn’t. Before you open a code editor, ask four questions — in order. Is the demand real and persistent? Can the SERP be materially improved? What evidence can you publish that nobody else can? Will the traffic compound into something you own?

Not all good markets make good websites. Some are IMPROVE decisions — real demand, wrong angle. The framework below is how you tell the difference before you spend six to eighteen months of your attention.

Search volume is not an opportunity.

Search volume tells you how many people type a query. It says nothing about whether you can earn their click — or what you’d do with it if you did.

A deeper question follows: if search volume does not decide value, what does? A keyword is worth what its buyers pay, not what its searchers click.

Keyword difficulty is equally silent. It estimates ranking competition; it does not determine whether entering the market makes commercial sense. Our positioning data shows the gap. The decision and method queries we anchor on — “how to evaluate a website niche,” “should I build a website for a niche” — carry difficulty between 12% and 21%, with forums and blogs on page one and no dominant incumbents. The tool-style query “keyword opportunity analysis” carries 82% difficulty, with a first page of almost all tool vendors.

Same job, opposite realities.

Why high-volume keywords trap operators

High search volume often means someone already built the pages, earned the links, and answered the query — often with testing depth a solo operator can’t match in a weekend. Small sites rarely enter a mature market through its head terms; they enter through fragments of the tail and work up.

What a real opportunity looks like

A real opportunity isn’t a big number. It’s a first page you can look at and name exactly what’s missing: a question the current results answer badly, a type of evidence none of them publish, an audience none of them serve.

The chair market draws real traffic — between 1,600 and 8,100 US monthly searches across four related queries, recorded from Semrush’s US database on 2026-08-06 — and its page one includes substantial, well-documented testing competition. If search volume isn’t the answer, what is?

Four questions before you build.

The framework is the Search Opportunity Framework: four sequential questions — Demand, SERP, Evidence, Asset. You answer them in order, and you don’t skip.

Think of them as decision gates. You can’t jump to question three before answering question one. Each gate has one job. Question 1 asks whether the demand is real and persistent. Question 2 asks whether the current first page can be materially improved. Question 3 asks what evidence you can publish that nobody else can. Question 4 asks whether the traffic can compound into an asset you own.

Order matters. Question 1 screens the wasted-year risk, Question 2 the wasted-content risk, Question 3 the copyable-edge risk, Question 4 the rented-traffic risk. Skip one and you’re guessing where the framework is trying to make you look.

This is not a scoring formula. There is no multiplication, no weighting, no opportunity score at the end. Markets don’t get better because you averaged four numbers; they get better because one of the four conditions is actually true.

If a question fails, the answer isn’t automatically stop. It’s stop — or change the angle and re-ask. IMPROVE is a real outcome of this framework, not a consolation prize.

Candidate market
  →  Q1: Is demand real and persistent?          NO → STOP
     YES ↓
  →  Q2: Can the SERP be materially improved?     NO → STOP, or narrow the angle
     YES ↓
  →  Q3: Can you publish evidence others can't?   NO → STOP, or build evidence first
     YES ↓
  →  Q4: Will the traffic compound into an asset? NO → STOP, or run short-term
     YES ↓
     GO — build, then re-evaluate on a schedule

We’ll run a real market through all four questions in a moment. First, question one — the one most people skip.

Question 1: Is the demand real and persistent?

Signals that lie

Search volume lies first. A spike can reflect a news cycle, a viral post, or a seasonal blip — none of which is a market. Demand that turns on a trend turns off with it.

Tool screenshots lie second. A keyword report shows how many times people typed a phrase; it doesn’t show whether anyone is paying to satisfy that need, or whether anyone will be typing it in three years. Treat a volume number as a starting hypothesis, never a verdict.

Your own belief lies third. The classic operator error is wanting a market to be real because you already like it. “There should be demand for this” is a feeling, not a signal.

Signals that matter

Real demand has three properties you can check.

It has a source you can name. Platform sales data, recurring community discussion, and replacement activity show that people reach for this need repeatedly, not once.

It persists across a meaningful time span. A market active for years, with new entrants still arriving, is different from one that peaked last quarter.

And — one signal worth checking — someone is already paying to serve it. Persistent paid participation can be one demand signal, but it does not by itself prove profitability or durable demand.

In the chair market, the demand check was straightforward. Amazon US lists 4,722 products in the category; 30% of the sampled products were listed within six months; the category has been continuously listed for nine years. Demand passed — clearly, and it wasn’t close.

But even real demand doesn’t mean you should build. The next question asks whether you can actually be seen.

Question 2: Can the SERP be materially improved?

Not sure how to read a SERP beyond the KD score? Read the first page like a market map — it tells you what Google is already rewarding, not just what is hard to rank for.

