Every keyword research tool shows you a number with no error bars. "blue running shoes — 8,100/mo." That precision is fake. The number is a rounded, averaged, aggregated estimate that originated in an advertising forecasting product, and by the time it reaches your spreadsheet it has passed through at least two lossy transformations.
This matters because volume estimates are usually the primary input to content prioritization. If you don't know how the number is produced, you can't know which decisions it's safe to make with it. Most of the time it's safe to use volume as a ranking between keywords and unsafe to use it as a forecast of anything.
What Keyword Planner actually returns
Google Keyword Planner is a media planning tool for advertisers, not a search analytics product. That shapes everything about the output.
It reports a 12-month average by default. A term that gets 60,000 searches in November and near zero the rest of the year shows up as roughly 5,000/mo. The monthly breakdown is available in the UI and the API, and it is the only version of the data worth using for anything seasonal. If your tool imports the average and throws away the curve, you have lost the most decision-relevant part of the dataset.
It buckets and rounds. Accounts without meaningful spend see ranges like "1K–10K" rather than point estimates. Even with spend, values are rounded to a grid — you will notice that volumes cluster on values like 720, 880, 1,000, 1,300, 1,600. Those are rounded, not measured. Two keywords both showing 1,600 may differ substantially in true volume.
It aggregates close variants. Keyword Planner groups terms it treats as effectively the same for ad matching: singular/plural pairs, common misspellings, accent and spacing differences, and often word-order permutations. You can verify this yourself in minutes — pull a singular and its plural, and a term and its common misspelling, and check whether the numbers are byte-identical. When they are, you're looking at one bucket reported twice, not two independent measurements. This is the single biggest source of double counting in keyword spreadsheets: people sum volume across a cluster that Google already summed.
It depends on your targeting settings. Location, language, and network settings change the number. A US-English pull and a US-all-languages pull are different questions. Tools that don't expose their targeting assumptions are giving you an answer to a question you didn't ask.
Where third-party numbers diverge
Ahrefs, Semrush, Moz and the rest do not have access to Google's query logs. They build volume estimates from some combination of:
- Clickstream panels — browser extension and ISP-derived data from a panel of users, extrapolated to a population.
- Keyword Planner anchoring — using Planner data as a calibration source, then modeling around it.
- SERP-derived features — inferring demand from what appears on the results page and from their own crawl data.
Each vendor weights these differently, which is why the same keyword returns three different numbers in three tools. Panel extrapolation has a specific failure mode worth understanding: relative error scales inversely with the panel hit count. For a term with millions of monthly searches, a panel of a few hundred thousand users gives a stable estimate. For a term with 40 searches a month, the panel probably saw it zero, one, or two times, and the extrapolated figure is essentially noise multiplied by a large constant.
That's the mechanism behind a pattern you've probably noticed: head terms are roughly consistent across tools, long-tail terms are wildly inconsistent.
Why "zero volume" doesn't mean zero demand
A reported volume of 0 or 10 means "below the reporting floor of this estimation method." It does not mean nobody searches this.
Google has said publicly that roughly 15% of the queries it sees each day are ones it has never seen before. A permanently refreshing long tail cannot, by construction, appear in a 12-month average built from historical data. New product names, new model numbers, newly phrased questions, and anything conversational will under-report or not report at all.
The practical consequence: for long-tail and question-shaped content, volume tools tell you almost nothing useful, and Search Console impressions from pages you already have tell you a great deal.
What Search Console impressions can and cannot substitute for
Search Console is real logged data about your property, which makes it more trustworthy than any estimate — within a narrow scope.
| Source | What it measures | Main distortion |
|---|---|---|
| Keyword Planner | Modeled ad-matching demand | Rounding, close-variant grouping, 12-month averaging |
| Third-party tools | Panel extrapolation + modeling | Huge relative error on low-volume terms |
| Search Console | Queries where you appeared | Only your impressions; anonymized queries omitted |
| Google Trends | Normalized relative interest | No absolute values; sampled |
Search Console impressions are a floor, not a volume figure. You only get an impression when your page was in the result set the user saw, so a page ranking on page four accumulates a fraction of the true query volume. Queries below Google's privacy threshold are dropped entirely from the query dimension, so summing query-level impressions understates the page total. And impressions count result appearances, not people — a user who searches twice generates two.
Used carefully, though, GSC gives you something no estimate can: the actual phrasing distribution real users bring to your topic, including the variants no tool reports.
Reasonable inference versus documented behavior
Be clear with yourself about which is which. Google documents that Keyword Planner reports averages, supports monthly breakdowns, and shows ranges for low-activity accounts. Google documents close-variant matching in Ads. The specific grouping rules Keyword Planner applies to keyword ideas are not fully documented — that's inference from observable identical values, and it's strong inference, but it's inference. Third-party methodology is partly disclosed and partly proprietary; treat vendor comparisons of "accuracy" as marketing unless they publish the validation set.
What to do with this
- Pull monthly breakdowns, never the 12-month average, for anything with a seasonal shape. Store the twelve values, not one.
- Deduplicate before summing. Within a cluster, check for identical volume values and treat them as one number. Sum across clusters, not within them.
- Use volume ordinally. "Term A is bigger than term B" survives rounding. "Term A will produce 8,100 impressions" does not.
- Set a floor for decision-making. Below a few hundred monthly searches, treat tool differences as noise and decide on strategic fit, intent, and existing GSC evidence instead.
- Calibrate against your own data. Take twenty keywords where you rank in the top three, compare your GSC impressions to the tool's volume, and derive your own correction factor per tool. It won't be perfect, but you'll learn whether your vendor systematically over- or under-reports in your vertical.
- For high-stakes bets, buy the data. A short exact-match paid search test on a handful of terms costs less than a content program built on a modeled number, and it returns real impression counts from Google's own inventory.
The goal isn't a more accurate volume figure. It's knowing which claims your number can support.