Written by Omri Kfir, DigiAssets · Last updated: September 2, 2026
The Generative AI Performance report is a new tab inside the Performance report in Google Search Console. It shows how many times your pages appeared in AI Overviews, AI Mode and AI answers in Discover. Right now it only has impressions: no clicks, no CTR, no queries. Its value: it reveals which of your pages Google already pulls into AI answers.
On August 31, Google officially opened the new, much-discussed Search Console report to every site in the world. Yes, yours too.
Google had been trickling the report out to site owners gradually, and some of you got it more than a month ago. But officially, everyone should see it now.
So if this is new to you, let me sort this report out for you in five minutes: what it means, what conclusions to draw from it, and more. Plus a small bonus at the bottom: a breakdown of a page structure that actually gets citations in this report (the way we do it at the agency, and it works for us).
What is the Generative AI Performance report in Search Console?
The report is called Generative AI performance, and it sits inside the regular Performance report.
Basically, it shows how many times your pages appeared inside Google’s AI answers: AI Overviews, AI Mode, and AI answers in Discover. If you haven’t yet looked at what a blog that gets traffic from AI engines looks like, now is the time to compare your data with the report.
You can segment it by page, country, device and date.
Until now it was open only to some sites… mostly in the UK. Since August 31 it’s open to everyone.
To get there, open Search Console, click Performance, and you’ll find a separate tab there.
If you don’t see it, that usually means your site doesn’t have enough appearances in answers generated by AI.
That’s an answer too. Not a very pleasant one.
Why are there no clicks in the Search Console AI report?
Right now, this report has no clicks.
No CTR and no average position. No queries.
There are impressions. That’s it.
Some sites are already reporting that query data is starting to show up, but it’s very, very early days.
Two things will happen in the coming weeks.
Some people will open it, see a big number, and feel great.
Others will see a small number and panic.
Both are wrong.
An impression in this report means one thing only: your page was among the sources Google pulled when it built an answer.
It doesn’t mean anyone saw your brand name…
It doesn’t mean anyone clicked…
And it definitely doesn’t mean any money came in.
So why bother opening it at all?
Because it gives you a list you didn’t have until now: which pages on your site Google is already willing to pull into an AI answer.
That’s your list of candidates.
And that’s the entire value of the report.
By the way,
in Hebrew, AI answer coverage is still lower than in English. Factor that in for Israeli sites.
If your numbers look small compared to what you’ve read on American blogs, that’s fine and normal. Look at your own trend over time, not at the absolute number.
How do you structure a page that gets cited in AI Mode?
First, the single most important thing to understand is that AI Mode doesn’t choose pages.
It chooses paragraphs.
It takes the user’s question, breaks it into several sub-questions, and then pulls from each source the passage that best answers one sub-question.
In other words, your page doesn’t compete as a page. Every paragraph on it competes separately.
Get it? That changes how you think about a “page” as the thing that ranks.
From now on, it’s chunks of content within a page. Or within the site as a whole.
From that, you get five actions. If you want to go deeper into the small changes with the biggest impact, we covered them in detail in 5 minutes, 5 changes: how to get AI engines to recommend your business.
| The rule | What to actually do | Why it matters |
|---|---|---|
| Subheadings as real questions | Phrase every subheading as a question in the customer’s words, for example “How much will it cost to install a mini-central AC in a five-room apartment?” and not “Our advantages” | The heading is a label the model uses to decide whether the section is relevant to it |
| A direct answer in three lines | Answer the question right under the heading, with no “In today’s world” or “Before we answer, it’s important to understand” | A paragraph that makes sense without its surrounding context is a paragraph that can be cited |
| One paragraph = one claim | Break every paragraph into two to four lines, with only one claim per paragraph | A paragraph with several claims has no clear boundaries, so it can’t be cited |
| Numbers in a table | Turn prices, specs, delivery times and comparisons into a table instead of free text | A table is an organized data structure that the model can pull whole, with no effort |
| Last-updated date and author name on the page | Show a last-updated date and the author’s name on the page itself, and actually update the content every time you change the date | Google compares content versions over time, and a match between the date and a real update builds trust |
Every subheading is a real question, plain and simple, in the customer’s words and not ChatGPT’s
Not “Our advantages.” Instead: “How much will it cost to install a mini-central AC in a five-room apartment?”
