How Search Console calculates it
A link to your site appearing in results someone sees is an impression. Them clicking it is a click. CTR is one divided by the other.
The complication is aggregation. Sitewide CTR averages across every query, page, country and device, so it moves without anything changing. Pick up impressions on a broad query you rank 40th for and it falls, even though every page performs exactly as before.
So the number is only useful filtered: one query, one page, India versus everywhere else, mobile versus desktop. Unfiltered CTR is a mood indicator, not a metric.
The curve everyone quotes is from 2019
Ask anyone what CTR position one gets and you'll hear about 31%. That comes from Backlinko's 2019 analysis of five million Google searches. Good study — of a SERP with no AI Overview, a smaller ad block and fewer features. That page has changed shape considerably since.
Published curves are useful as priors. They are not measurements of your SERP. Here's how we use them when forecasting, stated as assumptions rather than facts.
- The ranges are wide on purpose. Anyone quoting CTR to one decimal place for your query is quoting someone else's dataset, and query type moves this more than position does.
| Organic position | Rough planning CTR, no AI Overview | What we assume with one |
|---|---|---|
| 1 | Roughly 25–32% | Meaningfully lower — plan nearer 20% |
| 2–3 | Roughly 10–18% | Cut by around a third |
| 4–6 | Roughly 5–9% | Cut by around a third |
| 7–10 | Roughly 2–4% | Often close to noise |
| 11+ | Under 1% | Effectively zero either way |
What AI Overviews did to it
Two 2025 analyses are worth knowing together, because they measured different things and got different answers.
Pew Research Center studied real browsing behaviour and found users clicked through to a result on roughly half as many visits when an AI-generated summary was present. Ahrefs, studying a large keyword set, reported top-ranking pages seeing around a third lower CTR on queries carrying an AI Overview.
Neither is India-specific and neither is your site. They agree on direction and rough magnitude: the drop is real, big enough to matter to a forecast, and hardest on informational queries.
The practical version is arithmetic. Take the 2019 figure of roughly 31%, take a third off, and you land near 20% — which is why we forecast nearer 20% than 30% wherever an AI Overview appears. A derived assumption, not a measured 2026 statistic, and we'd rather label it than dress it up.
Branded versus non-branded, and why the split matters
Someone searching your company name has already decided; they're using Google as a navigation bar. Branded CTR runs many times higher than non-branded, so a healthy sitewide number is often brand demand in disguise.
Split it: filter queries containing your brand name, note that CTR, then invert the filter with "Queries not containing". The second number reflects how well your pages compete against strangers.
It also explains the pattern that panics founders in month four: sitewide CTR drops while everything improves. New non-branded impressions arrive at low positions and dilute the branded average. The business is growing; the blended metric is falling. Both are true.
Improving CTR, and when not to bother
CTR is one of the few SEO fixes with a same-week payoff: you're editing what appears in the result rather than moving the result. Work the queries where you rank 3–10 and underperform.
- When not to bother: position 11+, or low CTR on queries you shouldn't rank for. Nothing you write changes a result nobody scrolls to, and irrelevant impressions cost you nothing real.
- Rewrite the [title tag](/glossary/title-tag) around the promise, not the keyword. Google rewrites titles it finds unhelpful, so a stuffed one often isn't what's shown.
- Write the meta description as ad copy. It isn't a ranking factor; it's the only sentence you fully control in the result.
- Add a number, price or date where it's genuinely true. Specificity separates your result from four near-identical ones.
- Ship schema that changes the result's appearance — FAQ, breadcrumb, review where legitimately applicable.
- Match the query's language. If people search in Hinglish and your title is formal English, the result reads as something else.