If keyword rankings and monthly organic sessions are your main SEO metrics, you’re working with an incomplete picture.
The reason is simple. A growing share of people get their answer on the results page, or inside ChatGPT, and never click through. Your ranking can hold steady while your traffic falls. Both numbers are accurate. Neither one tells you what’s actually happening.
The good news is that measurement got a lot better in 2026. Two things shipped this year that most reporting templates haven’t caught up with yet. Here’s a framework built around what you can actually measure now, rather than the workarounds people used when there was no real data.
What Changed, With Actual Numbers
This is worth grounding in evidence rather than assertion, because plenty of agencies are selling panic.
The Pew Research Center tracked the real browsing behaviour of 900 US adults across nearly 69,000 Google searches in March 2025. When an AI summary appeared on the results page, users clicked a traditional search result 8% of the time. When no summary appeared, they clicked 15% of the time. That’s roughly half.
The more striking number: users clicked a link inside the AI summary itself in just 1% of visits. And they ended their browsing session entirely on 26% of pages with an AI summary, versus 16% of pages without one.
Pew isn’t the only study. Ahrefs analysed 300,000 keywords and found that AI Overviews correlated with a 34.5% lower click-through rate for the top-ranking page, comparing March 2024 with March 2025. When Ahrefs re-ran the study using December 2025 data, the gap had widened rather than closed. Amsive, looking at 700,000 keywords, found a smaller average drop of around 15%, with much bigger losses in specific situations.
Three caveats worth stating. Google disputes this research and says AI Overview citations produce higher-quality clicks. The Pew panel is a specific slice of US users, so your industry may look different. And the studies use different methods and reach different numbers, which is why the range runs from roughly 15% to well over 50%.
But every independent study points the same direction, and one Amsive finding is worth holding onto: branded queries rarely trigger AI Overviews at all, and when they do, click-through actually improves. That’s a strong argument for the brand-building measures further down this page.
The practical takeaway: being cited inside the answer is worth less traffic than people assume. Traffic alone is a poor measure of whether your content is working.
1. Use Search Console’s Generative AI Report
This is the biggest change, and it replaces a workaround you may still be using.
On 3 June 2026, Google launched dedicated Generative AI performance reports in Search Console. They show your impressions inside AI Overviews, AI Mode, and generative features in Discover, broken down by page, country, device, and date.
Before this, the standard advice was to infer AI visibility from rising impressions and falling click-through rate in the main Performance report. That guesswork is no longer necessary.
Three things to know:
There’s no click or query data. You can see visibility, not what it’s worth. Pair it with your analytics for the traffic side.
It’s a breakout, not new data. AI impressions were always folded into your overall totals, so your headline numbers won’t change.
The rollout started with a subset of UK site owners. If you don’t see it under Performance yet, you’re waiting your turn.
Check one setting while you’re in there. Google added a control that lets sites block their content from AI features. Opting out costs you AI impressions but doesn’t affect normal rankings. If your AI impressions are zero, confirm nobody switched that on by accident.
2. Track AI Referral Traffic Properly
This is the second thing that changed, and it fixes a real reporting gap.
On 13 May 2026, Google Analytics added a native “AI Assistant” channel to GA4’s default channel group. Sessions from recognised AI tools now get classified automatically, with no setup. Before this, a click from a ChatGPT citation landed in Referral or Direct, mixed in with everything else.
It’s useful but incomplete, in three ways that matter:
The platform list is limited and it has changed. Google’s launch post named ChatGPT, Gemini, and Claude. The live documentation has listed a different set since. Check Google’s current help page before you put the list in a report.
Perplexity is the gap people keep hitting. It generally still lands in plain Referral.
Google AI Overview clicks aren’t in it at all. Those keep a google.com referrer, so they count as Organic Search.
The practical fix is to run both: the native channel, plus a custom channel group with a source regex covering the platforms you actually see, ordered above Referral so it catches sessions first.
Also filter for utm_source=chatgpt.com. OpenAI adds that parameter to links automatically, which gives you a clean way to isolate ChatGPT referrals.
And for Bing and Copilot, Bing Webmaster Tools added its own AI performance reporting earlier in 2026.
3. Track Branded Search Volume
Branded search is the number of people searching for your company or product by name. It’s one of the better signals that your wider visibility work is landing, because someone has to have heard of you first.
Use Search Console to monitor impressions and clicks for queries containing your brand name. Set a baseline and watch the trend.
Rising branded search alongside flat organic traffic is a genuinely good sign. It suggests people are encountering you somewhere, then coming back to find you directly.
