Updated
Organic CTR (click-through rate) is the share of times your page is shown in Google’s organic results that someone clicks it: clicks divided by impressions, as Search Console reports it. It falls steeply with position, but by how much depends on the search, the result features above you and whether people searched your brand. So instead of comparing your pages with a published CTR curve, build the curve from your own Search Console data and look for pages clearly below it.
Why a fixed CTR-by-position curve misleads
Several studies publish average organic CTR by position, such as Advanced Web Ranking’s CTR tool and Backlinko’s CTR guide. They are useful for a sense of shape, and they disagree with each other, because an average over millions of searches hides the things that decide your click rate:
- Kind of search. Brand searches and searches for one business’s name get very high click rates at position 1. Broad “[service] near me” searches spread clicks across maps, ads and many results.
- Result features. A map pack, ads, images or a “People also ask” box above you push your result down the screen even when your position number is the same.
- AI Overviews. When an AI summary appears, people click fewer links. Pew Research found that users in its panel clicked a traditional result link in 8% of visits with an AI summary and 15% without one (March 2025 browsing data). In Search Console, an AI Overview takes one position and every link in it gets that position (Google’s note).
- Device and country. Mobile screens show fewer results above the fold.
- How the curve was measured. Some studies use rank trackers, some use Search Console. Search Console’s position is an average that can mix 3rd and 12th place for the same query.
Compare a directory’s category page with an industry average and it will look bad, because local searches show a map pack first. Compare it with your other category pages at the same position and you find the ones that really underperform.
Build your own expected CTR per position
You need query-level data for the last 3 months (long enough for volume, short enough that the results haven’t changed much). Then:
- Export queries with clicks, impressions and position. Use the Search Analytics API, the Looker Studio connector or the bulk export, because the Performance report stops at 1,000 rows (Google’s note on limits).
- Tag each query by kind. At the least: brand and non-brand (how to split them). On a directory, also business-name searches, “[service] in [city]”, “near me” and questions.
- Round each query’s position to a whole number and keep 1 to 20.
- For each kind of search and position, add up clicks and impressions, then divide. Total clicks over total impressions, not the average of each query’s CTR, so a query with 5 impressions doesn’t count as much as one with 5,000.
- Drop thin buckets. DirectoryLens keeps a position only when it has at least 100 impressions, and gives a kind of search its own curve only when at least 5 positions qualify; otherwise it uses the curve for all searches.
- Smooth it so it never rises with position. Noise can make position 7 look better than position 6. Where that happens, merge the two buckets (an impression-weighted average) until the curve only goes down. DirectoryLens uses isotonic regression, which does exactly this.
The result is a table: for non-brand “[service] in [city]” searches, position 3 earns 9%, position 4 earns 6%, and so on, on your site, for your kind of results.
Find pages under the curve (high impressions, low CTR)
“High impressions, low CTR” is a common Search Console filter, but impressions alone aren’t the test: a page at position 15 should have a low click rate. The test is whether a page’s click rate is below what its position earns, by more than chance.
For a search at a given position, the expected click rate is e from your curve. With n impressions, normal chance variation is about √(e × (1 − e) ÷ n). Count a gap only when the actual click rate is at least 2 of those below e:
| Search at position 4, expected 6% | Impressions | Clicks | Click rate | Standard errors below | Real gap? |
|---|---|---|---|---|---|
| Small search | 150 | 5 | 3.3% | 1.4 | No, could be chance |
| Bigger search | 1,500 | 50 | 3.3% | 4.3 | Yes |
Worked example with made-up numbers. Same click rate, different verdicts: the bigger search has enough impressions to rule out chance.
To judge a whole page, add up the expected clicks of each of its searches (impressions × expected CTR for that search’s own kind and position) and compare with the page’s actual clicks. A page that ranks 1st for its own name and 15th for everything else isn’t then judged by one blended position. DirectoryLens flags a page when it has at least 100 impressions in the top 20 and its clicks are at least 2 standard errors below the expected total.
How to improve organic CTR
Titles
The title link is most of what people see. Make it match the words people searched, put the specific part first (the service and the city, the business name), and keep it short enough not to be cut off; Google truncates titles to fit the device width (title link guidelines). If Google rewrites your title, it is often because the title is stuffed, vague or the same across many pages; Google may use your H1, other headings or link text instead.
Snippets
Google builds the snippet from page content, and uses your meta description when it describes the page better for that search (Google’s snippet guidelines). Write a unique description per page or per template with real details: how many listings, the areas covered, what makes the page useful. Boilerplate descriptions repeated across hundreds of pages get ignored.
Structured data
Structured data can make a page eligible for rich results, such as breadcrumbs instead of a raw URL (Google’s structured data introduction). It doesn’t guarantee them, and markup must match what is visible on the page.
Things a title can’t fix
- A map pack, ads or an AI Overview above you. That lowers clicks for every result below; your curve for that kind of search already accounts for it.
- A mismatch between the search and the page. If people searching “plumber salary” see your plumber directory, a better title won’t make them click.
- Brand searches where another site (a review site, a social profile) ranks above you for your own name.
Before you roll a new title template out to hundreds of pages, test it on half of them; see SEO A/B testing.
On a directory
A directory’s click rates differ by page type more than most sites’. Listing pages get high click rates for searches that name the business and low ones for generic searches. Category-in-city pages compete with the map pack. City hubs show up for broad searches with many results. Build curves per kind of search, then group the pages under the curve by page type. If most of them are category-in-city pages, look at that title template first: does it lead with the service and the city? Does it say how many businesses are listed? Is the state included where city names repeat across states?
These are also striking distance keywords when they rank 5 to 20, and the quick wins guide puts both checks in order.
How DirectoryLens does it
Quick wins > Click rate check fits your site’s click rate at each position for each kind of search over the last 90 days, smoothed so it never rises with position, with the brand terms you set in Site settings counted as their own kind. It then shows:
- Pages getting fewer visits than their ranking predicts: each page’s visits next to the visits its searches and positions predict, for pages shown at least 100 times in the top 20 with a gap too big to be chance.
- Searches with a click rate well below normal: searches shown at least 50 times and at least 2 standard errors below the curve, with estimated missed visits.
- Title patterns: whether titles with a listing count, the current year, “best” or more than 60 characters get more or fewer visits than their position predicts, within each page type. Shown only when the difference holds after correcting for the number of patterns tested, and at least 10 pages and 500 impressions sit on each side. It is a hint; check a change with Title tests.