Search Console export analyser

The write-up comes first

The tool opens with sentences, not tiles. Each one is a claim, the evidence under it, and the caveat that makes it safe to act on — ranked so the thing worth doing first is first, and tagged as a caution, an opportunity or context. There is a button to copy the lot as text.

That ordering is the point. A dashboard hands you five numbers and leaves the reading to you, which is the step everyone skips and the step that actually takes judgement. Where the file is too thin to support a claim, the write-up says so in as many words rather than going quiet and letting an empty table read as good news.

What the chart is showing you

A Search Console export has thousands of rows and about four things in it worth acting on. A table hides all of them. The plot doesn't.

Every bubble is one query. Its position on the horizontal axis is where you rank; its height is the click-through you actually get; its size is how many people saw it. The dashed line is the click-through that position normally earns.

Everything sitting below the line is a query that ranks and doesn't get clicked. That is the single most actionable thing in the file, and it is invisible in a spreadsheet because it only exists as a relationship between three columns.

Everything else in the file

Over the range. Impressions, clicks and average position as three stacked panels on one date axis, rather than one plot with two y-scales. Two measures of different size on a shared axis invent a correlation that isn't in the data; the alignment of the two scales is arbitrary and the reader can't see that it is. Days with no impressions have no position either, so the position panel breaks rather than drawing a fall from first place that never happened.

Who is seeing it. Device and country, which the export ships and most readers never open. The device split is often the sharpest thing in the file — Google ranks the same page differently on mobile and desktop, and a wide gap usually means the two are being shown for different queries entirely.

The tables sort, filter and expand. A thousand-row export is not served by showing you the first fifteen of anything.

The lists underneath

Closest to page one. Queries ranking 8–20 with real impressions. They are already close, so a title rewrite, an extra section, or a couple of internal links moves them further than a new page would. The bars show what each would add in clicks at position five, so you can start with the biggest one.

Ranking well, barely clicked. Top-ten positions earning less than half the click-through of that position. That is almost never a ranking problem — it is a title and description problem, and it is the fastest fix in search because you keep the position you already have.

Branded vs non-branded. Put your brand terms in and it splits the export. If most of your clicks are people typing your name, your search traffic is a reflection of your other marketing rather than a channel of its own.

Getting the file

In Search Console, open Performance → Search results, set the date range you care about, then Export → Download CSV. You get a zip. Drop the zip in whole — the tool opens it here in the tab and reads the tables it needs.

A single CSV out of it works too. The tool reads the header itself, so it also copes with exports that carry an extra title row above the columns, or that use slightly different column names.

Why the percentages don't add up

Queries.csv is not your traffic. Google withholds any query rare enough to identify the person who typed it, and on a small site that is most of them — the export this tool was built against showed ten queries covering 42% of the impressions the same export's chart reported.

So the tool reads Chart.csv as well and tells you what share of your real impressions the query analysis is actually describing. Anything drawn from the queries table is a statement about that share, not about your site. Pages.csv has no such suppression, which is why the pages list is often the more complete half of the file.

Nothing is uploaded

The file is read in your browser and parsed in memory — the zip is opened here too, with the decompression the browser already ships. It is never sent anywhere, there is no account, and nothing persists once you close the tab. Open the network panel and watch nothing happen.

That is not a gimmick for this one tool. It is how the main product works: the analysis produces a process that can run wherever the data already is, so confidential files stay on the machine that already has them.

The honest limits

The expected click-through curve is a blended average, not your industry's. Treat the under-clicked list as a starting point rather than a verdict — a query where you rank third but everyone wants a video, or where an AI overview answers it above you, will look like it is underperforming when it isn't.

Position is also an average across every impression in the range. A query that sat at 4 for a week and 30 for a month averages to something misleading. Shorter ranges give sharper numbers.

And click-through needs a sample before it means anything. Fifty impressions with no clicks is a finding; five is a Tuesday. The under-clicked list holds itself to that floor and tells you when nothing clears it, rather than showing you an empty table and letting you read it as good news.