SEO results are the measurable effects of work on organic search: how often the website appears on Google, how many clicks and visits it gets, and how many enquiries and sales come from them. Since 2025 every one of these numbers needs a correction: Google has changed the way it counts impressions, positions, AI answers and visits from assistants, and a report that ignores this compares apples with pears.
It sounds like the easy part of SEO.
Open Search Console, look at the line, if it goes up you're fine.
Not quite?
Not quite. The line can go down while the website improves, and up while enquiries fall. We see it every month in the reports people send us to review. The mistakes are almost always these:
- comparing the current year with a period before September 2025, when everyone's impressions dropped for a technical reason
- adding branded traffic to generic traffic, so the SEO work takes the credit for brand awareness (or the blame when it falls)
- counting appearances in AI Overviews as if they were clicks
- stopping at visits without getting to enquiries
Why are we the ones writing about it?
Because we measure results on real projects. Macropix, a Milan manufacturer of custom LED screens, has received around 7,000 enquiries in seven years and today holds more than a hundred keywords in the top three positions on Google Italy (the Macropix case study tells the story with the numbers).
On an Italian dental sector website we work on, by contrast, average position improved by 64% in eight months and clicks by only 36%: the same project, two "results" that say different things. The data is in our article on how long it takes to see SEO results, so we won't repeat it here.
Here you'll find:
- the map of the five SEO results, with the right tool and the distortion to correct for each
- the September 2025 change that skewed year-on-year comparisons until a few weeks ago
- our own check on 278 Italian keywords: AI Overviews appear on 13% of the results pages for one project and on 92% for the other
- six steps to set up measurement in an afternoon
Let's start.
Five SEO results, five tools (and what has skewed them since 2025)
SEO results fall into five families: visibility, position, clicks, visits and enquiries. Each has a different source and, since 2025, at least one factor that muddies it.
This is the table we use as the starting point on every project.
| Result | Metric | Where to read it | What has distorted it since 2025 | Correction |
|---|---|---|---|---|
| Visibility | Impressions | Search Console, Performance report | Removal of the num=100 parameter (September 2025), AI impressions inside the total | Comparisons only after 15 September 2025, separate AI report |
| Position | Average position, positions per keyword | Search Console, rank tracker | An AI Overview takes a single position for all the links inside it | A rank tracker that records SERP features |
| Clicks | Clicks, CTR | Search Console | Lower CTR where an AI Overview appears, 46.77% of clicks with no visible query | Read clicks by page, not just by query |
| Visits | Organic sessions | GA4 | Visits from ChatGPT and Gemini moved to the AI Assistant channel (May 2026) | Look at Organic Search and AI Assistant together |
| Enquiries | Leads, quote requests, orders | GA4, CRM | Last-click attribution | Source of the enquiry recorded in the CRM |
The column almost nobody fills in is the fourth.
The two guides on page one of Google UK for "seo results" (Backlinko, updated in February 2026, and Semrush) list the metrics well. Neither mentions the September 2025 change, the Search Console brand filter or GA4's AI Assistant channel. Three facts that move the numbers more than any optimisation.
The definition of each metric, and why it matters, is in our guide to SEO KPIs. Here we deal with how to read them without being misled.
The result that counts comes last (around 7,000 enquiries in seven years, for Macropix)
The SEO result a business cares about is the enquiry or the sale; all the others are indicators that come before it. Impressions and positions tell you whether the work is heading the right way. They don't pay invoices.
Macropix shows this well. In 2020 the website was in position 88 for "monitor pubblicitario" (advertising monitor); today it's in position 2. In July 2026 the domain held 25.55% share of voice on the "ledwall" cluster, ahead of Amazon.
Lovely. But the number that counts for the business is another one: enquiries, around seven thousand since the project began, on €420,000 invested across SEO and advertising.
So positions don't matter?
They matter the way a thermometer matters for a fever. They tell you whether something is moving, not what it's worth.
A concrete example: below position ten, traffic is close to zero. Going from 60 to 25 is half the work and doesn't change a line of turnover. That's why in our reports share of voice and enquiries go at the top, positions underneath.
And if you have to explain to a board which part of those enquiries is down to SEO and which would have arrived anyway, the method is different: you'll find it in our guide on how to prove the value of SEO.
