Keyword research when you are the small site

Most keyword research advice assumes a site that can already rank. It tells you to find high-volume terms, check the difficulty score, and write something better than the current results. On a domain with no age, no links and forty pages, that process produces a list of queries you will not touch for two years.

The method below optimises for a different thing: finding queries where the current results are beatable by a small site, and where the traffic is worth having even when it is small.

Start with what you already almost rank for

Before any tool, open Search Console and sort by impressions with a position between 8 and 30. Those are queries where you are already considered relevant and are being shown — you are just below the fold. Moving position 14 to position 6 is a fraction of the work of entering a new query cold, and it uses pages you have already written.

If a page ranks 8–20 for a query, the usual fix is one of three things: the query is not directly answered on the page, the page answers it in passing while being about something else, or the title does not contain the phrase people actually type. All three are edits, not new content.

Then mine the sources that are not keyword tools

• Your support inbox and chat logs. Every repeated pre-sale question is a query with commercial intent and almost no competition.
• Sales calls. The objections people raise are the "is X worth it" and "X vs Y" queries.
• Autocomplete and "people also ask". Free, and sourced from real searches rather than a model of them.
• Forums and communities in your niche — the phrasing real people use rarely matches the phrasing in keyword tools.
• Your own site search. People telling you exactly what they expected to find and did not.
• Competitor reviews. What customers complain about is what the next buyer will search before choosing.

These sources produce lower volumes and much higher intent than a tool export, and they surface phrasings no tool has data for — which is exactly why nobody is competing for them.

Judge difficulty by looking, not by the score

Difficulty scores are a model of link strength, and they are a blunt instrument at the long tail. Open the first page of results and answer five questions:

1. Are the top results actually about this query, or are they broader pages that happen to mention it? Broad pages losing to specific ones is the most common small-site win.
2. How many results are forums, aggregators or old listicles? Those are beatable by a page written this year with a real answer.
3. Is the intent uniform? Mixed results mean the engine is unsure, and an unsure result set is an opening.
4. Are there pages from sites your size? If every result is a nine-figure brand, move on — however good your page is.
5. How old is the content? A top result last updated four years ago in a fast-moving category is an invitation.

If four of those five point your way, the difficulty score does not matter much. If none do, no amount of content quality will fix it this year.

Prefer specific over big

A query with 60 searches a month, unambiguous intent and a weak result set is worth more to a small site than one with 20,000 searches you will never place for. Twenty of those is a real traffic channel, and each one is winnable in weeks rather than years.

There is also a compounding effect that volume-first research misses: ranking for the specific version of a topic is how a site accumulates the topical evidence that eventually makes the broad version reachable. You get to "project management software" by first owning forty questions about project management, not by attacking it head-on.

Group before you write

Several phrasings usually deserve one page, not several. The test is empirical: search two phrasings and compare the results. If the same pages rank for both, they are one query and one page. If the result sets differ substantially, they are two pages.

Getting this wrong in the direction of "more pages" is the more expensive error. Two pages targeting the same intent compete with each other, split their links, and often result in the engine picking the weaker one.

Assign a page type to each cluster

• How / what / why → an article that answers directly and links to the relevant product page.
• Best / top / comparison → a comparison page with stated criteria and an actual recommendation.
• "X vs Y" → one page per genuine pairing, and only for pairings people really weigh.
• Alternatives to X → a page that is honest about when X is the better choice. The honest version outperforms the self-serving one.
• Buy / price / near me → a product, pricing or location page. Never an article.

Sequence for momentum

Order the list so early wins fund the patience the rest requires — and remember that new pages take eight to sixteen weeks to settle:

1. Existing pages that rank 8–20 — edits, results in weeks.
2. Long-tail questions where the current results are forums and old posts — new pages, results in one to two months.
3. Comparison and alternatives pages — high commercial value, moderate difficulty, results in two to four months.
4. Category-defining terms — planned now, attempted after the first three tiers have built up authority.

What to track

Track the average position of each target query and the number of queries where you appear at all. Traffic is a lagging, noisy indicator on a small site; query coverage moves earlier and tells you whether the strategy is working before the traffic arrives to confirm it.

Frequently asked questions