A keyword research tool is software that estimates how many times a month a query is searched, how competitive it is and who already ranks for it. The word everything rests on is "estimates": none of these tools counts searches one by one, not even Google's own.
It sounds like nitpicking until you put a tool's numbers next to the real data for the same site. We did this on 16 September 2026 for Macropix, an Italian LED screen manufacturer we have worked with since 2020, and the result matters more than any ranking of tools. Then we repeated the volume check on Google UK.
Two queries, the same stated volume, different real demand
We queried the Google Ads API for the Italian market (location Italy, language Italian) for nine queries in the sector, and set them next to the impressions those same queries produced for the site in Google Search Console, read through Keyword.com on the same day. The queries are in Italian; the English meaning is in brackets.
| Query | Volume stated by Google Ads | Real impressions in Search Console | Position |
|---|---|---|---|
| ledwall | 8,100 | 2,125 | 2 |
| led wall | 8,100 | 951 | 1 |
| ledwall prezzo (LED wall price) | 720 | 418 | 2 |
| schermo led wall (LED wall screen) | 210 | 42 | 1 |
| schermi led pubblicitari (LED advertising screens) | 260 | 60 | 7 |
| schermo led pubblicitario (LED advertising screen) | 260 | 38 | 9 |
| monitor pubblicitari (advertising monitors) | 320 | 41 | 11 |
| monitor pubblicitario (advertising monitor) | 320 | 35 | 10 |
| monitor per pubblicità (monitor for advertising) | 320 | 38 | 13 |
The first two rows are the same thing written two ways. Google Ads returns 8,100 searches a month for both, the same CPC of €1.69 and a twelve-month history identical figure by figure. In Search Console the demand splits 69% to 31% in favour of the joined-up form. They are two queries that behave differently, and the only number available presents them as twins.
Further down the same happens with three different queries: "advertising monitors", "advertising monitor" and "monitor for advertising" all come out at 320 searches a month, with the same CPC of €1.06 and the same seasonal curve. Anyone who opens a spreadsheet and adds up those three rows is convinced they have 960 searches a month. They have 320, spread across three phrasings.
Google UK behaves the same way. We ran the English equivalents through the same source.
| Query pair | Stated monthly volume (each) | CPC (each) | Twelve-month series |
|---|---|---|---|
| ledwall / led wall | 1,000 | £3.03 | Identical |
| advertising screens / advertising screen | 480 | £4.21 | Identical |
| led advertising screens / led advertising screen | 480 | £4.92 | Identical |
| keyword research tools / keyword research tool | 720 | £6.50 | Identical |
Every pair returns one number twice. Add "ledwall" and "led wall" together and you get 2,000 searches a month in the UK; the source only knows about 1,000, and it can't tell you how they split.
This isn't a flaw in the tool; it is how the source works. Google Ads' official documentation says so: volumes are rounded, and when the list contains similar keywords the tool tries to divide traffic between the overlapping terms. Anyone taking data from Keyword Planner, directly or through another piece of software, inherits that grouping.
The effect has been measured at scale. Ahrefs compared Search Console impressions with Keyword Planner volumes on 72,635 randomly selected keywords in the 1,000 to 10,000 searches range: 91.45% of Keyword Planner volumes were higher than the real figure and 54.28% were well above it. The study dates from 2021 and the sample is from the US, so it doesn't transfer to every market as it stands: what it shows is that the direction of the error is systematic, not random. Our Italian check above points the same way.
Three families of tools that measure different things
Comparing keyword research tools on price is the quickest way to buy the wrong one, because three very different data sources sit inside the same category.
What Google states. Keyword Planner was built for people buying ads and returns numbers designed to estimate a budget, not to plan content. Rounded, grouped and calibrated on the demand an advertiser can buy.
The ones that rebuild the data. Ahrefs, Semrush and local suites such as SEOZoom in Italy start from that base and correct it with clickstream data and their own models. Ahrefs states 28.7 billion keywords filtered from 110 billion found, across 217 locations. Their numbers differ from the others by design, not by mistake.
The one that estimates nothing. Search Console reports the impressions and clicks your site actually got. It is the only unmodelled data you have access to, and it has one clear limit: it only covers queries you already appear for. It doesn't tell you what you are missing; it tells you what you have.
The practical consequence: a tool is for building the list of candidates and understanding the type of intent; Search Console is for calibrating that list against your site's reality. If you use only one, you are working with half the information.
The tools we use, with verified prices
Prices taken from the official price lists on 16 September 2026. They change often, so each one links to the page it comes from.
