Tools for keyword research, and the thing none of the roundups tell you

Almost every roundup of keyword research tools compares database size, interface and price, and skips the fact that changes the decision:

Nearly all of them are reporting the same volume, from the same place.

Google Ads is the origin of search volume data for the overwhelming majority of the market, either directly through the Keyword Planner or through providers that resell it. When two tools disagree it is usually rounding, bucketing, refresh cadence, or how each treats close variants, rather than one having better data than the other.

That single fact reorganises the whole comparison. If volume is broadly commodity, then the things worth choosing on are the three that are not.

What actually differs

1. Discovery, which is where the real variation is

Finding the phrasings is a harder problem than sizing them, and it is where tools genuinely diverge. Some expand from seeds using their own crawl of ranking pages. Some mine autocomplete. Some read what a competitor already ranks for and hand you the list.

That last approach is the highest leverage one available and it is worth choosing a tool for. Pointing at a competitor and asking what they rank for that you do not produces a better candidate list in ten minutes than seed expansion produces in an afternoon, because it is grounded in queries that demonstrably have pages competing for them.

Free discovery is also genuinely good, and underused. Autocomplete expanded across the alphabet and across question words returns real strings ordered by prediction likelihood. Related searches and People Also Ask at the bottom of a results page are free. Your own site search and support inbox contain the exact wording customers use, which no tool has.

We do this with a script and a prompt rather than a subscription. The last full run took 142 seeds to 16,332 distinct real queries at no cost, and the paid step afterwards priced a thousand of them in a single request. If you would rather run the expansion conversationally, the ChatGPT prompt for keyword research and the keyword clustering prompt cover expansion and grouping respectively.

The usual mistake is paying for discovery and guessing at volume. It should be the other way round.

2. How you are billed, which decides how you work

This gets almost no attention and it changes behaviour more than any feature.

Volume data is generally billed per request rather than per keyword, with a large allowance inside each request. That means asking for eight hundred keywords in one call and asking for one keyword cost close to the same. Anybody who does not know this checks keywords one at a time and pays many times over for what a single request would have covered.

Seat based desktop tools invert the incentive: once you own the seat, running another analysis is free, which makes them excellent for the exploratory work that metered tools make you ration.

Match the billing model to how you actually work. Continuous exploration wants a flat cost. Occasional planning wants metered access you can stop paying for.

3. What happens after the export, which is where most tools stop

Every tool gets you to a spreadsheet. Almost none help with the part that decides whether the work pays, which is the cut: can we say something here that is not already being said, would this person ever buy from us, and could we win it from where we stand.

No tool answers those. If a product claims to, look closely at whether it is scoring difficulty and calling that a recommendation, which is not the same thing.

Google published its own guidance on evaluating third-party SEO tools in June 2026, and the sentence worth carrying out of it is that no third-party tool has access to Google’s internal ranking systems. That applies to every difficulty score, every “SEO score” out of 100, and every visibility index in the category, including ours. They are models of the outside of a system, which is useful, and none of them are readings from inside it.

The categories, briefly

The all in one platforms. Semrush, Ahrefs, Moz and similar. Strong discovery, competitor ranked keyword data, difficulty scoring, and everything else an SEO team needs in one subscription. You are buying the breadth, and if you only need keyword data you are overbuying substantially.

Google Keyword Planner. Free, and the origin of most of the data everybody else resells. Its limitation is well known: without meaningful ad spend it reports bucketed ranges rather than numbers, which makes separating two candidates in the same bucket impossible. It is also built for advertisers, so its notion of a related keyword skews commercial.

Data providers sold by the request. You call an API, you get raw volume, CPC, competition and trend, and you pay per request. No interface, complete control, and by far the cheapest way to get real numbers if somebody on the team can handle JSON. This is what a lot of tools you might buy are themselves using underneath.

Free and freemium point tools. Browser extensions and single purpose sites. Good at discovery, unreliable at volume, and best treated as a way to generate candidates rather than to size them.

Autocomplete and the results page itself. Free, real, and the source most likely to surface phrasings nobody else in your industry has thought to target.

What we do, since it is relevant

We buy volume from a provider by the request rather than holding a subscription, because our usage is bursty: a planning session needs a few hundred keywords priced once, and then nothing for weeks.

Two habits came out of that and both generalise:

Write the entire raw response to a file before filtering it. Paid data you did not keep is paid data you will buy again. We have made that mistake and had figures floating around in notes with no file behind them, which meant nobody could check them and they eventually turned out to be slightly wrong.

Pull volume after discovery, never before. Discovery should be wide and cheap. Sizing should happen once, on a shortlist, in a single request.

Choosing, in one paragraph

If you already own an all in one platform, you have a keyword tool and buying another is unlikely to change any decision you make. If you do not, start with free discovery and Keyword Planner, and only pay once you have a shortlist that needs real numbers to separate. If you are technical and your usage is occasional, a data provider billed per request will cost a fraction of a subscription. And whatever you use, remember that the tool’s job ends at the spreadsheet: the method is what turns that into a plan, and the newer question of whether the click survives at all is not something any keyword tool currently reports.

The half no keyword tool covers

Worth naming, because it is the gap that has opened since these roundups were first written. Keyword tools size typed queries. They do not tell you whether an assistant names you when somebody asks it for a recommendation, because there is no advertising platform selling against prompts and therefore no volume data to resell.

The measurement for that half is a rate rather than a ranking: run the prompts your buyer would plausibly ask, count how often you are named. Our AI visibility checker does it, the prompt library is the working set, and the free tools cover the technical floor underneath both, which is whether anything can fetch and read the page in the first place. For where you actually sit on the typed half, a rank tracker is still the right instrument.

Sources

Read on 15 August 2026. Vendor pricing changes constantly and none of it is quoted here for that reason; the links go to each vendor’s own page so you are reading today’s version rather than ours.

• Google Ads, use Keyword Planner, for the bucketing behaviour and what the tool is built for.
• Google, third-party SEO tools, services, and advice, for the statement that no third-party tool has access to Google’s internal ranking systems.
• Vendor pricing pages, for what each currently costs: Semrush, Ahrefs, Moz.