In topic-first SEO, keyword research answers two questions: does anyone search for this, and roughly how many. That is a demand check. It takes an afternoon, and it comes after the topic list from the previous page, never before it. The tool confirms the topics and hands you the phrasing people type. It does not decide the pages.
A headcount, not a ranking
I think of search volume as 'a headcount': how many people show up at the door each month asking that question. Useful to know. Not the reason to open the door, and not a number to sort your whole plan by.
The rule: use volume as yes or no and as order of magnitude, never as a ranking. The consequence is that most of the sorting in keyword-first SEO is sorting noise. 100 against 300 searches a month is noise. 100 against 10,000 is real. A tool that shows you 590 and 880 to the last digit is showing you an estimate with a lot of confidence, and in SEO land those estimates move every month.
This confuses people, but it's simple: keywords still matter, because they are how we know anyone is asking and they give us the words people use. They just don't decide how many pages you build. The search results decide that, and that is step 3.
| What the tool shows | Good for | Not good for |
|---|---|---|
| Search volume | Yes or no. Tens, hundreds, thousands | Ranking topics against each other to the nearest hundred |
| Related keywords and questions | The phrasing, and the candidates for one cluster | A page per row |
| Cost per click | A hint that buyers are behind the search | Anything precise |
| Trend | Seasonality (espresso machines peak before Christmas) | Predicting next year |
| Keyword difficulty | Very little on the long tail. See below | Deciding whether you can win |
The score that isn't there. I tested 400 long-tail keywords from one niche against the difficulty data the big tools buy from their vendor. A value came back for 8 of them. Two percent. A control run on 12 keywords that do have scores came back 12 of 12, so the blanks were real absence, not a broken lookup. A public example anyone can check: "harbor freight air compressor" gets just under 50,000 searches a month and its difficulty comes back as zero. Not easy. Absent. Most tools fill the blank with a default number and show it as if it were measured.
Worked example: difficulty coverage, 400 long-tail keywords, one niche
Step by step
- Put one seed phrase per topic into the tool. "Espresso machine" for the choosing topic, "descale espresso machine" for maintenance. The obvious words, not clever ones.
- For each topic, write down yes or no and a size bucket. Tens, hundreds, thousands, tens of thousands. Nothing finer. The finer digits change every month anyway.
- Collect the related phrasings and questions the tool lists under each seed. These are the candidates for clusters. Keep them with the topic. Do not turn them into pages yet.
- Ignore the difficulty column for anything under a few thousand searches. It is mostly a default in a blank. You will judge winnability from the actual results in step 3.
- Drop a topic only if nobody searches for it and it is far from the money. A question with no volume that a customer asks before paying still gets a page. The tool cannot see the counter.
- Write the numbers down once and stop refreshing them. A demand check is a check. It is not a dashboard.
When you need this, and when you don't
You need the demand check once per topic list, and again when you add a theme to the site. An afternoon with a free tool, Google autocomplete and People also ask covers most small businesses.
You don't need it to decide how many pages a topic gets, and you don't need a paid tool to decide whether a keyword is winnable. Both of those come from reading the results, which is what the next two pages and step 3 are about.
Next
- Next page: Search intent: the four things people want when they search. The results tell you what kind of page can win before you write a word.
- See the actual results for any keyword right now, and whether an article can win there, with the free What Ranks Here tool.
- SitePerfector runs this check on up to 1,500 keyword ideas per map and shows a dash where the data is missing instead of a default: Keyword suggestions.