This is a market diagnosis, not a chair buying guide. We’re here to show you how to read a SERP — not which chair to buy.

What the SERP rewards today

Forget the difficulty number for a moment. Read the first page as a record of what the market already rewards. Ask three things: which page types hold the rankings, how strong is the evidence behind them, and are the domains contestable?

Here’s what page one rewards for the four chair queries, recorded from Semrush’s US database on 2026-08-06:

For the record, Semrush records “best home office chair” at a keyword difficulty of 57%, meaning roughly 33 referring domains are typically needed to compete. A useful number — but KD estimates ranking competition; it doesn’t decide whether the market makes sense.

We verified the testing claims by opening the pages. BTOD’s founder states he has personally tested more than 200 chairs since 2018. Wirecutter reports more than 175 collective hours of group testing. WIRED says it tested 70 chairs, sitting in each for at least two weeks. Architectural Digest uses its editorial team; Consumer Reports runs a three-person panel.

The price band is covered, not open. WIRED has tested chairs at $150, at $120, and below $80 on promotion — not just premium models. The “nobody reviews cheap chairs” gap is not supported by the pages.

When the SERP says no

For the broad chair queries, the read is clear: the queries include several well-documented testing publishers. With the evidence available today, we have not identified a defensible, rankable improvement angle. The incumbents are professional, well-documented, and cover the price spectrum. A new page would enter against organizations with years of testing behind them, without a gap we can name.

There’s a second problem, subtler than competition: intent. Chair queries carry commercial intent — people searching to buy. The Search Opportunity Framework serves a different intent: deciding whether to build. The two don’t match. That’s an intent mismatch, not a content gap. No amount of optimization makes a chair-buying query serve a market-evaluation article, and we won’t pretend otherwise.

A SERP with no identified improvement angle doesn’t mean the market is worthless — it means you need a different angle. But before you hunt for that angle, ask the next question: can you publish something nobody else can?

Question 3: What can you publish that nobody else can?

Not every observation qualifies as evidence. How to build publishable evidence — the three criteria and four layers that separate what you can publish from what you should not.

Evidence is not experience

We have hands-on operating experience in furniture. Experience, though, is not evidence. To turn this market into publishable content we would need order structure, return rates, cost structures, material comparisons, and de-identified customer feedback. None of that data has been assembled. As of this writing, the operational data to convert founder experience into published evidence has not been assembled.

So the honest self-check runs like this: we thought we had an evidence advantage. We don’t — yet. What exists is a notebook, not a page. That’s the difference the framework is built to force out into the open.

An evidence advantage is not a resume. It is something a reader can hold: a data table, a documented test, a comparison no one else has run. If a competitor could reproduce your edge with a weekend of research, it was never an edge.

What publishable evidence looks like

Publishable evidence has three properties. It is original — you generated it, not reworded it. It is verifiable — you show the method and the numbers, with dates. It is defensible — a competitor can’t fake it in an afternoon.

In the chair market, the evidence question does not pass today. The demand is real. The competition is documented. The evidence is still in a notebook. That combination — not any single number — is what makes this an IMPROVE decision rather than a GO.

We keep our standards public so you can check us. The where the evidence comes from page explains which claims are verified, which are sourced, and which are judgment.

Evidence is what turns a good SERP read into a real asset. That brings us to the final question.

Question 4: Can the traffic compound?

The framework is not a formula for always finding a market. It is a filter for stopping you from building in the wrong one. We evaluated two real markets through all four questions — and both failed. The failures are the proof.

The framework was built for a search landscape that is already changing. Three structural pressures are reshaping search — and Amazon evolution happens to illustrate all three.

The first three questions decide whether you can earn attention. The fourth decides whether that attention becomes something you own.

What compounding looks like for a content site

For an informational site, the loop is simple. A ranking page earns a reader. A reader becomes a return visitor. A return visitor becomes a subscriber, a data point, a piece of brand trust. Every page you publish after that starts from a small bank of people who already know you, instead of from zero. The loop is why the fourth question exists.

If that step never happens, the visitor is rented. A ranking position is not an asset — you hold it for as long as the position holds, and when the ranking moves, the visitor goes with it. That’s not necessarily bad; it’s simply not compounding. Compounding is the difference between a channel you’re building and a position you’re borrowing.

For the chair market, the asset question is conditional — and it separates the two kinds of asset this framework produces. The generic framework assets — the four-question worksheet and the decision record you see below — are publishable now; they don’t depend on chair-category evidence. The chair-category assets — a sizing guide, a price-band analysis, an Amazon-data view of the category — do. Each feeds the others and the next worked example, but the chair-specific pieces can’t be built credibly until the chair evidence is assembled. The evidence question doesn’t pass today; it stands open, waiting on evidence.