Understand that a heading isn’t decoration. It’s a label, the kind of label the model uses to decide whether this section is relevant to it… Write that down.
Three lines under the heading answer the question. Period.
I’ve talked about this several times and I’ll keep repeating it: stop with the “In today’s world” openers. Or “Before we answer, it’s important to understand.”
Get good at giving a direct answer in the very first line.
It’s fine to use Claude to write the content, but make sure you fill it with real context about the subject,
and give yourself a simple test: cut the paragraph out of the page and show it to someone without the context. If they understand it on its own, it can be cited. If not, it can’t.
One paragraph = one claim, not twenty
A 300-word paragraph with four claims and two conclusions is a paragraph that can’t be cited. It has no boundaries.
Break it up: two to four lines per paragraph. Train the AI that writes your content to do this too.
Want to include numbers? Put them in a table
Prices, specs, delivery times, comparisons. It’s fine to mention them in the text in general terms, when the point is general.
But know this: a table is a structure the model pulls whole, and easily. It’s organized data. It’s always better to turn your stats and numbers into a table, and if that’s hard, just ask Claude.
A last-updated date and the name of whoever wrote it… remember?
Not somewhere in the site’s general area. On the page itself.
And don’t touch the date without actually updating the content… that doesn’t fly anymore. It worked back in 2015.
Today there’s correlation and there are content versions, and Google compares one version to the next across the dates you claim for your changes. So when the real update and the date line up, it pays off. Speaking from experience.
What happened when we restructured a client’s page?
A client of ours in cybersecurity had one core service page: well written, ranking nicely, and zero presence in AI answers.
We didn’t rewrite a single word… really. We only restructured the page according to the five rules above. And it simply took off. This isn’t the first time a proper restructure has brought a fast jump in results: we shared a similar case in how a client jumped 75% in clicks within a month.
3 weeks later, the site shows up in AI answers and in the citations Google pulls, with thousands of impressions.
Hard to believe how simple it is.
What is google.com/goto in search results?
Not sure if you’ve noticed, but Google has started routing links in its search results through its own internal redirect, google.com/goto, before they reach the destination site. Google has officially confirmed this and says it’s a technical measure against abuse of search results.
For now there isn’t much to conclude from it, except this: if you see goto redirects in the URL in your direct traffic, that’s most likely search traffic.
It doesn’t affect rankings or traffic.
It does make scraping search results harder for third-party tools.
Which means the rank trackers we all pay for may show slightly odd data in the coming weeks. So hold tight and don’t panic. If you’ve already built a dashboard that shows how much traffic comes from ChatGPT and other models, you’ll be able to tell a real drop from temporary third-party noise.
So, as always: track in Search Console, not so much in third-party tools.
FAQ
What is the Generative AI Performance report in Search Console?
It’s a new tab inside the Performance report in Search Console that shows how many times your pages appeared in Google’s AI answers: AI Overviews, AI Mode and AI answers in Discover. You can segment it by page, country, device and date.
Why are there no clicks or CTR in this report?
Right now the report shows impressions only, with no CTR, average position or queries. Some sites already report seeing query data, but it’s still very early.
If I have an impression in the report, does that mean someone saw my brand?
No. An impression only means your page was among the sources Google pulled when it built an answer. It doesn’t mean anyone saw your brand name, clicked, or that any money came in.
How does AI Mode choose what to cite from a page?
AI Mode doesn’t choose whole pages, it chooses paragraphs. It breaks the user’s question into sub-questions and pulls from each source the passage that best answers one sub-question, so every paragraph on the page competes separately.
What is the google.com/goto that shows up in search results?
It’s a technical redirect Google adds to links in its search results as an anti-abuse measure. It doesn’t affect rankings or traffic, but it can make third-party rank trackers show unusual data in the short term.
How do I know if my page structure is ready to be cited in AI Mode?
Test it simply: cut a paragraph out of the page and show it to someone without the surrounding context. If they understand it on its own, it can be cited. If not, break it up according to the five rules.
