Last weekend, I had the pleasure of presenting my first ever session at an (un)conference called MeasureCamp. Specifically, MeasureCamp Milan. I mean - if you're going to test material for the first time, better do it outside your home country, right?
So here I am, with KPIs for AI (Search) Optimization. And on how to actually define success. Let's get into it.
It's actually a very relevant question. And I'm going to be honest before we get anywhere near KPIs.
I'll give you 2 case studies.
One from a well-established B2C e-commerce client, operating across 4+ central European markets, with zero focus on AI optimization:
People on LinkedIn say it's around 1%.
Spoiler: it's even worse than that.
Even for a stronger case - a 25+ year old B2C/B2B SaaS in photography with solid E-E-A-T - the numbers aren't jaw-dropping. Product web sits at 0.008% AI traffic. Their content magazines do better (0.2-1.2%, EN performs stronger), but we're still talking ~70 sessions from AI per month.
So yeah. AI is not a traffic channel. Not yet.
Because AI Overviews are already eating your clicks - even without sending you any back.
Ahrefs re-ran their study on 300K keywords using December 2025 GSC data. CTR impact when an AI Overview is present:
For every 100 clicks you used to earn at position #1, Google now keeps 58. Search is becoming zero-click.
And it gets more interesting. Lily Ray analyzed 11 sites hit by Google's unconfirmed January 2026 algorithm update. All 11 lost AI citations alongside organic drops. Average AI citation drop was -22.5%, ChatGPT dropped -27.8%, Google AI Mode -23.8%.
The one outlier? Perplexity - actually showed citation growth despite organic drops. Though in general, Perplexity is pretty "special" in how it behaves - it runs on Brave Search pipeline and diverges from the rest quite consistently.
The takeaway is straightforward: your organic SEO strategy is your AI visibility strategy. They're not separate things. And investing in tactics that hurt organic will hurt AI presence too.
Before KPIs, let's make sure your setup isn't lying to you.
GA4: two quick sanity checks
Check 1 - Unassigned Traffic. Go to Reports > Acquisition > Traffic Acquisition, filter by Channel = Unassigned. As a rough signal, anything over 10% is worth investigating - though this really depends on your setup. S2S tracking vs client-side behaves very differently, but I've seen one of my own sites sitting near 30% unassigned so you can tell that is certainly a problem. There's no hard line, just pay attention to it.
Check 2 - Key Events & Conversion Rate. Just make sure you have them set-up correctly.
These can be done in 5 minutes but save up to 5 days of headache.
Setting up an AI Channel Group in GA4
Julius Fedorovicius from Analytics Mania has a solid guide on this. The idea: create a custom channel group called "Artificial Intelligence" using a regex of known AI referrer domains. The regex covers everything from ChatGPT and Gemini to Perplexity, Claude, Grok, and a few dozen more - I shared the full version during the talk.
You can download the full regex and the full presentation here: bmateas.com/kpis-aio-measurecamp-milan-2026.pptx
As you may've seen in the data above, I shared with the Google AIO numbers. This is not something you track by default. Moreover, you can track which parts of your website Google actually used! The best tutorial I've found on setting it up properly is Marek Lecián´s guide on GA4 + AI Mode measurement.
I structure AIO/SEO KPIs in three tiers. It simplifies the understanding and gives you an actual framework for better thinking.
Tier 1 - Primary OKRs. "Are we growing?" Organic Users and Organic Conversions. Organic Users is the north star. Conversions matter too - but not just as a growth metric. A drop in conversion rate can tell you something completely different, like a design issue, a landing page that doesn't match intent, or a new audience segment that behaves differently. So watch both, but for different reasons. And as always: CR must not drop as traffic grows.
Also worth noting - smaller markets tend to show higher relative growth rates. Less competition, lower baseline, more room to move.
Tier 2 - Growth Key Results. "Why (or why not)?" Checked quarterly. Content KRs and Ranking KRs.
For content, I track: keyword coverage % (share of target KW pool with a ranking URL), topic cluster completion (pillar + supporting pages per cluster), new/optimized URLs per quarter (minimum 20/Q as output accountability), and E-E-A-T signals on new content - author bio, citations, expert review. Technical baseline is Core Web Vitals 100% green and index coverage at least 95%.
For rankings, the targets roughly look like this over two years: +30-40% keywords in Top 10, +30-50% priority keywords in Top 10, and featured snippets going from 10+ to 50+.
One important note on CTR: track it, don't target it. AI Overviews and new content additions both independently deflate CTR. Use it as context, not a hard KR.
Tier 3 - Monitoring. "Early warnings?"
This is where AI citation tracking comes in - as a diagnostic layer, not a primary KPI.
The metric: citation rate = cites per AI answer, per platform. My benchmarks based on Peec AI's citation research and Lily Ray/Amsive research:
Prompt intent matters here too. Informational and commercial intent together account for roughly 10-11% of URLs but capture 37% of all citations. Transactional and navigational are spread thin. So if you want AI visibility ROI, prioritize BOFU and comparison content.
Warning signals worth watching:
Secondary: track % share of organic traffic for context, and don't sleep on Bing Webmaster Tools - the AI performance tab is surprisingly useful.
1. Keyword analysis should become a core part of your AIO strategy. Not just for ranking - but for understanding what intent you're actually covering and where AI-visible gaps are. Quarterly keyword audits aren't optional anymore.
2. Affiliate and comparison posts are more important than ever. BOFU and commercial-intent content is disproportionately cited by AI tools. If you're not producing it, someone else is getting cited instead of you.
3. Think of your site as a brand and an ecosystem, not just a collection of pages. Find your E-E-A-T angle and commit to it. AI citation models reward consistent expertise signals, not one-off content sprints.
I got this question at MeasureCamp and it's a good one. If you're a startup with limited content, what do you do?
My answer: obviously, brands with bigger presence have it a lot easier. But I would say: generate as much content as possible. And ideally try to catch somewhat viral and trendy keywords. In general though, a knowledge base is a pretty cheap win - from my experience it costs around 20 cents per LP article with AI.
But the work is also changing. It's no longer about one finalized version of a page. It's becoming more iterative - you publish, track which parts of the site are getting cited or performing, and optimize from there. Some pages will punch above their weight, others won't. That feedback loop is the new workflow.
If you need help building your knowledge base in your brand's voice and style - reach out. It's something I can help with.
AI isn't a traffic channel yet. But it's already reshaping how clicks are distributed. And the sites that will win AI visibility in 18 months are building organic authority right now.
The KPI framework isn't complicated: grow organic users, understand why (content + rankings), and monitor citation signals as an early warning system. Everything else is noise.
Thanks for reading - and if you were at MeasureCamp Milan, thank you for coming. It was a blast for a first talk. Had so many great convos on different topics. Stay updated, as I will come back to some of them in the near future.