Every keyword plan starts with a number: monthly search volume. Agencies build forecasts on it, clients approve budgets against it, and content teams rank their backlog by it. It is worth asking how real that number is.
The short answer is that no tool reports real search volume, and the best-known vendors say so themselves. The longer answer is more useful: the three most widely used sources compute volume in different ways, independent tests have measured how far each drifts from Search Console, and the gap between volume and the clicks you can actually win has widened sharply since AI Overviews arrived.
Key takeaways
- Google Keyword Planner, Semrush and Ahrefs all report a 12-month average, and all three are estimates. Keyword Planner groups close variants and rounds its figures.
- An independent 2023 study of 33,377 keywords found that 69% of top-10 keywords had more Search Console impressions than Keyword Planner reported.
- A separate 2023 test of Semrush on 87 keywords found a median deviation of 46%, with 71 of 87 underestimated.
- The only independent test to put Ahrefs and Semrush against the same Search Console data dates from 2020. There is no current independent ranking of the tools, and anyone claiming one should be asked for the study.
- Volume is not traffic. In a 2025 Pew study, users clicked a traditional result on 8% of visits when an AI summary appeared and 15% when it did not.
Where the three numbers come from
The three sources start from different raw material, which is the first reason they disagree.
| Source | How the vendor describes it | Averaging |
|---|---|---|
| Google Keyword Planner | Searches for a keyword and its close variants, based on location and network settings. Statistics are rounded. | Past 12 months |
| Semrush | Third-party data overlaid with historical clickstream data, then run through a computational model | Total of the past 12 months divided by 12, mobile and desktop combined |
| Ahrefs | Keyword Planner data combined with Google Trends and other third-party sources, refined with click-through rate models | Last known 12 months |
Two details in Google’s own documentation explain a lot of downstream confusion. Because Keyword Planner reports a keyword together with its close variants, several keywords can share one figure. Because its statistics are rounded, volumes for multiple locations may not add up as you would expect, and Google notes that results will be less accurate for a small geographic location, since there is less data to base them on. A tool that starts from Keyword Planner inherits both properties unless it corrects for them.
Ahrefs is candid about the limits. In a March 2024 piece its CMO put it bluntly: there is no such thing as an accurate search volume. The metric is designed to be directionally accurate, useful for comparing keywords rather than for predicting a figure.
What independent tests found
Measuring accuracy needs a ground truth, and the usual one is Google Search Console impressions for keywords where a site ranks at the top. That benchmark has limits of its own, covered below, but it is the best available.
Keyword Planner
The largest recent independent test came from the agency Upgrow in May 2023, which compared 33,377 ranking keywords over 16 months against Search Console:
| Finding | Share |
|---|---|
| Top-10 keywords with more impressions in Search Console than in Keyword Planner | 69% |
| Top-ranking keywords extremely underestimated by Keyword Planner | 63% |
| Top-ranking keywords extremely overestimated | 24% |
| Keywords with Search Console impressions but none in Keyword Planner | 7% |
The same study found Keyword Planner exaggerated impressions across top-10 terms by 163% in aggregate. Both findings are true at once: most individual keywords were underestimated, while a smaller number of heavily overestimated ones dominated the total. Which one matters depends on whether you plan around a long tail or a few head terms.
Semrush
In August 2023 the agency Altitude Marketing compared Semrush estimates against its own June 2023 Search Console data for 87 US keywords where it ranked seventh or better. The average deviation was 99% and the median 46%. Semrush underestimated 71 of the 87. For 26 terms, real impressions ran to more than twice the estimate, and one term matched exactly. It is one B2B site and a small sample of low-volume keywords, so treat it as an illustration of the direction of error in niche B2B, not a verdict on the tool.
Ahrefs, and the head-to-head that is six years old
The only independent comparison that ran several of these tools against the same Search Console data was published in Search Engine Journal in June 2020. It used 160 keywords that held a top-three position for twelve consecutive months with at least 1,000 monthly impressions, and compared nine tools including Ahrefs, Semrush and Keyword Planner. Its conclusion was that Ahrefs and KWFinder came closest to actual values. The data covers December 2018 to November 2019, and every vendor has rebuilt its models since.
Vendors have run larger comparisons, and they should be read for what they are. Ahrefs published a study of 72,635 keywords in November 2021 finding 91.45% of Keyword Planner volumes were overestimates, and reporting its own figures as roughly accurate, within 50% of Search Console, in 60% of cases against 45% for Keyword Planner. Semrush published its own multi-tool study in 2022 and ranked itself first. Both are serious pieces of work with published methods, and both were run by a company scoring its own product.
The honest summary: there is no current, independent ranking of these three tools on the same keywords. If someone tells you one is “the accurate one”, ask which study.
Why no tool can be fully accurate
Even a perfect model would be chasing a moving target, for three reasons.
- Impressions are not searches. Google counts an impression when a user has seen, or potentially seen, a link to your site. It is the best proxy available, not a count of searches.
- Grouping and rounding. Close-variant grouping means the upstream figure already describes a cluster, not a keyword. Rounding hides small differences.
- Averages flatten seasonality. A 12-month average for a keyword that peaks in one month tells you little about that month.
