Keyword research is the work of finding what people actually search, and deciding which of those searches are worth competing for. The output is not a list of keywords. It is a mapping of keywords to pages, ordered by commercial value.
The most common mistake is prioritising by search volume. A keyword with 10,000 monthly searches and informational intent produces readers. A keyword with 200 searches and transactional intent produces customers. The 200 is usually worth more, and cost-per-click will tell you so, because advertisers have already worked this out and are paying real money for it.
Volume is the worst signal you have
And it is the one every keyword tool puts in the biggest column.
what is a sales funnel has 1,300 monthly searches. white label seo pricing has a fraction of that. The second one is worth more, and it is not close. One is somebody learning. The other is somebody buying.
The fix: sort by intent, then by CPC.
CPC is the market's own valuation of a keyword. If advertisers are paying $38 a click for seo for b2b saas, that is thousands of businesses collectively concluding that the traffic converts. That is a far better signal than volume, and it is free.
Keyword difficulty scores are not reliable
Say this out loud, because the entire industry plans around them.
Keyword difficulty is a vendor's model. It is not a Google metric. Different tools give different scores for the same term. In a recent 30,000-keyword dataset we analysed, only 56% of rows even had a difficulty score, and terms that were obviously competitive were being scored as easy.
What to do instead: look at the actual SERP.
Who is ranking? Are they national brands or small businesses? How many referring domains do the ranking pages have? That is a real number, not a model.
Are directories and forums ranking? That is the strongest opportunity signal there is, because it means no business has covered the topic properly.
Ten minutes looking at a SERP beats any difficulty score.
The intent taxonomy
| Intent | Pattern | Funnel | Build |
|---|---|---|---|
| Transactional | "buy X", "X pricing", "hire X", "X near me" | Bottom | A page. Not a blog post. |
| Commercial | "best X", "X vs Y", "X alternative", "X review" | Middle | Comparison content |
| Informational | "what is X", "how to X", "why X" | Top | Guides, explainers |
| Navigational | "[brand] login" | Existing | Nothing |
Build bottom-up. Destination first, then the road. Agencies that start with the blog spend a year sending traffic to nowhere.
How to actually do it
1. Seed. What does the business sell, and what would someone call it? Ask sales. Ask the support inbox, which contains the questions customers actually ask, in their own words.
2. Expand. Ahrefs, Semrush, Search Console, Google's autocomplete, People Also Ask, Reddit, and the site's own internal search logs.
3. Clean. This is the step everyone skips. Strip out geographic contamination, brand queries for other companies, and terms from adjacent industries that happen to share a word. In a real 30,000-keyword dataset we cleaned recently, "SEO" surfaced a Korean surname and "white label" surfaced whisky brands. If you do not clean, you will build a strategy on noise.
4. Classify by intent.
5. Cluster. Group keywords that should be answered by one page. seo retainer, seo retainers, monthly seo retainer are one page, not three. Building three is how you cannibalise your own rankings.
6. Map to pages. One primary keyword per page, with secondaries. Never two pages targeting the same primary.
7. Prioritise. Intent, then CPC, then volume, then SERP difficulty as judged by the actual SERP. In that order.
Long-tail keywords
Lower volume, higher intent, far easier to win, and there are vastly more of them.
plumber is unwinnable and half the people searching it are doing a school project. emergency boiler repair Leeds Saturday is a customer with a wallet out.
And long-tail is where programmatic SEO lives. Hundreds or thousands of pages, each targeting a specific, low-competition, high-intent query, generated from structured data. → Programmatic SEO
One caution: long-tail informational is being absorbed by AI Overviews. Long-tail commercial is not.
Using ChatGPT for keyword research
What it is genuinely good for: brainstorming seed terms, generating question variants, grouping a messy list into clusters, and articulating the intent behind an ambiguous term.
What it is bad for: volume, difficulty and CPC. It does not know these numbers and it will invent plausible ones. We have seen agency keyword decks built on hallucinated volume figures, presented to clients with confidence.
Use it for the thinking. Use a real tool for the data. Never let it produce a number.
A keyword gap you should look for
Every keyword export is bounded by the seeds you chose. If you never seeded "AI content", you will never see that cluster.
Test for this deliberately. List the questions you know clients ask. Search your dataset for them. If they are not there, your seeds were incomplete, not the market.
What is keyword research?
Finding what people actually search and deciding which searches are worth competing for. The output is a mapping of keywords to pages, ordered by commercial value.
Should I target high-volume or high-intent keywords?
High intent. A keyword with 200 searches and transactional intent usually outperforms one with 10,000 and informational intent. CPC is a good proxy.
Are keyword difficulty scores accurate?
Not reliably. Keyword difficulty is a vendor's model, not a Google metric. Look at the actual SERP instead: who ranks, how many referring domains they have, and whether directories or forums are ranking.
What are long-tail keywords?
Longer, more specific queries with lower volume and much higher intent. They are far easier to win and there are vastly more of them.
Can I use ChatGPT for keyword research?
For brainstorming and clustering, yes. For volume, difficulty or CPC, no. It will invent plausible numbers.
What is keyword cannibalisation?
Two pages on the same site targeting the same primary keyword, splitting the ranking signals between them. Cluster related terms onto one page.
Which keyword should a page target?
One primary, with related secondaries clustered onto it. Never two pages competing for the same primary term.
Still Have Questions on Your Mind?
Contact our support team and we’ll guide you every step of the way.