The argument against keyword research is that it kills creativity, and the argument for it is that publishing into silence is worse. After ten years of writing for sites that nobody read and sites that read better than they deserved, I fall on the research side, provided the research stays short.
Write down the words your customers already use
Start with a plain list of twenty phrases: the questions in your inbox, the terms people use on the phone, the wording on the invoices. This list beats any tool's suggestion, because it comes from people who were ready to pay. Tools then extend the list, they do not replace it.
Sort by intent, not by volume
Two searches with identical volume can be worth completely different amounts. One describes a problem, one describes a purchase. Group your list into three buckets before you compare numbers.
- Problem: something is broken, confusing or urgent.
- Comparison: two or three options are on the table.
- Decision: the reader has chosen and wants the shortest path.
A page written for a problem query that ranks for a decision query converts badly, and the reverse is worse: you end up answering a buying question with a definition.
Judge difficulty honestly
For a small site, the honest question is not whether a term has volume, it is whether page one currently contains anything you could improve on. Open the results, read the top five, and ask whether yours would be the most specific answer. If the first page is made of original research, government pages and decade-old threads, choose a narrower phrase and win that instead. Long phrases with modest volume and clear intent are the whole game for a site under a few years old.
Build a plan you can actually publish
- Pick one main phrase and three to five variations that fit the same page.
- Check whether two of your existing posts compete for it, and merge them if so.
- Decide the format from the results: if the page one is lists, write a list; if it is steps, write steps.
- Schedule the piece for the week you can give it proper editing, not the week you have a headline.
- Revisit the ranking after six to eight weeks and improve the section that underperforms, rather than rewriting the whole thing.
Refresh what already almost works
Before commissioning anything new, look at what you have published and pick the pieces sitting on page two. Improving a post that already has links and impressions beats starting from zero, and the work is mostly the same either way.
- List every page that ranks between positions eight and twenty for its main phrase.
- Open page one and note what the top three results cover that yours does not.
- Add the missing section, cut whatever no reader reached, and update the examples.
- Change the title only if it genuinely misrepresents the answer, because titles carry history.
- Give it four weeks before deciding again.
Three checks before anything gets published
- Intent match: does the result page for this phrase look like the page you wrote?
- One-page rule: is there already a post on your own site that would compete with it?
- Specificity: does your version say something the top five cannot say without you?
A yes to all three is rarer than it sounds, and when all three fail the honest move is to choose a narrower phrase rather than to write a longer page. Length does not fix a mismatch, it just delays the same disappointment by two months.
Signals I stopped trusting years ago
- Raw monthly volume without intent attached to it.
- Difficulty scores for niches the tool has clearly never seen properly.
- Any metric that cannot be tied to an action you would take differently.
- Competitor calendars that exist because their agency needs billable output.
Where I keep the working list
Everything lands in one document with three columns: the phrase, the intent bucket, and the decision. The decision can only be three words: write, merge or ignore, because anything longer means I have not actually decided.
- One sheet per topic cluster, not one giant sheet that nobody scrolls.
- A dated note on every decision, so the quarterly review can see why it was made.
- Dead phrases struck through rather than deleted, which prevents relitigating them.
- The competitor pages that beat us linked from the phrase, so the gap stays visible.
- One owner per sheet, because a shared list with no owner is a wish list.
- New phrases added only after the current batch has been acted on.
- Two competing drafts merged before either of them is written, never after.
- A note on the phrase we abandoned last quarter, so the argument stays settled.
What to ignore for now
Skip keyword density calculations, skip chasing every seasonal spike, and skip anything a tool marks red without you having read the page one yourself. A keyword strategy that fits on one sheet of paper and gets reviewed every quarter will outperform a spreadsheet nobody opens, and the review is where the actual decisions happen.
The next step is the twenty-phrase list: write it today from memory and your last ten customer messages, then sort it into the three intent buckets before you open any tool at all. Once the buckets exist, picking the first topic takes minutes rather than an afternoon, because the decision has already been made by the sorting.