SEO analytics is the work of turning organic search data into decisions: which pages to rewrite, which to leave alone, when to admit that a strategy isn't working. It isn't the dashboard and it isn't the list of metrics, which come later.
In the UK, 25% of businesses that handle digitised data say they analyse it to generate new insights, according to the UK Business Data Survey 2026 published by the Department for Science, Innovation and Technology. Among large businesses the share reaches 69%, among small businesses it stops at 35%, and among micro businesses at 26%. These are numbers about who analyses data, not about who decides anything with it. The gap between the two is the subject of this article.
Start from the question, not the metric
The quickest way to lose two hours is to open GA4 and look around. Organic data answers a few precise questions well. For each one there is one source that counts and two that mislead. The table below is the one we use internally before opening any report: on the left the client's question, on the right where it gets answered, how late the data arrives and where people most often go wrong.
| Question | Source that answers it | Data delay | Typical mistake |
|---|---|---|---|
| Do people find me when they search for what I sell? | Search Console, Performance report, non-brand queries | 2-3 days | reading average position as the position the user sees |
| Has this page improved since the change? | Search Console on the single URL, 28 days against the previous 28 | 2-3 days for the data, 4-8 weeks for the verdict | judging after a week |
| Does organic traffic bring enquiries? | GA4 with conversions reconciled against the CRM | 24-48 hours | counting conversions that never reached the sales team |
| Am I losing ground or has the market moved? | rank tracker on a keyword cluster plus Search Console impressions | weekly | mistaking seasonality for a drop in position |
| Do I appear in AI answers? | Search Console report on generative AI features | daily | reading impressions as if they were clicks |
| What does it earn? | revenue attributed in the CRM, not in GA4 | monthly | stopping at last click |
It works the other way round too: if a question isn't in this table, it probably doesn't need measuring. The word count of a page, keyword density, the score from an audit tool are not data on how people behave, they are a piece of software's opinion. On how to write an objective that can be checked later, we have a dedicated guide to SEO goals.
Three sources, three numbers, and none of them is wrong
Search Console says 15,000 clicks, GA4 says 10,000 organic sessions, the rank tracker says you are fourth while Search Console says eighth. Anyone who has worked with the data for a few years has stopped being surprised; anyone opening their first report thinks something is broken.
Google puts it in writing in the Search Central documentation on Search Console and Analytics: clicks and sessions are calculated differently, so the numbers won't match. Google lists seven causes. They are all mundane: the Analytics tag is missing from some pages, some users reject cookies, Search Console uses Pacific time by default, the two tools attribute differently, Search Console only reports canonical URLs, it includes PDFs and non-HTML content, and Analytics automatically excludes known bot traffic while Search Console doesn't necessarily filter it.
The working rule that follows is a single one: Search Console clicks and GA4 sessions never go on the same line of a chart. They answer different questions. Search Console tells you what happens before someone lands on the site, GA4 what happens afterwards. We have explained in detail how much data Search Console really shows and what Google Analytics doesn't let you see. On how to lay all this out without building a chart that lies, there is our guide to the SEO dashboard.
The threshold below which a number says nothing
This part is missing from almost every SEO report we see during audits, and in my view it is the mistake that costs the most money. A small change on little data isn't a result, it's noise. Interpreting it leads to work on pages that were perfectly fine.
The three rules we apply on our projects, with no claim that they are universal laws:
- 28-day windows compared with the previous 28, not week against week: B2B traffic has a strong weekly cycle, and on the B2B projects we follow Monday is worth twice as much as Saturday
- for positions, look at the 30-day moving average, never the value for the day
- below 100 impressions in 28 days a query isn't interpreted, it stays on the list and you wait
Then there is the opposite case, a real drop mistaken for noise. When traffic falls, the order in which you check the causes changes how long the diagnosis takes: we have written it up step by step in the article on how to find the cause of a traffic drop.
The click is no longer the unit of measurement
Until three years ago, SEO analytics rested on one assumption: someone who searches and finds a useful answer clicks. Recent data says that assumption holds for a shrinking share of searches. Three studies, none of them on UK samples. Worth saying before the numbers.
The Pew Research Center tracked the real behaviour of 900 US adults in March 2025, for a total of 68,879 Google searches, 12,593 of which had an AI summary on the page. When the AI Overview was there, users clicked a traditional result on 8% of visits; when it wasn't, on 15%. The link inside the summary was clicked in 1% of cases. And 26% of sessions with an AI summary ended there, against 16% without.
Ahrefs measured the same thing from the publisher's side: Ryan Law and Xibeijia Guan compared March 2024 and March 2025 across 300,000 informational keywords, desktop, United States, and found a 34.5% lower CTR for position one when an AI Overview is present.
Across all searches, Rand Fishkin published data from a Similarweb panel on the US market in June 2026: 68.01% of searches in the first four months of 2026 produced no click at all, against 60.45% in 2024. The closest European figure is still the 2024 SparkToro study on the Datos panel, which recorded 59.7% zero-click searches for the European Union and 360 clicks to the open web for every 1,000 searches, against 374 in the United States. Fishkin wrote that separate data for the UK, Italy, Germany, France and Canada would follow later: at the time of writing there is no UK figure for this, and anyone quoting one is estimating it. What zero-click searches are and how much they weigh is covered in our piece on zero-click searches.
What to measure when the click doesn't come
Since 3 June 2026 Google has had a partial answer to this problem. Hillel Maoz and Moshe Samet announced on the Search Central blog performance reports for generative AI features: the site's impressions inside AI Overviews and AI Mode, the pages that appear, countries, devices and granularity down to the hour. The metric missing from the list is clicks, and at announcement the rollout was limited to a subset of sites. Visibility in generated answers can be counted. The effect on traffic can't.