None of this makes the tools useless. It makes them comparative. A keyword a tool rates at five times another is very likely searched more often. Neither figure should go into a revenue forecast unchecked.
Volume is not traffic: the SERP reality
The larger problem is not whether a volume estimate is 20% off. It is that a growing share of searches never sends a click anywhere.
| Study | Finding |
|---|---|
| SparkToro and Datos, July 2024 (US and EU clickstream, September 2022 to May 2024) | 58.5% of US and 59.7% of EU Google searches ended with zero clicks. Per 1,000 US searches, 360 clicks reached the open web. |
| Pew Research Center, July 2025 (900 US adults, 68,879 searches, March 2025) | Users clicked a traditional result on 8% of visits with an AI summary, against 15% without. They clicked a link inside the summary on 1% of visits. About one in five searches produced a summary. |
| Seer Interactive, April 2026 (53 brands, 2.43 billion organic impressions, January 2025 to February 2026) | Per million informational impressions: about 33,500 organic clicks with no AI Overview, about 20,743 when the brand is cited in it, about 9,445 when it is not. |
Datos is a Semrush company, which is worth knowing when reading the first row. The Seer data also shows the picture moving: organic click-through on AI Overview queries recovered from 1.31% in December 2025 to 2.36% in February 2026. Whatever the current figure, two keywords with identical volume can differ several times over in the clicks available, depending on what sits above the organic results.
That is the part a volume column cannot show you, and the part you can check directly.
Check the SERP, not just the number
Before a keyword goes into a plan, look at the results page a real searcher in your market gets, on the device they use:
- What occupies the top of the page. Ads, a featured snippet, a People Also Ask block, a map pack or an AI summary all take clicks before the first organic result.
- Who ranks. If the page is all marketplaces or all publishers and you are neither, the volume is not your volume.
- Whether it changes by location. Keyword Planner itself warns that small-area figures are less reliable, and local intent reshuffles results city by city.
- Whether it changes by device. Mobile and desktop results differ in features and in rankings.
Doing that by hand for a handful of keywords is fine. For hundreds, across several markets, it needs structured SERP data. Shifter’s SERP API returns parsed JSON for organic results, ads, featured snippets, knowledge panels and People Also Ask, with country, language, location and device set per request:
curl "https://serp.shifter.io/v1?engine=google&api_key=YOUR_API_KEY&q=crm+for+startups&gl=us&hl=en&location=Austin%2CTexas%2CUnited+States&device=mobile"Set gl, hl, location and device explicitly on every request; leaving them to defaults produces results that are not comparable between runs. For exact cities, resolve the place through the Locations API first, as the locations documentation describes. Local tracking in more depth is covered in geo-targeted SERP APIs for local SEO tracking.
A workflow that survives contact with reality
- Use tool volume to rank, not to forecast. It is good at telling you which of two keywords is bigger.
- Cross-check with your own Search Console wherever you already rank. Your impressions beat any third-party estimate for keywords you appear on.
- Pull the SERP in each target market and device before committing, and record which features appear above the first organic result.
- Forecast clicks, not searches. Discount volume by what the SERP leaves on the table, and revisit the discount as AI summaries spread or retreat.
- Re-check periodically. Volumes are 12-month averages; SERP layouts change far faster.
The bottom line
Keyword Planner, Semrush and Ahrefs all publish estimates, all average twelve months, and all say, in their own documentation, that the figure is approximate. Independent tests show errors large enough to matter for forecasting and point in different directions depending on the keyword set. No current independent study ranks the three on the same data.
What each tool cannot tell you is what happens on the page. The same volume can hide a results page that sends clicks freely or one where most searches end without any. The number starts the conversation. The SERP, in the market and on the device that matter to you, settles it.
Sources and references
- Google Ads Help, About Keyword Planner forecasts. Close-variant grouping, rounding, and accuracy for small locations.
- Semrush Knowledge Base, What is search volume in Semrush. Data sources, modelling and 12-month average.
- Ahrefs Help Center, How accurate is keyword search volume in Ahrefs and Monthly search volume vs average search volume, and Tim Soulo, There’s No Such Thing as “Accurate” Search Volume, 18 March 2024.
- Ryder Meehan, Upgrow, Google keyword tools data study, 2 May 2023. 33,377 keywords over 16 months.
- Adam Smartschan, Altitude Marketing, Search volume estimate accuracy, 14 August 2023. 87 US keywords.
- Bartjan Sonneveld, Search Engine Journal, Keyword data accuracy study, 30 June 2020. Data December 2018 to November 2019.
- Tim Soulo, Ahrefs, GSC vs. GKP: Comparing Search Volumes for 72k Keywords, 3 November 2021. Vendor research.
- Rand Fishkin, SparkToro, 2024 zero-click search study, 2 July 2024. Datos clickstream data.
- Athena Chapekis and Anna Lieb, Pew Research Center, Google users are less likely to click on links when an AI summary appears, 22 July 2025.
- Seer Interactive, AIO impact on Google CTR: 2026 update, April 2026.
- Google Search Console Help, Performance report metrics. Definition of an impression.
- Shifter, SERP API documentation and locations documentation.