Four questions, asked in order. Let’s see what happens when you run them all on a real market.

Running the four questions: Home Office Chairs

Here’s the worksheet, filled in — the whole framework applied to one real market. Treat it as a live demonstration, not a retrospective: these are the notes you’d take at the moment of evaluation.

Demand: Passed.

The demand question passed clearly. Amazon US lists 4,722 products in the category; nine years of continuous listing; 30% of the sampled products listed within six months. Real, persistent, and still admitting new entrants. This sets the baseline: it’s a market worth evaluating seriously — which is exactly why the next two failures matter.

SERP: Failed.

The broad chair queries include several well-documented testing publishers, and with the evidence available today we have not identified a defensible, rankable improvement angle. This is the decision-driving failure: real demand with no identified broad-level improvement angle under current evidence.

Evidence: Not yet.

Experience without data is not publishable evidence. This failure compounds the SERP one: even a narrower angle would lack the original data that gives the page distinct, publishable value.

Asset: Possible — if evidence is built first.

The worksheet and decision record are already on this page — generic framework assets, publishable now. A sizing guide, a price-band analysis, and an Amazon-data view of the category are chair-category assets that become buildable once the chair evidence exists. The decision is IMPROVE because the failed SERP and Evidence questions still have actionable remediation paths: narrow the angle, assemble publishable evidence, then re-evaluate. Asset remains conditional on that work.

Home Office Chairs passes one of four questions. The demand is real. The broad queries include several well-documented testing publishers — and the evidence is still in a notebook, not on a page. That’s IMPROVE — not GO, not STOP.

IMPROVE means the direction is right, but the angle is too wide and the evidence is not yet publishable. Two of the four questions fail today. This is not GO with a caveat, and it’s not failure. It’s a precise diagnosis with a clear next move: narrow the angle, assemble the evidence, then re-run the questions. When those two conditions are met, the market gets re-evaluated — not assumed. And if the re-run still fails, the answer drops honestly to STOP — which is a decision, not a disappointment.

IMPROVE is not failure. But what about markets where the answer is a clear no?

When the answer is stop.

Every niche tutorial makes “stop” sound like failure. The framework treats it as a successful decision with a precise job: it rules a market out before it can cost you six to eighteen months, thousands of dollars, and attention you can’t get back.

The math is simple to state and hard to feel. The resources you don’t spend on the wrong market are the resources you have for the right one. Saying no to a market you’ve evaluated honestly isn’t quitting; it’s redeploying.

A stop verdict has to be earned. It comes from a specific failed question, recorded with evidence — not from a vague feeling that a market is “too competitive.” If you can’t name which question failed and why, you haven’t decided anything; you’ve just paused.

After a stop, the process doesn’t end — it moves to the next candidate with sharper standards. Each honest no makes the next yes cheaper to find. That’s why stop is a win: it turns a year of expensive guessing into a decision you can defend in writing.

Whether the answer is go, improve, or stop — writing it down is what makes the next decision better.

Record the decision.

Recording a decision isn’t administration. It’s calibration. The value shows up at the second and third decision, not the first. Six months from now, you’ll read back “I rated this SERP improvable because I assumed the reviews were shallow” and learn more about your judgment than any course could teach.

Here’s what a decision record looks like with the chair market filled in:

A few habits make this compound. Keep the basis to a phrase you’ll still understand in six months — a date and a reason, not a feeling. Record the decision, not the hope: write what the evidence said, not what you wanted it to say. And set a re-evaluation date, because IMPROVE is a plan of work, not a verdict on life. A record that sits unread is paperwork; a record you re-read before the next evaluation is leverage.

This isn’t our template — it’s one way to do it. What matters is that the decision is written where you can read it later and be corrected by it. That’s how judgment compounds: the same person, re-reading their own evidence, gets more accurate each time.

You’ve seen the framework, run it on a real market, and recorded the result. What’s next?

What to do next.

Don’t go niche hunting. Take the candidate market already on your desk and run it through the four questions. Keep a reference copy of the framework open while you work.

Apply the four questions to your own market

If you’re new here, start here — the reading path orients you to how Search Asset Lab thinks about search assets.

Key claims in this article are tied to named sources and dated data notes so they can be checked. Run your next market through the four questions. If it passes all four, you have a case worth pursuing. If it doesn’t, you have learned what to change—or what to stop—before committing a year. Both are progress.

Data notes: Chair-query SERP data from Semrush US database, 2026-08-06. Positioning-query data from Semrush US database, 2026-08-05. Amazon category data from SellerSprite 2026.07 report, Amazon US, 100-product sample, OCR review pending. Competitor testing claims verified against page text on 2026-08-06.