Every keyword tool will give you a different number for the same phrase. That is not a bug — they measure different things, and knowing which measures what is more useful than picking a favourite. Here is the map of the categories, what each is genuinely good for, and the free stack that covers most of the job.
“keyword research for seo” — 22,200 searches/mo (Google, US)Average monthly US search volume. Source: Google autocomplete demand via AnswerThePublic, 2026.
Four categories, four jobs
Your own data (free)
Search Console: the queries you already appear for, with real positions. Nothing else knows this about your site.
Autocomplete miners
Aggregate what people type — questions, prepositions, comparisons. Best source of phrasings; volumes are modelled, not measured.
Ad-platform data (free)
Keyword Planner reports grouped ranges from advertiser data. Honest but blunt, and vaguer without an active campaign.
Full platforms
Ahrefs, Semrush: volume, difficulty, competitor gaps, SERP history. The most complete picture, priced for teams.
Start with the data nobody else has
Before paying anyone: open Search Console, filter queries to positions 8–20, and sort by impressions. Those are searches where Google already thinks you are relevant and you sit just out of reach. No paid tool can tell you this, because no paid tool knows your site the way Google does. It is consistently the highest return per hour in keyword research.
Then find the phrasings
Autocomplete miners answer the question the volume tools cannot: how do people actually say this?The gap between the phrase you would write and the phrase people type is routinely a multiple. Our own research found “seo audit for website” carrying several times the demand of the phrasing we would have guessed — and the demand charts throughout this blog come from exactly that source.
Add difficulty only when you need it
Difficulty scores matter when you have more candidates than capacity and must choose. For a small site with five pages to write, the choice is usually obvious from intent alone, and a platform subscription is an expensive way to confirm it. Check the actual results page instead: if the top ten are national brands with deep pages, you are not winning that term this quarter.
The method matters more than the tool
Whichever source you use, the sequence is the same: seed from your business, expand into real phrasings, qualify by intent, judge winnability, map one keyword per page. Full walkthrough in how to do keyword research — and when you have chosen your targets, run the free audit to make sure the pages carrying them are technically able to rank.
Frequently asked questions
- What is the best keyword research tool?
- For free, Google Search Console plus autocomplete data beats most paid tools for one specific job: finding the queries you already half-rank for. For breadth and difficulty scoring, Ahrefs and Semrush lead. For WordPress users who want research where they write, ProRank brings volume and difficulty into the editor so the keyword decision and the page happen in one place.
- Why do keyword tools report different search volumes?
- Because they measure differently. Google Keyword Planner reports grouped ranges from ad data, third-party tools model volume from clickstream and index sampling, and question miners aggregate autocomplete rather than measuring searches at all. Treat every number as an estimate of relative size, not an exact count — the ranking between keywords is far more reliable than the absolute figure.
- Can I do keyword research for free?
- Yes, and the free stack is unusually good. Search Console shows real queries and positions for your own site, Google autocomplete and People Also Ask reveal phrasings, and Keyword Planner gives volume ranges. What you give up is difficulty scoring and competitor gap analysis.
- What is keyword difficulty and can I trust it?
- It estimates how hard the current top ten would be to displace, usually from their backlink profiles. Use it comparatively — keyword A is harder than keyword B — rather than as an absolute threshold, and always eyeball the actual results page, which tells you things no score captures, like whether the top ten are all forums or all brand pages.
- How does AI help with keyword research?
- It is genuinely good at clustering hundreds of raw phrases into topics and labelling intent, which is the tedious middle step. Be sceptical of AI-generated search volumes — models will confidently invent numbers. Volume should come from a data source; let AI organise, not measure.