September 2025: impressions down with nothing changed (87.7% of websites in a US sample)
Between 12 and 14 September 2025 Google stopped supporting the num=100 parameter, which let rank trackers and scrapers load a hundred results on a single page. Those hundred-result pages generated impressions in Search Console even for websites in position 70 or 90, which no human being saw. When the parameter went, those impressions went with it.
The change is documented by LOCOMOTIVE, a US agency. Its head of technical SEO, Tyler Gargula, analysed 319 Search Console properties:
- 87.7% lost impressions
- 77.6% lost unique keywords they had been ranking for
It's one US agency's sample, with no breakdown by country. No equivalent UK figure has been published. The mechanism applies to any property, though: according to LOCOMOTIVE's reading, part of the impressions came from hundred-result pages loaded by tools, not by people.
And average position?
It improved on its own. Take away the appearances in position 80 and the average goes up, without a single page moving.
That's why a report that in September 2026 compares "year on year" with August 2025 shows impressions falling and positions rising, and neither is true. How to read average position without falling for it is explained in our guide to keyword rankings.
And now?
Now there's some good news the industry rarely mentions: from mid-September 2026 the year-on-year comparison is clean again, because both periods fall after the change. For a whole year it wasn't.
Count brand and non-brand separately (the filter has existed since 20 November 2025)
Branded searches are the ones that contain the name of the company or its products: they measure awareness, not SEO work. Someone searching "Macropix" finds Macropix anyway. Someone searching "ledwall" doesn't.
On 20 November 2025 Google introduced the branded queries filter in Search Console, extended to all eligible websites on 11 March 2026. It isn't based on a list of words to exclude: the classification is done by an internal AI-assisted system that recognises the brand in every language, with typos, and the website's unique products. Google itself warns that some queries may be misclassified.
How much does it weigh, in practice?
Ahrefs removed branded searches from the count on its own blog: clicks dropped from 1.37 million to 686,000. About half the traffic was down to the name, not to SEO.
The share varies a lot from one business to another, which is why it has to be measured, not estimated. Either way, the number to put in the report as the SEO result is non-brand, and brand should be read as a separate indicator: we explain why in our guide to SEO branding.
One detail to know before you trust the total: according to Ahrefs' study of 887,534 properties and around 22 billion clicks, 46.77% of Search Console clicks (April 2025) come from anonymised queries, meaning with no visible text. The brand filter only works on the half you can see. How to recover the rest is in our guide to Google Search Console.
AI Overviews are a result that doesn't bring clicks (on 278 Italian keywords: from 13% to 92%)
Appearing in an AI Overview is an SEO result, but it has to be counted separately from clicks: Google records it as an impression, gives it a single position for all its links, and the click rarely comes. The Search Console documentation says so: an AI Overview takes a single position, and a click on a link inside it counts as a normal click.
In the UK, AI Overviews have been live since 15 August 2024, AI Mode since 28 July 2025.
How much do they cost in clicks? The most solid measurements are from the US:
- Pew Research Center, 900 adults in the United States, March 2025: with an AI summary on the page, people clicked a traditional result in 8% of visits, against 15% without. 1% clicked inside the summary (Pew study)
- Ahrefs, 300,000 keywords, desktop CTR from Search Console, December 2023 against December 2025: the first position loses 58% of its CTR where an AI Overview appears (Ahrefs update)
And on our own projects?
A national average says little, because how often AI Overviews appear varies enormously from one sector to another. We measured it on 8 October 2026 on the keywords we track with Keyword.com for two B2B projects, both on Google Italy. It's a check on the Italian market, but the gap between the two types of query is the point, and you can repeat it on your own keywords in the UK.
| Project | Type of keywords tracked | Results pages with an AI Overview | Results pages with "People also ask" | Website citations in AI Overviews |
|---|---|---|---|---|
| Project A | Product and category (manufacturer) | 13% | 87% | between 1 and 3 from July to mid-September, 0 in October |
| Project B | Informational and local (B2B services) | 92% | 100% | 7 on 5 October, 4 a month earlier |
| Total | 278 keywords | 53.6% | 94% |
Three things this table shows that averages hide:
- On product queries the AI Overview rarely appears (13%): there the SEO result to measure is still the click
- On informational queries it almost always appears (92%): there the result to measure is the citation, not just the position
- Citations jump around from one day to the next: on project B they went from 6 to 2 between 21 and 22 September 2026, and from 6 to 1 between 28 and 29 September
A citation read on one day isn't a result. An average over four weeks is.
