SEO ·Explanation

Landing page performance: find pages worth improving

Read landing page performance by query, country and device: pages that earn their impressions, under-click at a strong position or rank for the wrong query.

Dana has last week’s query list; landing page performance is the next question. Dana leads SEO at Marlowe, a mid-size clothing brand with a handful of its own shops and an online store, and the question on her Monday list is blunt: of all the pages pulling organic impressions, which ones are actually cashing them in for clicks, and which are sitting on visibility they waste? A property total will never answer that. Neither will a page’s own average, because one page ranks for dozens of queries across markets and devices, and the average describes none of them. This piece is about reading organic search at the page grain: which pages earn their impressions, which hold a strong position but leave the click on the table, and which rank for a term they were never built to satisfy.

Marlowe is a fictional worked example. Its figures illustrate the reading method; they are not customer results or independent benchmarks.

What is landing page performance, and why read it at the page grain?

Landing page performance is how well an individual page earns organic impressions and converts them into clicks for the queries it ranks on. It is the step that comes after the query read: you know which terms drive the visibility, now you find the pages standing behind them and judge each one on its own numbers. The reason to read it at the page grain, and split further by country and device, is that visibility is wildly uneven. Ahrefs studied a billion pages and found that 90.6% of them get no organic search traffic from Google at all (Ahrefs). The pages that do earn traffic carry the whole estate, so knowing exactly which ones, and why the rest don’t, is the entire job.

landing page performance := how well a single page earns organic impressions and turns them into clicks for the queries it ranks on, read at the page grain and split by country and device rather than as one property or one page-level average.

Organic search does awareness work: it puts a page in front of a searcher and earns the click that hands them to the on-site funnel. For how that stage sits in the wider picture, see the pillar’s funnel section.

Read the page, not the property

A page-level average is two averages stacked on top of each other, and both of them lie. The first blend is by query: a single page ranks for a spread of search terms, a few strong ones and a long tail of position-twenty impressions nobody will ever click, and the page’s headline click-through rate mixes them into a number that fits none of them. The second blend is by country and device, and it is the one most reports skip. Click-through at a given position is not a constant. It shifts by device and by market, so a page can look healthy on desktop in its home country and be close to invisible on mobile abroad, with the property average sitting comfortably in between and hiding both.

That is not a hunch. Both Search Console and GA4 expose device as a first-class breakdown, and GA4 reports every traffic and behavior metric by device by default, precisely because the same position earns a different share of clicks and the same landing page converts at a different rate on a phone than on a desktop. The platforms segment by device because a blended device number hides more than it shows. Read a page as one number and you average a strong desktop story with a weak mobile one and act on neither. The grain is not a nice-to-have refinement: it is the difference between finding the problem and papering over it.

So the sequence is: take the page, drop to country, then to device, and only then read impressions, clicks, click-through and position together. One caveat rides along with all of it. These figures come from Search Console, which reports average position as an average that hides the distribution, lags two to three days, and samples the long tail, so read position as a trend and pair it with click-through before you act.

The three page-level stories

Once you read landing page performance at the page grain, almost every page tells one of three stories. Naming which one you are looking at is what turns a table into a decision.

The story The signature The fix, and what it is not
Earning its impressions click-through in line with its position, on the queries it was built for Leave it alone, or fund more of it. Not a problem to solve.
Strong position, weak click good average position, click-through well below what that position should pull Rewrite the title and snippet. Not a ranking problem, do not build links.
Ranking for what it cannot serve impressions on a valuable query the page does not actually answer Build or point the right page at the intent. Not a snippet tweak.

The earning page needs no defending, only feeding. If a page pulls the click-through its position deserves, the work is to give it more of the same: more depth, more internal links, more coverage of the intent it already serves well. The top three organic positions take 68.7% of all clicks (Sistrix), so a page earning cleanly in that zone is an asset to extend, not audit.

Strong position, weak click: the snippet fix

The strong-position-weak-click page is the highest-return fix in the channel, and the most misread. The ranking is already won. The searcher sees the result and scrolls past, which is a title-and-snippet failure, not a content or link one. Backlinko’s analysis of four million results puts the number-one organic result at a 27.6% click-through rate (Backlinko), so a page holding position two or three at a fraction of that is leaving clicks on the table that no amount of link-building would recover. Google’s own guidance is that unique, descriptive title links and meta descriptions are what earn the click from the results page (Google Search Central), and Google primarily draws snippets from page content and may use the meta description (Google Search Central). Rewrite the snippet to match the query intent and the click comes back at the rank you already hold.

The page ranking for what it cannot serve is a trap dressed as a win. It shows impressions, sometimes a lot, on a term worth real money, so a property-level glance calls it a success. Drill it to its queries and the mismatch appears: a thin explainer collecting impressions on a transactional query, or a style article surfacing for a buying term it was never written to close. Google is explicit that thin, low-value pages made to rank rather than to help are the pages it aims to demote (Google Search Central). Worse, when two of your own pages surface for the same valuable query, they split the signal and cannibalize each other, a self-inflicted wound Semrush flags as one of the more common causes of pages stalling below the position they could reach (Semrush). The fix is never a better title. It is a page that actually answers the query, or a redirect to the one that does.

