Every CTR benchmark table you have seen disagrees with the others
Search for "Google CTR by position" and you will find a dozen tables of neat percentages. They do not agree. Some put the first result near a fifth of all clicks, others at more than half. They cannot all be right, and none of them is exactly wrong either, because each measured a different thing.
The number a study reports depends almost entirely on the mix of queries in its sample:
- Brand queries send most clicks to one result. A dataset heavy with brand searches reports a very high position 1 CTR.
- Generic informational queries spread clicks across the page, and increasingly lose them to answers shown directly in the results.
- Local queries put a map above the organic results, so organic position 1 is really the fourth thing on the screen.
- Shopping queries put product listings above everything.
"Position 3" is not one thing. It is a different amount of screen, and a different amount of competition, on every one of those result pages.
The mistake this causes
The practical damage is not that people quote the wrong percentage. It is that they compare their site to a stranger's dataset, conclude they have a CTR problem, and rewrite title tags that were fine. Or worse, the reverse: their site-wide CTR looks healthy against the benchmark, so they stop looking.
That second failure is worth spelling out, because it happened to us.
In a recent 28 day window our own Search Console reported a site-wide CTR that would look respectable in any benchmark table. Nearly every one of those clicks came from people searching for our brand name. Filtered to non-brand queries, the click count was zero. The aggregate number was arithmetically true and completely useless: it described people who already knew us, and told us nothing about whether we could reach anyone who did not.
Any benchmark that mixes brand and non-brand queries will hide the same thing on your site.
Build your own curve instead
This takes about twenty minutes in Search Console and gives you a number that actually applies to your pages.
1. Open Performance and set a long enough window. Use 3 months minimum. Short windows on low-traffic sites produce CTR figures driven by a handful of clicks.
2. Filter out your brand. Add a query filter, set it to "Queries not containing", and enter your brand name. Add a second filter for common misspellings of it. This is the step that matters most, and the step almost everyone skips.
3. Split by device. Phone and desktop result pages look different enough that combining them averages away the thing you are trying to see. Do the whole exercise twice.
4. Export the Queries tab. You want query, clicks, impressions, and average position.
5. Group by rounded position and weight by impressions. For each position bucket, divide total clicks by total impressions. Do not average the per-query CTR values: a query with four impressions would count as much as one with forty thousand.
6. Throw away thin buckets. If a position bucket has fewer than a few hundred impressions, it is noise. Leave it out rather than plotting it.
What comes out is your CTR curve: what a given position is actually worth on your site, for the queries you are actually trying to win.
Reading the result
Compare individual pages against your own curve, not against anyone else's.
A page sitting well below your curve for its position is worth investigating. Usually it is one of:
- The title answers a different question than the query it ranks for. This is the most common cause and the most fixable.
- The result is being pushed down the page by features that the position number does not capture. Search the query yourself and look at the actual page.
- The query is ambiguous and most of the impressions are people looking for something else entirely. No title will fix that; the page is ranking for the wrong meaning.
A page well above your curve is also information. Whatever its title is doing, do it elsewhere.
What about clicks as a ranking signal
Documents and testimony from the United States antitrust case against Google described a system called NavBoost that uses aggregated click data as an input to ranking. So click behaviour is used, and that is now a matter of public record rather than speculation.
What that record does not contain is a figure for how much it is worth on your query, in your market, at your position. Anyone who gives you one is guessing. Treat CTR as a metric worth improving because more clicks are better than fewer clicks, and treat any ranking effect as a bonus you cannot forecast.
We sell SERP click services, so take our view on that with the appropriate caution: the honest version is that influencing a click signal is a lever with genuinely uncertain returns, not a ranking button.
The short version
- Published CTR tables measure query mixes that are not yours.
- Filter brand queries out before you believe any CTR number about your own site.
- Weight by impressions, split by device, and ignore thin buckets.
- Compare pages to your own curve, and investigate the outliers in both directions.
Frequently asked questions
What is the average CTR for position 1 on Google? There is no single number worth quoting. Published studies put position 1 anywhere from roughly a fifth to well over half of clicks, because each one measures a different mix of queries. A keyword set full of brand searches produces a high figure; a set of generic informational queries produces a low one. The only average that tells you anything about your site is the one you calculate from your own Search Console data.
Does CTR affect Google rankings? Testimony and exhibits in the United States antitrust case against Google described a system called NavBoost that uses aggregated click data as a ranking input. That confirms click behaviour is used. It does not tell you how much weight it carries for your query, and nobody outside Google can tell you that.
What is a good CTR in Google Search Console? Better than your own average for that position, on that device, for that type of query. Comparing your CTR against a table built from somebody else’s dataset produces a number you cannot act on.