What stays measurable with the tools you already have:
- Impressions in AI features, from the new Search Console report, by page. It tells you whether you are in the pool the model draws from.
- Brand searches, filtering the queries in Search Console that contain the company name. If people read about you in an AI answer without clicking, the trace they leave is a search for your name a few days later.
- Presence as a cited source, checked by hand on the ten or twenty queries that really matter, or with one of the AI visibility tools. On a B2B site, twenty queries checked one by one say more than an aggregate score.
- Direct and referral traffic, read together with organic impressions and not on its own: when an AI answer cites a company, the visit often arrives without a referrer.
We have covered the resulting change of approach on the strategy side in generative engine optimisation and on the numbers side in the impact of AI Overviews on SEO, where ten Italian B2B queries are checked one at a time.
How often to look at what
Half of all useless SEO reports come from a cadence problem: a number that moves every quarter gets looked at every day, and a number that needed checking the same day gets looked at every quarter. This is the grid we keep on our projects, with the column that is usually missing.
| Cadence | What you look at | What you don't look at |
|---|---|---|
| Daily | indexing errors, 404s on converting pages, a vertical drop in impressions | positions, traffic, conversions |
| Weekly | new pages entering the index, emerging queries between positions 8 and 20, anomalies on key pages | ROI, year-on-year comparisons |
| Monthly (28 days) | non-brand clicks and impressions, moving average of positions on the cluster, conversions reconciled with the CRM | individual low-volume queries |
| Quarterly | share of visibility on the cluster against competitors, attributed revenue, the decision to continue or change | anything that moves week by week |
The quarterly row is the one that leads to real decisions, and it is also the only one that needs a comparison with cost. The sum is done as explained in the article on SEO ROI, bearing in mind that an organic position takes months to build, not weeks.
One project read over seven years
Macropix makes custom LED screens in Milan and has passed five thousand installations worldwide. We have worked on the project since 2020. The numbers are public, in the case study on SEO for manufacturing.
The line that best explains how to read an industrial project is "monitor pubblicitario" (advertising monitor): from position 88 to position 2. Position 88 is Google's ninth page, meaning zero traffic and zero useful impressions. A report that looked at the site's total traffic would have seen nothing for months, because a keyword that moves from 88 to 40 still brings zero clicks. The signal was in the impressions and average position of that single query, not in the total.
On the paid side of the same project the maths is quicker: €805 invested, €44.93 per conversion, average CTR 3.59% and average CPC €0.81, with conversions up 84.2% on the previous twelve months and cost per conversion down by €11.77. These numbers can be read within weeks. The same project's nine organic positions took years. The difference in speed between the two channels is why we measure them on different cadences, and why SEO and Google Ads are read together.
Other projects with their starting and finishing numbers are on the projects page, and if the measurement setup is the part you are missing, we can talk about it under SEO services.
Four mistakes we find in almost every report
- Brand and non-brand mixed together. If brand traffic grows because an offline campaign has started, the organic total rises and it looks like SEO's doing. Of the nine keywords that bring Macropix the most qualified leads today, none contains the company name: that is the filter that makes everything else readable.
- Average position for the whole site. It averages queries with 10 impressions and queries with 5,000, and it moves when new queries enter at the bottom of the list. It goes up when you cut content and down when you publish: it says almost nothing about the work done.
- Conversions never reconciled. If GA4 says 40 enquiries and 24 reached the CRM, the problem isn't SEO, it's tracking, and every decision taken on those 40 is a decision taken on nothing.
- Bounce rate used as a quality score. In GA4 it no longer exists as you knew it, and a page that answers a question well can end after thirty seconds without that being a problem. How to read it depends on the type of page, and a product page and an article aren't judged by the same threshold.
If you are deciding which indicators to keep in a monthly report, the short list is in the article on SEO KPIs.
Frequently asked questions
It is the work of collecting, cross-checking and interpreting organic search data to decide where to act on a website. The basic sources are Google Search Console for what happens before the click and Google Analytics 4 for what happens after, with the CRM as the final judge on conversions.
Because they measure different things and calculate them differently. In the Search Central documentation Google lists the causes: Analytics tag missing from some pages, cookie rejection, Pacific time zone used by Search Console, different attribution models, only canonical URLs in Search Console, PDFs and non-HTML content counted, bot filtering active in Analytics and not necessarily in Search Console. There is always a gap: when it is very wide, check the tag implementation and cookie consent first.
Non-brand clicks from Search Console over a 28-day window, the moving average of position on the keyword cluster that brings enquiries, and organic conversions reconciled with the CRM. The others are there to explain these three.
Since 3 June 2026 Search Console has a report for generative AI features showing impressions, pages, countries and devices inside AI Overviews and AI Mode, but not clicks. Alongside it you look at the trend in brand searches, presence as a cited source on a short list of queries checked by hand, and direct traffic read together with organic impressions.
Technical errors and sudden swings in impressions are checked daily, emerging queries weekly, clicks and conversions over 28-day windows, and the decision to continue or change course every quarter. Reading positions every day only produces pointless changes.
One last metric, which isn't in any tool: decision latency, meaning how many days pass between the moment a number is available and the moment someone touches the site because of it. On projects where that number is under a week, the reports are short and ugly. Where it is over a month, the reports are beautiful and nobody reads them. If last month's report didn't change anything, it doesn't need improving: it needs switching off, and rebuilding one question at a time.