Since this summer there's also an official report. On 3 June 2026 Google launched generative AI feature reports in Search Console, live for all websites from 31 August 2026: they show impressions, pages, countries, devices and dates. Not clicks. How to read that report alongside the other data is in our analysis of the impact of AI Overviews.
Your impressions are falling and you don't know if AI is to blame?
On project B, 92% of the tracked keywords show an AI Overview, and the website's citations change from one day to the next. Let's look together at which of your queries have ended up under an AI answer and which clicks you're really losing.
Book a strategy consultationWhere visits from AI end up (GA4's AI Assistant channel, since 13 May 2026)
Since 13 May 2026 Google Analytics 4 has a default channel called AI Assistant, which collects visits from assistants such as ChatGPT, Gemini and Claude. It's announced in the GA4 release notes.
Before that, those visits ended up in Referral, or in Direct if the link was copied.
The point that causes confusion is another one. According to the official channel definitions, visits from AI Overviews and AI Mode stay in Organic Search, because they come from Google. Visits from ChatGPT don't.
In practice: if your SEO report only looks at Organic Search, since May it has lost a slice of traffic that until April was counted elsewhere, and which partly comes from the same work on content. The rule we use is simple: Organic Search and AI Assistant go on the same row, with the total next to them.
How the credit given to SEO changes depending on the model you choose, with three GA4 properties compared, is in our guide to attribution models.
How to set up SEO results measurement in an afternoon (six steps)
To measure SEO results reliably you need Search Console, GA4, a rank tracker and your CRM, set up in that order. There are six steps, and only the third needs a paid tool.
- Mark external changes in Search Console. Since 17 November 2025 the Performance report accepts custom annotations of up to 120 characters: add 12 September 2025 (num=100), the UK launch of AI Mode (28 July 2025) and every major change to the website. Anyone with access to the property can see them
- Turn on the brand filter and save two views: brand and non-brand. The SEO result is the second
- Track the keywords that matter with a rank tracker that records SERP features, AI Overviews included. How to choose the tool is in our guide to AI Overview tools
- In GA4, compare Organic Search and AI Assistant over the same period, with key conversions next to them
- Record the source of every enquiry in your CRM, even by hand. It's the only place where the SEO result becomes turnover
- Choose the right comparisons: month on month for corrections, year on year only with both periods after 15 September 2025
If you want to turn all this into a monthly document for a client or a manager, the five-block structure is in our guide to the SEO report. If the numbers are falling and you can't see why, start with the procedure for a traffic drop.
And if you'd rather have the measurement set up by someone who does it every day, you'll find how we work on our SEO consultancy page.
Frequently asked questions
No. Google says so in its guide for people hiring an SEO: no one can guarantee a #1 ranking. What can be guaranteed is the work, the delivery dates and honest measurement of the results, not the ranking.
It depends on the comparison period and the type of query. A drop that started between 12 and 14 September 2025 is due to the removal of the num=100 parameter, not to the website. A drop on informational queries with stable clicks is often an AI Overview that has appeared above the results.
No, not on its own. It is an average weighted by impressions: it changes when the queries you appear for change, even if no page moves. For the keywords that matter you need a rank tracker that checks them one by one.
Yes. But it has to be measured separately: Search Console records it as an impression, gives it a single position and clicks are few. The figure to follow is how often you are cited over four weeks, not on a single day.
It depends on where the website starts from. Usually at three months you see impressions and first positions, at six months clicks, and after twelve months a steady flow of enquiries. The timings for each phase, measured on Italian projects, are in our guide to how long SEO takes.
It depends on the metric. Positions and AI citations every week, clicks and visits every month, enquiries and the comparison with the previous year every quarter. Looking at everything every day only produces anxiety.
So which SEO results should you look at? (in six lines)
Enquiries and sales at the top, everything else underneath.
Non-brand separate from brand.
Year on year only after 15 September 2025.
AI citations separately, on four-week averages.
Organic Search and AI Assistant on the same row.
An annotation in Search Console for every change.
One last thing the industry rarely says. The most misunderstood SEO result of 2026 isn't a drop, it's a rise: the average position that improves after September 2025 without anything changing on the website. Before you celebrate a number, ask yourself whether the website changed or the way of counting it.
If you'd like to redo the count on your own data, get in touch. Speak soon!
