A CTR-by-position curve built from your own Search Console queries is the fairer yardstick because it reflects your own results pages and query mix, while an industry average describes other people's pages. Google says what a position number means depends on the situation, and the Google pages we read give no CTR figure for any position.
- Google defines CTR in the Performance report as clicks divided by impressions (Google, Search Console Help, read 30 October 2026).
- Google says what a position number means depends on the situation, and defines average position by the topmost result from your site (Google, Search Console Help, read 30 October 2026).
- Penlid's CTR gap tool draws a point on its curve only where a position holds at least 1,000 impressions from at least 5 queries of 100 impressions or more, and it needs at least 4 such points (Penlid, CTR gap tool, read 30 October 2026).
- Penlid's CTR gap tool lists a query only if it has at least 100 impressions, a CTR at most 75% of the curve, and a Wilson interval, at 2.576 standard errors, whose top is below the curve (Penlid, CTR gap tool, read 30 October 2026).
What is a CTR-by-position curve, and what does a gap against it show?
A CTR-by-position curve shows the typical click-through rate at each average position for a set of queries, and a gap is a query whose own CTR sits below the curve at its position. Google defines CTR as the click count divided by the impression count
and average position as the average position of the topmost result from your site
, so a curve is a ratio worked out per position: group the queries by position, add up each group's clicks and impressions, and divide. Google lists finding queries with high impressions and low CTR
as a common use of the queries data; a curve makes "low" mean low against your own queries at the same position (our reasoning), and the free CTR gap tool draws it from your export.
Why is an industry average a weak yardstick for your own queries?
An industry average is a weak yardstick because it describes other people's results pages, and a position number alone does not say which layout, searcher or query mix produced it. Each difference below comes from Google or the tool's page; the conclusion is our reasoning.
- Position means different things. Google says
a position number can mean different things in different situations
and that results arespecific to the time, place, device, and recent history of the person searching
. An average from elsewhere mixes situations you cannot see. - Layouts differ. Google says an AI Overview
occupies a single position in search results
, and the tool's page adds that queries with and without an AI answer sit in one file, which cannot separate them, so your curve blends them in your proportions and an outside average blends them in someone else's. - Query mix differs. Google says branded queries
typically lead to higher-ranking pages from your site and result in higher click-through rates
, and the tool's page says they pull the curve up, so it leaves out the brand terms you type; see how to split branded from non-branded traffic.
We cite no industry average: none appears on the Google pages we read, and studies by other companies are not primary sources.
How does the CTR gap tool build a curve from your own export?
The CTR gap tool builds the curve by grouping the queries in your Queries export by whole position, dividing each group's clicks by its impressions, and smoothing the result so it never rises with position. Brand terms you type are left out of the curve and the list, and the rules below decide which groups count.
| Rule | Value |
|---|---|
| Positions covered | 1 to just under 21, in groups 1 position wide |
| Impressions in a group | At least 1,000 |
| Queries in a group | At least 5, each with at least 100 impressions |
| Share one query may hold | At most 50% |
| Points needed for any curve | At least 4 |
| Smoothing | A weighted fit that never lets the curve rise with position |
Each point sits at its group's impression-weighted average position. If your file is too thin, the tool says how far short it is and draws nothing, so a longer date range helps. Export the Queries table with Clicks, Impressions, CTR and Average position switched on.
How does the noise guard tell a real gap from luck?
The noise guard lists a query only when three tests all pass: at least 100 impressions, a CTR at most 75% of the curve at its position, and a Wilson interval whose top, at 2.576 standard errors, is still below the curve. The click gap is impressions × (the curve's CTR minus the query's CTR), and the "at least" figure, taken from the top of the interval, is the part not counting noise.
Wikipedia says the Wilson interval can be safely employed with small samples and skewed observations
, which suits a CTR of a few percent on a few hundred impressions (our reasoning); Penlid's methodology page explains why Wilson is used rather than the textbook normal interval. The 2.576 is stricter than the usual 1.96 because hundreds of queries are checked at once. It is 99.5% one-sided, so about 0.5% of queries exactly on the curve would pass the third test by chance: with 400 queries judged, 0.005 × 400 = 2 would be expected to pass if the curve were exact (our arithmetic). The tool's page says the real count is higher, because the curve is estimated from the same queries it judges.
The lowest CTRs are partly bad luck too. Picking the queries furthest below a curve picks some whose CTR was low by chance, so their gaps tend to shrink on another export. That is regression toward the mean, the effect behind pages that dropped recovering on their own.
What should you do with a query the tool lists?
Treat a listed query as a place to look: click it in the Queries table, use the Pages tab to see which URLs were shown
, and compare that page's title and snippet with the query and with the rest of the results page. Re-export a few weeks later; a gap that persists is a better candidate for a change than one that shrank (our reasoning). If your blended CTR is falling, see why impressions rise while clicks fall.
Example (illustrative numbers)
These counts are invented to show the arithmetic; they are not data from any site, and the "published" CTRs below are made up, not real averages. Brand terms are already left out of these position groups.
| Position | Impressions | Clicks | CTR | Curve |
|---|---|---|---|---|
| 1 | 17,500 | 2,100 | 12.0% | 12.0% |
| 2 | 14,200 | 1,136 | 8.0% | 8.0% |
| 3 | 11,800 | 590 | 5.0% | 5.47% |
| 4 | 10,600 | 636 | 6.0% | 5.47% |
| 5 | 9,400 | 282 | 3.0% | 3.0% |
The tool's curve never rises with position, but position 4 (6.0%) sits above position 3 (5.0%), so the fit pools them, weighted by impressions: (590 + 636) ÷ (11,800 + 10,600) = 1,226 ÷ 22,400 = 5.47%. Points sit on whole positions here for simplicity.
