You checked a keyword in RankDots and in another tool, and the volumes disagree. Both tools are working correctly. Search volume is an estimate everywhere — including in Google’s own tools — and different estimates built from different sources, models and rounding land on different numbers. Every SEO tool gets this question; here is exactly where the differences come from, and which number to use when.
Nobody counts. Everybody estimates.
Google does not publish how many times a phrase is truly searched. Every volume figure in every tool is therefore a reconstruction — and reconstructions differ by what they are built from. RankDots takes its volume from Google Keyword Planner, for your project’s search engine and country: Google’s own figure, defined as an average of monthly searches for a keyword and similar terms over twelve months. Most other tools build volume differently — typically by watching a panel of real users’ browsing and extrapolating from that sample, blended with the tool’s own index of the web. Different raw material, different arithmetic, different number. What is search volume? [link is coming]
The five things that vary
| What varies | Planner-based (RankDots) | Clickstream-based tools |
| Raw source | Google’s advertising data | A panel of real users’ browsing, plus the tool’s own crawled index |
| What the model does | Averages twelve months; folds similar terms into one figure | Extrapolates from the panel sample to the whole population |
| Variant grouping | Close phrasings share one number | Phrasings often split, each with its own smaller number |
| Rounding and window | Bucketed, rounded steps; a twelve-month window | Smoothed by the tool’s own rules; windows and refresh dates differ |
| Scope defaults | Your project’s engine and country | Whatever country set the tool defaults to — sometimes worldwide |
Two of these deserve a closer look, because they cause the biggest and most confusing gaps.
Variant grouping cuts the same demand differently
Planner folds close phrasings into one figure; many tools split them. So the same real demand appears in RankDots as one larger number on one phrasing, and in another tool as several smaller numbers spread across variants. Neither is wrong — they are different slices of one pie. But comparing a folded number against a split one is comparing a whole against a piece, and the gap that produces is not an error in either tool.
Even Google disagrees with Google
Compare Planner’s figure with your Search Console impressions for the same query and you will often see a large gap — and both come from Google. They measure different things: Planner estimates total demand, averaged and folded; impressions count the occasions your site actually appeared in results, exactly as queried, no averaging. A measured count of your appearances and an estimated average of everyone’s searches have no obligation to match. What is measured and what is estimated is the full map of that distinction.
Difficulty and cost-per-click differ even more
Volume at least aims at one underlying quantity. Difficulty does not: every tool’s difficulty score is its own formula computed over its own index — RankDots’ is an average built from the results pages, others weigh links, authority or their own signals. The scores share a name and a scale, and nothing else. Compare difficulty within one tool, never across tools. Cost-per-click ranges vary for the same reason: different advertising data, different aggregation.
Which number to use when
- Pick one tool per decision and stay in it. Estimates exist to compare options against each other, and that only works inside one consistent model. The delta between two tools’ estimates is not a signal about the keyword — it is a fact about the tools.
- Let measurements outrank all of it. When your Search Console shows what actually happened, that number wins over every estimate in every tool, ours included.
- Check the scope before calling anything wrong. A worldwide figure against a one-country figure is the most common “huge discrepancy” there is. Match engine, country and language first.
When another tool’s number is the right one
- Your team already standardises on it. For reporting, continuity beats source: numbers tracked in one model for years are worth more than a switch to a different estimate.
- The job is outside our data. Backlink profiles, crawl diagnostics, ad auction planning — tools built on those datasets should be read in those datasets.
- Fast-moving queries. Different pipelines refresh on different clocks, so on volatile topics another tool’s estimate can move sooner — or later — than a twelve-month average. On such queries, treat every volume figure loosely.
And if a gap still looks wrong after scope, grouping and kind are accounted for, The numbers look wrong [link is coming] walks the remaining causes step by step.
What next
- What is measured and what is estimated
- What is search volume? [link is coming]