Keyword research has a reputation problem. The word brings to mind a tool subscription, a spreadsheet with forty columns, and an afternoon sorting numbers that turn out to be estimates anyway. Most people who run a small blog do it once, badly, and then never again.
You do not need the spreadsheet. You need a short list of questions your customers already ask, a way of telling which of them a blog post can win, and the discipline to write those and not the rest. That is the whole playbook. It fits on one page.
Start from questions, not keywords
A keyword is a guess about how a person types a question. The question is the thing. If you start from the question, the keyword follows, and if you start from a keyword you can spend a week polishing a phrase nobody means.
So the first step is collecting questions, and the best sources are already in your building:
- Sales and support conversations. Whatever prospects ask on the third call, and whatever customers ask in the second week, is a topic. Ask the people who take those calls for the five questions they are tired of answering.
- Your own search box and your competitors' help docs. Both are lists of what people could not find on their own.
- The "People also ask" box. Search your core topic in a private window and expand the box a few times. It is a free map of the adjacent questions searchers actually have.
- Community threads. The forums, subreddits and Slack groups your buyers sit in. Look for questions that get asked repeatedly and answered badly.
Write each one down as the person would say it, not as a phrase. "How do I know if my agency needs a payroll provider" is a topic. "Agency payroll" is a fragment.
Aim for thirty questions. You will not write thirty posts. You will write the eight that survive the next two steps.
Three checks that replace the spreadsheet
Volume numbers from tools are estimates, and for the queries a small site can actually win they are usually rounded to zero anyway. Three quicker checks tell you more.
Is anyone asking?
Search the question. If autocomplete finishes it for you, or a "People also ask" box repeats it, people are typing it. If the results page is nothing but a few forum threads and an outdated PDF, demand is thin but the field is open, which is often the better bet for a small site. If nothing on the page is even about the question, nobody asks it in those words. Reword it and search again before dropping it.
Can a blog post win it?
Look at what the first page is made of, the way the field guide to search intent walks through. Ten explainers means a blog post is the right shape. Ten product pages means the searcher wants to buy, and a blog post will not rank. A page full of the biggest names in your category means you would need a very different article to break in.
The useful question is not "is this competitive" but "is there a gap on this page". A thin answer, an old answer, an answer written for a different audience than yours: any of those is a gap you can write into.
Would you want the reader?
A query can be popular, winnable and useless. "Free invoice template" ranks easily and brings in people who will never buy an accounting product. Ask what a reader who typed this is likely to do next, and whether that next step has anything to do with what you sell. If the honest answer is no, the post is a vanity metric waiting to happen.
Keep the questions that pass all three. Cut the rest. This is the step most people skip, and it is why most blogs have sixty posts and no traffic.
Cluster before you write
You now have a short list. Do not treat it as a list. Group it.
Questions cluster naturally around a handful of themes. Put the ones that share a reader and a stage of the buying journey together, and the outline of a small cluster appears: one broad post that answers the theme, and several narrow ones that answer the questions inside it. Planning clusters on a small budget covers how to sequence and finish one, so here the only point is this: a question that belongs to no cluster is a lower priority than a question that completes one, even if its numbers look better on their own.
This is also where you catch cannibalisation early. Two questions that would produce the same post get one post. Merge them now, not after both are published and splitting the ranking between them.
Where the agent fits, and where it does not
If you run your blog through RankMill, the topic finder does a version of the first step for you, and it is worth being precise about what that version is.
The topic finder reads the project's website profile, the document the research run wrote about your business, audience, voice and positioning, and suggests five topics at a time, each with a title and an outline of two or three sentences saying what the post would cover. It is told to stay relevant to the audience the profile describes, to differ from what the competitors already cover, to follow the project's writing rules, and never to repeat a title the project already has. It runs on the project's cadence, weekly by default, or whenever someone presses "Find topics". Either way a hunt spends one run.
What it does not have is a keyword tool. There is no search volume, no difficulty score and no results page in front of it. It works from the profile, plus whatever it can see on your own site, which is closer to the "collect questions" step above than to the three checks. Treat its five suggestions the way you would treat five questions from a colleague: a good starting list, not a verdict.
That makes the approval step the natural place for the checks. Every suggestion arrives as a post marked Suggested, with its outline, and nothing is written until it is approved, by you or by auto-approve if the team has switched that on. With auto-approve off, read the outline, search the question, look at the first page, and approve the ones that pass. Leave the others unapproved. Nothing is lost by leaving a suggestion where it is, and nothing is spent.
Two more things help the loop work in your favour:
- Put your themes in the profile. The profile is editable, and the topic finder reads it on every hunt. If you have decided which clusters you are building this quarter, say so there, and the suggestions will lean that way instead of wandering across everything the business could plausibly write about.
- Bring your own winners. A question you found through the checks above, that the topic finder never suggested, does not have to wait for a future hunt. "Write from an idea" takes a title and a short brief and sends it straight to the writer.
One detail that surprises people: a suggestion sometimes arrives with questions attached, up to four of them. They are never about keywords, tone or structure. They are about facts the writer cannot get from the profile or the site, such as a price, a customer you can name or a date something changed, and they exist so the post is written from your answers instead of the agent's guesses. Answer them if you can. If you do not, the writer leaves those facts out rather than inventing them.
A weekly rhythm that does not need a tool
Once the list exists, keeping it alive takes a small habit rather than a project:
- Add questions as they arrive. A sales call, a support ticket, a thread someone forwards. One line each, in the customer's words.
- Run the three checks on anything new, once a week, in ten minutes. Most new questions fail one of them, which is fine.
- Slot survivors into a cluster, or start a new one only when the current cluster is finished.
- Search the questions you already ranked for every couple of months. Results pages shift, and a post that slid usually slid because the page changed shape, not because your prose got worse.
That is the whole system. No subscription, no export, no forty columns.
The short version
- Collect questions in the customer's words, from the people who already hear them
- Check each one three ways: is anyone asking, can a post win it, would you want that reader
- Cut what fails, and cluster what survives before writing anything
- The topic finder is a good source of questions, not a substitute for the checks; run them at the approval step
- Add your own finds through "Write from an idea" and steer future hunts by editing the profile
The reason keyword research feels like a chore is that most versions of it try to make the decision for you with a number. The number is not the decision. Whether a real customer asks the question, whether a post can answer it, and whether you want that person on your site: those are the decision, and you can make it without a spreadsheet.