A landing page performance drilldown, by country and device

Here is the landing page performance read the way Dana actually runs it. She takes three of Marlowe’s pages, drops each to country, then to device, and reads the four numbers together. All figures are illustrative.

Landing page › country › device Impressions Clicks Click-through Avg position
/outerwear/wool-overcoats 161,000 8,690 5.4% 6.1
› UK › desktop 42,000 3,360 8.0% 4.2
› UK › mobile 88,000 4,400 5.0% 6.8
› DE › mobile 31,000 930 3.0% 9.1
/guides/how-to-care-for-a-wool-coat 66,000 580 0.9% 2.8
› UK › desktop 26,000 260 1.0% 2.6
› UK › mobile 40,000 320 0.8% 3.1
/blog/what-is-merino-wool 54,000 210 0.4% 11.4
› UK › mobile 39,000 150 0.4% 10.9
› DE › mobile 15,000 60 0.4% 12.7

Three pages, three different stories, and the page totals alone would have hidden two of them.

/outerwear/wool-overcoats is earning its impressions, but read the split before you relax. Desktop in the UK pulls a healthy 8.0% at position four, while the same page on German mobile sits at position nine with a 3.0% click-through, dragging the blended page average to 5.4%: a strong home market subsidizing a weak overseas one that the page total hid.

Whether the German-mobile position is worth fixing depends entirely on your business, and that is the whole point of reading by geography and device. Geography is about where your audience actually lives and which markets you can serve. If Germany is a market Marlowe targets, that position is the fix; if it is not, the honest move is to filter Germany out so the numbers reflect the demand you can actually win, not noise from a place you do not ship to. Device is the other half of the same lens: where your audience is more active, mobile-first in one market and desktop-heavy in another, so you read the device your buyers actually use rather than a blend of all of them. Scope to the country and device that carry your business case, then read the chain inside that scope.

/guides/how-to-care-for-a-wool-coat is the snippet shortlist, top of the list. It holds position 2.8, genuinely strong, and converts under 1% of its impressions. Nothing about its ranking is broken. The title and meta description are not pulling the searcher who already sees the result near the top of the page. One rewrite here is worth more than a quarter of link-building on a page that ranks worse, because the expensive part, the rank, is already paid for.

/blog/what-is-merino-wool is ranking for what it cannot serve, and the page average almost buries it under a shrug: 0.4% click-through at position eleven looks like a page that simply ranks poorly. Drill it to the queries feeding those impressions and the real story lands:

/blog/what-is-merino-wool › query Impressions Clicks Avg position
› “what is merino wool” 12,000 180 3.4
› “merino wool overcoat” 29,000 20 12.1
› “merino wool coat sale” 13,000 10 14.8

The page answers its informational query fine at position three. But it is also collecting 42,000 impressions on two transactional buying terms it has no way to satisfy, ranking at the bottom of page two because a fabric-explainer is not what a searcher typing “coat sale” wants. That demand belongs on a product or category page. The fix is not a new title on the blog post. It is pointing the wool-overcoat product page at those buying terms, and letting the explainer keep the informational query it earns cleanly.

Reading a product line, not a page list: clusters

Dana’s real question is bigger than three pages. She wants to read every category she sells, at every level of depth, without hand-tagging a thousand-row export each month into product lines. A flat page list cannot answer “how is the wool line doing across all its pages,” because the line is spread across a product page, a care guide, a lookbook and three blog posts, and no column in the export knows they belong together.

In Maya’s interface she builds that grouping as a hierarchy of landing-page clusters from simple text-matching rules, no code, up to ten levels deep. Level one groups pages by type, the homepage in each language, the blog, the shop. Level two by product category, shirts, trousers, outerwear. Level three by specific line, linen, oxford, wool. And on down if she needs it. Each level previews live as she tweaks the rule and refreshes on its own as new data loads into the model, so a cluster she defines once never goes stale and never needs rebuilding.

flowchart TD
  Shop["Marlowe shop · level 1"]
  Blog["Blog · level 1"]
  Shop --> Shirts["Shirts · level 2"]
  Shop --> Outerwear["Outerwear · level 2"]
  Shirts --> Linen["Linen · level 3"]
  Outerwear --> Wool["Wool coats · level 3"]
  Outerwear --> Rain["Rain shells · level 3"]

Now she reads the same figures at any altitude, the whole shop, then outerwear, then just the wool line, as one consistent hierarchy. All figures illustrative.

Cluster Impressions vs last quarter Click-through Avg position What Dana reads
Outerwear (all pages) +9% steady mid healthy in aggregate, read the lines
Outerwear › Wool coats +6% rising strong earning cleanly, feed it more depth
Outerwear › Rain shells +31% falling slipping impressions up, clicks not following, snippet or intent gap

The category looked fine as a total. One line inside it, rain shells, is gaining impressions while its click-through falls, which is exactly the strong-position-weak-click signature emerging at the cluster level, and only reading at the line grain surfaces it before a quarter of traffic quietly leaks away.

