What is a good CTR by position? A measured curve

Search "average CTR by position" and you get a table. The same table, mostly, with the same numbers: 31.7% at position one, 24.7% at two, 18.7% at three, decaying smoothly to nothing by ten.

That table gets used to make decisions. Someone divides their impressions by it, decides a move from eight to four is worth 6,000 clicks a month, and commits a quarter to link building on the strength of it.

Here is the same measurement taken on one real site, over 216 days, on 410,000 non-brand impressions.

Position bandMeasured CTR
1–329.95%
3–410.81%
4–57.57%
5–63.44%
6–83.12%
8–101.47%
10–152.69%
15–301.94%

Two things in that table should stop you.

The curve is not smooth, and that is not an error

Look at positions 8–10 against 10–15. Click-through rate goes UP as the ranking gets worse: 1.47% at eight to ten, 2.69% at ten to fifteen.

Every published curve says that is impossible. On this data it happened, on tens of thousands of impressions per band, over seven months.

The explanation is not that page two beats page one. It is that position is not the only variable, and a band is not a population. The queries a site ranks 10th for are not the same queries it ranks 9th for. If the 10–15 band happens to contain more specific, lower-competition, higher-intent phrasings, the kind where someone knows exactly what they want and clicks the thing that says it, that band converts better despite ranking worse.

Averaging across positions hides the query mix. The curve looks like a law of physics and is actually a summary of one site's particular collection of queries.

The gap between 5 and 6 is bigger than the gap between 1 and 5

The other thing in that table: 7.57% at position 4–5 falls to 3.44% at 5–6. Click-through rate halves across a single position.

Meanwhile 1–3 to 4–5, three positions, costs 29.95% to 7.57%, which is a bigger absolute fall but a far more gradual one per place.

If your page sits at 5.4, moving it to 4.6 is worth more than the position number suggests, because you are crossing the edge of the visible block rather than moving within it. Where those edges sit depends on what the results page looks like for your queries: how many ads, whether there is a featured snippet, how tall the "People also ask" box is. It is not a property of position; it is a property of the layout above you.

Why the borrowed table is the wrong instrument

The industry tables are not lies. They are averages across categories that behave nothing alike, and the averaging is the problem.

Here is how far apart two markets on the same site, the same pages, the same day can be:

MarketAvg positionCTR
Mozambique4.8232.93%
United States12.991.55%

Identical content. The only variable is who else showed up. One curve cannot describe both, and the average of the two describes neither.

The same split appears by device, mobile ranks 1.2 positions better and converts at 1.5× desktop's rate on the same site, and by whether the query contains your brand name. A single number sitting where three populations used to be is a number about nobody.

What to use instead

Build the curve from your own export. You already have the data: Search Console gives you clicks, impressions and average position per query. Group your non-brand queries into position bands, divide clicks by impressions per band, and you have a curve measured on your own queries, your own results pages and your own audience.

It takes about ten minutes, and it is the difference between a forecast and a guess.

Three rules that matter when you do it:

Exclude brand queries first. Someone searching your name clicks you at rates that have nothing to do with position: in this dataset brand queries were 44.1% of all clicks. Leave them in and your curve is mostly a measurement of how many people already knew you.

Ignore any band under about 5,000 impressions. Below that you are reading noise, and noise at the top of a curve is the most dangerous kind because it is the band you will use to price your biggest decision.

Rebuild it every quarter. Results pages change. A curve measured before an AI overview appeared on your main query is describing a page that no longer exists.

The honest limits of the table above

This is one site, in one category, over 216 days, weighted toward markets outside the United States. It is a reference, not a benchmark, useful for sanity-checking whether your own numbers are plausible, useless as a target.

I am publishing it because a measured curve with its provenance stated is more useful than an unsourced one presented as universal, not because this one is more universal.

If your curve looks nothing like this, that is the expected result. The point is to have one.

Questions

What is a good CTR at position 1?

On this dataset, positions 1–3 averaged 29.95%. But "good" depends entirely on the query: an informational query answered by a featured snippet above you can hold position one and convert in single digits, while a navigational query can exceed 60%. Compare against your own curve rather than a published figure.

Why is my CTR higher on page two than page one?

Usually because the two bands contain different queries rather than the same queries at different ranks. Long, specific queries are easier to rank for and often convert better; if those cluster in your lower bands, the band inverts. Check the actual queries in each band before concluding anything about position.

Should I use industry CTR benchmarks at all?

For a rough sanity check, yes. For pricing a decision, no. In this dataset the same pages converted at 32.93% in one market and 1.55% in another, which is a range no single benchmark can span. Use your own export.

How many impressions do I need before a CTR figure means anything?

About 5,000 in a band, as a working floor. Below that a handful of clicks moves the percentage by whole points, and the resulting curve will look convincingly detailed while being mostly random.

Run this on your own data