Beyond the Click: The Metrics That Matter in AI Search
In my last post, I made the case that AI search is changing the funnel, not eliminating it. Buyers still move through the same stages, they're just doing it inside a single AI conversation instead of a string of Google searches. If you haven't read that one, start there. This post picks up where it left off.
In this post I want to dig into what's actually happening to your metrics, which channels are quietly dying, and what to tell your team when they ask, "But is any of this actually working?"
The Data Behind the Shift
Let's start with the number that's hardest to argue with. SparkToro's research, built on Similarweb clickstream data, found that 68% of U.S. Google searches ended without a single click in early 2026. This is up from about 60% just two years earlier. A separate randomized study (Agarwal and Sen, posted to SSRN in April 2026) found something more specific: when researchers used a browser extension to randomly show or hide AI overviews for the same pool of searchers, the group that saw them clicked through to websites 38% less often, with no measurable difference in how satisfied they said they were with their search.
That's the traffic side, what about the business side? 10Fold's 2026 report, The Visibility Reset (based on a survey of 400 B2B tech marketing leaders) found that 52% of them now rank AI-generated search as their most effective content distribution channel, ahead of traditional SEO, for the first time. As 10Fold CEO Susan Thomas put it, the marketers who win won't be the ones publishing the most content. They'll be the ones creating something "worth finding, citing and believing."
I'd push back gently on one claim floating around right now: that AI-referred traffic converts dramatically better than everything else. Most multipliers you see (5x, 10x, 20x) have no named source. If someone hands you a conversion multiplier with no named source and no methodology behind it, treat it as marketing, not data.
What I can point to with confidence comes from Conductor's 2026 AEO/GEO Benchmarks Report, an analysis of 3.3 billion sessions across more than 13,000 enterprise domains: AI referral traffic is still tiny, about 1% of total site traffic on average, and 87% of what exists comes from ChatGPT alone. Though AI search isn't about to become your biggest traffic channel, it is becoming something else — a visibility layer that happens before a click, and sometimes instead of one.
What This Means for Your Metrics
If traffic isn't the win condition anymore, what is? Here are a few things worth tracking:
Citation share. Are you showing up when someone asks an AI model a question in your category? This doesn't live in Google Analytics — you have to test it yourself, by running the same set of prompts against ChatGPT, Perplexity, and Google's AI Overviews on a regular cadence and logging whether you show up, and whose URL gets linked when you do.
Direct and branded search. When someone gets your name from an AI summary, they often don't click a link inside that conversation. They likely open a new tab and type your company name, or navigate straight to your pricing page. In Google Analytics, this shows up as "direct," which is exactly why it's so easy to miss.
Self-reported attribution. Add an open-text field to your demo request form: "How did you hear about us?" Skip the dropdown — a dropdown forces someone who used Perplexity to pick "Google" because it's the closest option available. You'll be surprised how often the honest answer names a specific AI tool and a specific prompt.
One technical check before you invest in any of this: make sure your robots.txt and your CDN (Cloudflare especially) aren't blocking the bots these engines use to crawl your site. OAI-SearchBot, GPTBot, and PerplexityBot are the ones to look for. If they can't crawl your content, none of the above matters.
The Channels That Are Actually Shifting
The middle of the distribution stack is thinning out. Generic "what is X" explainer content is fully absorbed into AI summaries now — nobody needs to click through to read what the AI already told them. Gated PDFs have a similar problem: if your best data lives behind an email form, a crawler can't read it, and the citation goes to whoever published the same stat in an open blog post instead.
What's growing instead sits at two extremes. On one end: public, ungated, machine-readable assets. That includes documentation, benchmark pages, open pricing calculators — the kind of thing a crawler can actually parse and cite. On the other end: private, high-trust human spaces like Slack and Discord communities, small newsletters, and live conversations. This is the kind of trust an AI model can't fabricate on your behalf.
If I had to put a limited budget somewhere, it'd be these three things, in order:
Your own public data in the form of fact sheets, benchmarks, open docs, or similar
The communities where your buyers are already asking questions, spaces where genuinely helping is what earns citations later (Reddit, GitHub, Stack Overflow, or similar)
One durable, direct-to-human channel you actually own
Stop Chasing the Algorithm
Here's the trap I see marketers falling into: treating every update from ChatGPT or Google as something to react to. Don't. While the specific mechanics change every few months, the underlying thing these models reward doesn't.
The foundational research here is the GEO paper out of Princeton, Georgia Tech, and IIT Delhi. It's from late 2023, so think of it as the study that named the field and tested which techniques move the needle, not a snapshot of the latest algorithm. It tested nine content optimization methods against real generative-engine responses. The ones that worked — adding citations, direct quotations, and statistics — lifted visibility by 30–40%. The one that's core to old-school SEO, keyword stuffing, did essentially nothing.
More recent data points the same direction. Ahrefs' March 2026 analysis of 863,000 keywords and 4 million AI Overview citations found that only 38% of cited pages ranked in Google's organic top 10, down from 76% just seven months earlier. Ranking well used to more or less guarantee a citation. Now it's closer to a coin flip.
That's the argument, old study and new data both pointing the same way. Formatting a page for machines doesn't mean writing worse content for humans. Rather it means adding things a model can't invent on its own: your own numbers, your own case studies, a counterintuitive opinion you can actually back up.
Bring Sales Into the Loop
The self-reported attribution field on your website only catches people who fill out a form. A lot of your best signal is sitting in sales calls instead, going unrecorded. I've written before about why this needs to be a two-way street, not marketing handing leads over the wall. This is a small, concrete version of that: have your SDRs ask a single question on discovery calls. "When you were researching this, did an AI tool point you our way, or did someone recommend us?" You'll start hearing specific prompts back, "I asked ChatGPT for alternatives to X and your benchmark came up." That's more useful than any dashboard, because it tells you exactly which page or stat is doing the work.
Where This Leaves You
None of this means content stops mattering. It means the content that matters now is the kind an AI model literally cannot generate without you — your data, your case studies, your specific point of view. So as you work on new content. make sure what you do publish is something worth citing.
This is Part 2 of a series on content strategy in the age of AI search. Read Part 1, AI Search Is Changing the Funnel, for the framework this post builds on.