How Maya turns scattered data into answers

Most SEO tools stop at description. They hand you impressions and click-through by page and, on a good day, a chart of it. The work that matters lives higher: diagnosis (why this page under-clicks), decision (rewrite the snippet, or build the page the query needs), outcome (did the fix land). Maya is one implementation of the idea that an analytics layer should climb that value pyramid for you, on definitions it cannot quietly reinterpret.

The foundation comes first, and the order matters:

  1. A marketing data warehouse unifies every platform’s raw exports in one cloud database, reconciled so numbers across sources become comparable.
  2. A governed Power BI semantic layer defines every metric once, so a ratio like click-through rate is rebuilt from its parts, total clicks over total impressions, at whatever grain you read it. Read the page, the country, the device or the cluster and the number stays correct instead of averaging averages into a figure that is true of nothing.
  3. A conversational layer, a Model Context Protocol (MCP) server, lets an AI assistant (Claude, ChatGPT, a Slack bot) query that governed layer in plain language. You steer with prompts; the enforced definitions decide what the numbers are.

Prompts steer, the semantic layer enforces, and that separation is why a click-through rate read at the device grain still reconciles to the page and the property. It is not a chat box on a mess.

For reading pages, that stack buys three things. The grain holds. Because click-through is a governed ratio, dropping from page to country to device never corrupts the number, so the strong-desktop, weak-mobile split shows up honestly. Clusters you define, ten levels deep. The hierarchy above is built once from text-matching rules and read at any altitude, so Dana asks about the wool line, not a list of URLs. The store sits in the same model. Because Marlowe’s e-commerce orders live alongside the search data, a page earning impressions on a buying term can be read next to what that term is actually worth in sales.

Without Maya: an analyst exports Search Console by page, exports it again by page-and-country, tries to line the two up in a spreadsheet, eyeballs the position column, hand-tags pages into product lines, and a day later you learn that one page holds position three at a 1% click-through. With Maya: you ask “Which pages hold a strong position but under-click, split by country and device?” Maya reads impressions, clicks, click-through and position at the page grain, separates the snippet problems from the ranking problems, and hands back the shortlist with the diagnosis. Minutes, not a day.

The questions a marketer actually asks Maya:

  • “Which pages earn the most impressions but the fewest clicks?” The snippet-rewrite shortlist, at the page grain.
  • “Show me this page split by country and device.” Where a healthy average hides a weak market or device.
  • “What queries is this page actually ranking for?” The intent-match check that finds a page collecting impressions it cannot satisfy.
  • “How is the wool line doing across all its pages?” The landing-page cluster read, at any altitude.

One honest caveat: these numbers inherit Search Console’s limits, the two-to-three-day lag and the long-tail sampling, and the assistant flags that rather than paper over it. Maya does the assembly, the diagnosis and the guardrails. You write the title, build the page, and decide which line to fund.

The bottom line

Read landing page performance at the page, never the property, and never even the page’s own average without splitting it. Three stories cover almost everything you will find: the page earning its impressions, which you feed; the page holding a strong position at a weak click-through, which needs a new title and snippet, not links; and the page ranking for a query it cannot answer, which needs the right page built or pointed at the intent, not a cosmetic tweak. The grain is what separates them, because a page average blends a dozen queries and stacks a strong market on a weak one, and the blend flatters you into fixing the thing that was already working.

So the question is not whether your pages are getting impressions. It is whether the pages getting them are the ones that can turn them into clicks, and whether you would even see the ones that can’t from where you are currently reading.

Glossary

  • Landing page performance: how well a single page earns organic impressions and turns them into clicks for the queries it ranks on, read at the page grain and split by country and device.
  • Organic impressions: the number of times a page appeared in unpaid search results. The reach signal.
  • Click-through rate: organic clicks divided by organic impressions. The transition from visibility to the click. A strong position at a weak click-through is a title-and-snippet problem, not a ranking one.
  • Average ranking position: the average ranking a page holds, weighted by impressions. An average that hides the distribution, so read it as a trend.
  • Read at the grain: reading impressions, click-through and position at the level where they mean something (page, then country, then device, then query), instead of a property or page average that blends behaviors that share nothing.
  • Title link and snippet: the clickable title and the descriptive text a result shows on the results page. Google draws them from the page’s title element and meta description, and they are what earn the click at a rank you already hold.
  • Thin content: a low-value page that ranks for a query it does not actually answer. Detected by drilling a page to its queries and finding a mismatch of intent.
  • Keyword cannibalization: two or more of your own pages competing for the same query, splitting the signal so neither ranks as well as one focused page would.
  • Landing page cluster: a set of landing pages grouped by a shared text-matching rule (a section, a template, a product line), read together at any level of a hierarchy rather than page by page. Built in Maya’s UI, nested up to ten levels deep.

Where this fits: the insight roadmap

This is step 3 of the SEO insight roadmap, landing-page performance, the read that comes right after the query grain: once you know which terms drive the impressions, you find the pages behind them and judge each on whether it earns, under-clicks, or ranks for what it cannot serve.



Ask this about your own account

Connect your sources, then ask Maya the question this piece just answered, on your own numbers.

Start with your store