Google Search Console (free)
The first place to look, and the one most guides mention last. The Performance report contains the queries the site has already appeared for, with impressions, clicks and average position. On a site that has been live for a few years, that is tens of thousands of rows no paid tool can sell you, because they are yours.
The most profitable filter: queries with lots of impressions, few clicks and a position between fifth and fifteenth. These are already relevant pages two or three changes away from generating traffic.
Google Keyword Planner (free, requires a Google Ads account)
It is still the reference for CPC and for seasonality, which it reports month by month over the last twelve. For "led wall" on Google UK the curve runs from 1,600 searches in September and November 2025 down to 720 in August 2026, a spread that no annual average reveals. As a measure of demand, take it for what it is: an order of magnitude.
Google Trends (free)
It doesn't give absolute volumes and is underrated for that reason. It does two things the others do worse: it tells you whether a query is growing or a flash in the pan, and it compares two terms with each other. If you have to choose between two phrasings for the same page, Google Trends settles it in thirty seconds.
SEOZoom (Italian market, from €69 a month)
It is only relevant if you also sell in Italy, but it shows why local databases matter. It is the only suite built on an Italian database, and on Italian-language queries it shows: it picks up long tail the international suites have never seen. The official price list starts at €69 plus VAT a month for the Essential plan (1 site, 100 tracked keywords) and rises to €99 for Evolution, €199 for Power and €689 for Enterprise, with 15,000 tracked keywords. GEO audit and AI engine analysis features unlock from the Evolution plan.
For a company selling only in Italy it gives the most information per euro spent. For anyone working across several markets it becomes a supporting tool.
Semrush (from $139 a month)
The price list starts at $139 a month for the SEO plan ($117.33 with annual billing), then $199 for Starter, $299 for Pro+ and $549 for Advanced, with 500 to 5,000 keywords tracked a day. From the Starter plan up it includes prompt tracking on ChatGPT, Perplexity and Gemini, 50 to 200 a day depending on the plan.
The Keyword Magic Tool is still the fastest way to go from a seed to a few thousand variants grouped by subtopic. It is also where lists nobody will ever use pile up.
Ahrefs (from $29 a month)
The Starter plan costs $29 a month, then Lite at $129, Standard at $249 and Advanced at $449, according to the official pricing page. Brand Radar, the part dedicated to visibility in AI answers, starts at $199 a month as a separate product.
The metric worth the price is Parent Topic: it tells you whether your keyword deserves its own page or whether Google treats it as part of a broader topic. It is the quickest way to avoid producing three articles that cannibalise each other, which is the real job of keyword clustering.
The pay-as-you-go free tools
AnswerThePublic, KWFinder and the free versions of the suites live on daily limits. Zapier, which tried more than ninety before choosing five, reports three searches a day on AnswerThePublic, five on KWFinder and ten reports a day with ten tracked keywords on the free Semrush account. Enough to validate an idea, not to build an editorial plan.
The comparison in one table
| Tool | Entry price | Where the data comes from | Local-language coverage | Tracks AI answers |
|---|---|---|---|---|
| Google Search Console | Free | Real data from your site | Complete, but only for queries where you already appear | No |
| Google Keyword Planner | Free | Google advertising data, rounded and grouped | Good on commercial queries | No |
| Google Trends | Free | Sample of Google searches, indexed 0 to 100 | Good, with regional detail | No |
| SEOZoom | €69/month + VAT | Proprietary Italian database | The best on Italian-language queries | Yes, from the Evolution plan |
| Semrush | $139/month | Clickstream and its own models | Good, strongest on multi-country projects | Yes, 50 to 200 prompts a day from the Starter plan |
| Ahrefs | $29/month (Starter) | Its own crawler and database, 28.7 billion keywords | Fair, strong on backlinks | Brand Radar, from $199/month |
The part almost nobody covers: researching prompts, not keywords
Most guides to keyword research tools stop at Google. Meanwhile, some of the questions you used to answer with an article now go through ChatGPT, and those questions have a different shape.
Similarweb, which measures traffic with browsing data, reports that the average ChatGPT prompt is around 60 words against 3.4 words for the average Google query. The data is international and mostly from the US. The direction is clear, though: a list of two- or three-word keywords doesn't describe that behaviour.
Three ways to work on it without buying another subscription. First: in Search Console, filter queries with at least ten words. There are still few of them, but they have the shape of questions put to an assistant. Second: take the questions your sales team hears on the phone and use them as test prompts on ChatGPT, Gemini and Perplexity, noting who gets cited. Third: look at your content that already answers a complete question rather than a keyword. That is the content that gets picked up.
If you want to automate it, prompt tracking is available in Semrush from the Starter plan and in SEOZoom from the Evolution plan. We have written a dedicated guide on how the work changes with generative engine optimisation, and the projects where we are applying it are described in our AI SEO services.