One honest caveat: it isn’t quite as manipulation-proof as it’s often described. Paid campaigns, PR spikes, and brand name collisions all move the number. It’s a strong signal, not a pure one.
4. Track AI Mentions, Carefully
How often does your brand come up when someone asks an AI tool about your category? Are you described accurately? Are competitors named more often?
You can test this manually. Pick 5 to 10 prompts your buyers would realistically use, run them across ChatGPT, Perplexity, and Gemini, and log what appears.
But you have to interpret it correctly, and this is where most people go wrong. These systems aren’t deterministic. The same prompt returns different sources on different days, in different accounts, and in different countries. Chat memory and personalisation shift results further.
So one test proves nothing. A pattern across many tests over several weeks means something. Always test in a fresh, logged-out session, and record the date and platform every time.
Dedicated AI visibility tools now exist and can automate this. Two notes before you buy one. General brand-monitoring tools like Brand24 and Mention track the web and social media, not the inside of AI answers, so they’re the wrong tool for this job. And Google explicitly warns that no third-party tool has access to its internal ranking or AI systems, so treat any product claiming “internal Google metrics” with suspicion.
5. Segment by Intent, Not Just Volume
Not all traffic loss means the same thing, and total session counts hide that completely.
Split your Search Console data by query intent:
Informational queries are the most likely to trigger AI Overviews. Pew found that question-based searches produced AI summaries 60% of the time, and searches of 10 or more words did so 53% of the time. Expect click-through compression here. It’s a structural shift, not a failure.
Transactional queries still need people to visit a site to buy or enquire. Click-through should hold up better. If it doesn’t, that’s the urgent problem.
Then measure conversion rate inside each segment. That gives you a far more honest read on commercial performance than a single traffic number.
6. Be Careful With Direct Traffic
You’ll see advice telling you to treat rising direct traffic as proof of growing brand strength. Be careful with that one.
Direct traffic is a catch-all bucket for any session whose source couldn’t be identified. That includes bookmarks and typed URLs, but it also includes clicks from email clients, messaging apps, some mobile browsers, and privacy tools that strip referrer data.
Here’s the part that matters for AI: a substantial share of AI referral sessions arrive with no referrer header at all and land in Direct. Estimates vary widely, which itself tells you how murky this is.
So rising direct traffic might mean your brand is getting stronger. It might just mean AI referrals are being misfiled. You can’t tell the two apart from the number alone.
Watch direct traffic as a diagnostic. If it moves sharply with no technical explanation, investigate. Just don’t put it in a board deck as a brand authority KPI.
7. Set Your Own Baseline
There’s no industry benchmark for AI mention frequency or generative share of voice. These measures are too new and too unstable.
So your benchmark is you. Spend a month recording where you stand:
Branded search impressions and clicks
Generative AI report impressions, if you have access yet
AI Assistant channel sessions in GA4
AI mention frequency across your test prompts
Conversion rate by intent segment
That’s your starting line. Set improvement targets from there and review quarterly, not weekly. Short-term movement in AI visibility is mostly noise.
Mistakes to Avoid
Using rank tracking as your headline metric. Rank tracking is still useful for spotting technical problems and sudden drops. It just isn’t a success measure anymore. A page at position three, sitting under an AI Overview, may get a fraction of its old traffic. The rank looks fine. The outcome isn’t.
Still using impression data as a proxy for AI visibility. If you have access to the Generative AI report, use it. The proxy method was always a rough guess.
Reading single AI prompt tests as data. One check on one day in one account tells you almost nothing.
Treating all traffic loss the same. Losing informational traffic to AI Overviews is a structural change. Losing transactional traffic is a fire.
Ignoring branded search entirely. Plenty of businesses spend heavily on content and never look at whether anyone is searching for them by name. Flat branded search alongside good rankings usually means your audience isn’t retaining you. That’s the real problem.
The Underlying Principle
Measure presence, not just position.
The goal is to be a recognised, trusted, frequently cited source across search results, AI answers, and your buyers’ memory. No single number captures that, which is exactly why the old dashboard stopped working.
One more thing. This landscape has changed twice in the last four months. Build your framework to be revised, review it quarterly, and be willing to throw out a metric when a better one becomes available.
Where to Go From Here
For the technical side of AI visibility, see our guides to showing up in ChatGPT answers and answer engine optimisation.
If you’d rather have someone build and run this for you, Digileads’ reporting and analytics services are designed around it. We help businesses set the right baselines, separate real signal from noise, and stop reporting on numbers that no longer mean anything.