Two queries sit at position 4: query A has 84 clicks from 2,400 impressions (3.5%) and query B has 132 from 2,400 (5.5%). Each is tested against your own curve, 5.47%, and against two made-up published values for position 4, 8.0% and 4.0%. The Wilson top is (p + z^2/(2n) + z*sqrt(p(1-p)/n + z^2/(4n^2))) / (1 + z^2/n) with z = 2.576; for A it is (0.035 + 0.00138 + 0.00976) ÷ 1.00276 = 0.0460, and for B it is 0.0683.
| Query | CTR | Wilson top | Own curve, 5.47% | Made-up table, 8.0% | Made-up table, 4.0% |
|---|---|---|---|---|---|
| A | 3.5% | 4.60% | Listed, gap 47 | Listed, gap 108 | Not listed |
| B | 5.5% | 6.83% | Not listed | Listed, gap 60 | Not listed |
Against your own curve, A passes all three tests: 2,400 is at least 100 impressions, its 3.5% is below 0.75 × 5.47% = 4.10%, and its top, 4.60%, is below 5.47%. Its gap is 2,400 × (0.0547 − 0.035) = 47 clicks, and at least 2,400 × (0.0547 − 0.0460) = 21. B sits above the curve, so it is not listed.
The 8.0% value lists both: A with 2,400 × (0.080 − 0.035) = 108 clicks and B with 2,400 × (0.080 − 0.055) = 60 clicks, although B is an ordinary query for this site. The 4.0% value lists neither, because 0.75 × 4.0% = 3.0% is below both CTRs (3.5% and 5.5%), so A's real gap is missed. A yardstick from other pages can invent gaps or hide them; which way a real average errs, we cannot say.
What this cannot tell you
- A gap is a place to look, not a diagnosis. A query can sit below your curve because its searchers want an answer the results page already gives, because of what else is on that page, or because your title and snippet miss what they typed, and the file does not say which. The click gap is also not a forecast.
- The curve is relative to your own site. If weak titles hold down every query, the curve sits low and few queries stand out, and it cannot show how you compare with other sites (our reasoning). Each query is also part of its own curve and pulls it toward itself.
- Averages blend different pages. Each point averages queries whose positions are averages too, so the curve says nothing about a single results page. The noise guard allows for counting noise only, not for queries being different in kind.
- The export is partial. Google says
Data is truncated to 1,000 rows of representative examples of your data
and thatanonymized (rare) results are omitted from the table
, so the curve rests on the rows present; see why an export stops at 1,000 rows. - Impressions carry a documented error. Google says impressions were not accurately reported from 13 May 2025 until 27 April 2026, that CTR and average position were affected too, and that
clicks were not affected by the error
, so a curve drawn from those days carries it (our reasoning); see whether a drop was real or a data anomaly.
Questions people also ask
What is CTR by position?
CTR by position is the click-through rate measured separately at each average position, for example across every query that averages position 3. Drawn as a line, it shows how clicks per impression change as the position number grows. Penlid's tool forces that line never to rise with position, a modeling choice and not a Google rule.
How is CTR calculated, and how do you work it out for several queries together?
CTR is clicks divided by impressions. For several queries together, add the clicks, add the impressions and divide once; averaging the separate CTRs would let a query with 4 impressions count as much as one with 3,000. With 1 click from 4 impressions (25%) and 90 from 3,000 (3%), the pooled CTR is 91 ÷ 3,004 = 3.0%, while a simple average of the two would say 14%.
What do clicks, impressions, CTR and position mean in the top queries table?
A click is someone following a link from Google to your site, and, in general, an impression is counted when a result of yours appears on the page of results the searcher is on, whether or not it is scrolled into view. CTR is the first divided by the second. Position is the average rank of your highest result for that query, so a row showing 4.0 and 3.5% means your best-placed result averaged fourth place and 3.5 in every 100 impressions ended in a click.
Sources
- Performance report (Search results): Overview and basic setup, Google, Search Console Help, read 30 October 2026.
The click count divided by the impression count
The average position of the topmost result from your site
specific to the time, place, device, and recent history of the person searching
- Performance report (Search results): Dimensions and data groupings, Google, Search Console Help, read 30 October 2026.
to find queries with high impressions and low CTR for opportunities
- What are impressions, position, and clicks?, Google, Search Console Help, read 30 October 2026.
a position number can mean different things in different situations
An AI Overview occupies a single position in search results
- Introducing the branded queries filter in Search Console, Google, Search Central Blog, posted 20 November 2025, read 30 October 2026.
Branded queries typically lead to higher-ranking pages from your site and result in higher click-through rates
- CTR gap by position, Penlid, read 30 October 2026.
- Methodology: how we measure, and what we can't, Penlid, read 30 October 2026.
- Binomial proportion confidence interval, Wikipedia, read 30 October 2026.
It can be safely employed with small samples and skewed observations
- Performance report (Search results): Common tasks and use cases, Google, Search Console Help, read 30 October 2026.
View the Pages tab to see which URLs were shown
- Export data directly from a Search Console report, Google, Search Console Help, read 30 October 2026.
Data is truncated to 1,000 rows of representative examples of your data
- Performance report (Search results): Troubleshooting data discrepancies, Google, Search Console Help, read 30 October 2026.
anonymized (rare) results are omitted from the table
- Data anomalies in Search Console, Google, Search Console Help, read 30 October 2026.
clicks were not affected by the error