How to choose, in three scenarios
You sell in one country and have a budget under £150 a month. Search Console plus one entry-level plan, such as Ahrefs Starter at $29 a month, with Semrush's free account for spot checks. For a single market you don't need the multi-country depth you pay for on the higher plans.
You work across several countries. Semrush or Ahrefs, with a local tool as a cross-check where one exists (SEOZoom for Italy) when the number doesn't add up. The typical case is a niche query the international suite marks as zero that generates enquiries every month in that market.
You run an ecommerce site. Add your site's internal search suggestions to the tools above, which no suite can see. The queries typed into the site are the most honest source of commercial demand you have, and they are free. We cover this in the guide to ecommerce SEO.
How to check a volume before trusting it
The method we use when a number decides whether a page gets made or not.
- Look up the keyword and its variants in two different tools. If they return exactly the same number and the same CPC, they are reading the same grouped source: you have one figure, not two.
- Look at the twelve-month series instead of the average. An 8,100 made up of three months at 20,000 and nine at zero is not a keyword to build a page on.
- Search the query on Google and see who is there. If the SERP is made of service pages and your idea was an article, the volume doesn't matter: you won't rank anyway.
- Cross-check with Search Console. If the site already appears for that query, you have the real number and can stop estimating.
Step two is the one most often skipped, and for Macropix it is what stopped us treating "ledwall" as a stable keyword: between September 2025 and August 2026 it swung between 5,400 and 9,900 searches a month in Italy, with the low point in August against 9,900 in May.
The rest of the method, from choosing the keyword to the outline, is in the guide to keyword research and the one on SEO strategy.
Three things no tool can tell you
Keyword difficulty is a score every piece of software calculates with its own formula, almost always based on the backlinks of the ranking pages. Two suites can give the same query 34 and 61 without either being wrong. Read it as a traffic light, not a measurement.
Zero volume doesn't mean zero searches. Google's John Mueller, at Search Central Live New York in March 2025, repeated that 15% of queries are new every day. No database can contain what nobody has searched for yet, which is why long-tail keywords often show zero and still bring in enquiries.
And none of the numbers you read accounts for how much of that volume ends in a click. Between AI Overviews, answer boxes and local packs, a growing share of searches end without leaving Google: a high volume on an informational query is worth less today than it was three years ago. The demand is there and growing; what is missing is a reliable way to count it from the outside.
Where to start tomorrow morning
Over the last few years the value of keyword research has shifted. Generating keywords has become free and instant: any suite, and by now any language model, will churn out two thousand in a minute. What has become hard, and what no subscription solves, is throwing away 1,980 of them.
The criterion for doing that isn't in the tools; it is in your data. A site that has had Search Console running for three years owns the most accurate dataset there is on its own market, and almost nobody opens it before buying a $139-a-month subscription. For Macropix that comparison rewrote the order of priorities in an afternoon: the full journey, from position 88 to second, is told in the piece on SEO for manufacturing and in the case study among our projects. If you want us to run that comparison on your data, start with our SEO services.
Frequently asked questions
For a site selling in one market, an entry-level plan of an international suite such as Ahrefs Starter, from $29 a month, plus Google Search Console covers most needs. Semrush and Ahrefs offer the deepest databases for work across several countries, and local suites such as SEOZoom in Italy are stronger on their own language. In every case, compare the figures with Google Search Console, which is free and reports real numbers rather than estimates.
Yes, combining Google Search Console, Google Keyword Planner and Google Trends covers most of the work. The limits come with competitor analysis and with volume for queries the site doesn't appear for yet. The free versions of the suites are rationed: three searches a day on AnswerThePublic, five on KWFinder, ten reports a day on the free Semrush account.
Because neither of them counts searches. They start from the data Google makes available to advertisers, already rounded and grouped, and correct it with clickstream data and their own models. Different numbers are the norm: what matters is the order of magnitude and the trend over time, not the exact figure.
It is reliable as a rough guide and as a source for CPC and seasonality. Google's documentation states that volumes are rounded and that similar keywords are treated together. In a check on nine Italian queries in September 2026 we found three different phrasings with exactly the same volume, CPC and history, and on Google UK led wall and ledwall both return the same 1,000 searches a month.
You need a different kind of research. The average ChatGPT prompt is around 60 words against 3.4 for the average Google query, according to Similarweb's international data, so a list of short keywords describes that behaviour poorly. Prompt tracking is included in Semrush from the Starter plan and in SEOZoom from the Evolution plan, but you can start by hand by filtering Search Console for queries of at least ten words.
