Pew Data Shows AI Summaries Cut Click Rates in Half

Google users click a traditional search link on just 8% of visits when an AI summary appears, versus 15% without one. That finding, from Pew Research Center's analysis of a month of real browsing data, is the hard number behind a new argument from PAN Communications: B2B marketers are measuring the wrong things, and their best buyers are the ones who never click.

The essay, by PAN Vice President of Marketing Lauren Hill, makes a direct case. Marketers traded impressions for clicks over the past decade, she writes, not because clicks told the whole story, but because they were countable and easy to defend in budget meetings. AI search has now broken that bargain. Buyers get answers inside ChatGPT, Claude and Perplexity without visiting sources, and Pew found users clicked sources cited within AI summaries in only 1% of visits.

The pressure compounds from another direction. Privacy expectations are rising, and the conversations that shape B2B decisions keep moving into Slack groups, DMs and niche communities no analytics tool can see — the so-called dark funnel.

What does the data say about silent audiences?

PAN's own research pushes back on the instinct to treat untraceable exposure as waste. The agency's 2026 Brand Experience Report analyzed more than 20 B2B brands and found that the fastest revenue growers often saw conversation volume decline. In 40% of brands studied, reach grew faster than mention volume. More people were seeing these brands than talking about them.

Judged on activity alone, that looks like a problem. Judged on revenue, it looks very different. Lurkers, Hill argues, are frequently further along the buying process — reading, comparing and forming opinions without tipping their hand to sales.

Which leading indicators should CMOs track?

Hill proposes three measures that correlate with impact without forcing clicks:

• AI citation appearance. Content surfacing in AI answers within days of publishing signals durable organic visibility. PAN's C-Suite Signals study analyzed more than 66,000 citations from AI answers to C-suite research questions and found LinkedIn was the top-cited source for eight of nine executive roles.
• Scroll depth and time on page in the first few hundred visits. A strong early ratio — 60% or more of readers scrolling through a long-form piece — predicts the asset will keep earning attention, links and citations after launch.
• Dark funnel tracking. Reverse-IP tools show which companies visit without filling a form. Ten visits from ideal-customer-profile accounts can mean more than a thousand from people who will never buy.

One caution on AI citations: don't treat a single check as a result. PAN re-ran identical prompts across ChatGPT, Claude, Gemini and Perplexity, and 50.4% returned a substantially different answer in the same session. For CMO queries specifically, it was 61%. Hill's advice: track citation appearance over time and read it as a trend.

Why attention and action are different jobs?

PAN's Brand Experience Report tested identical claims with the same CMO audience. The unpolished, human versions delivered 10x the click-through rate and held attention 2.3x longer in-feed than polished versions — yet every landing page click came from the polished variants.

The implication for measurement stacks is blunt. A stack that only rewards action may starve the content that actually earns attention. Much of what shapes AI answers, PAN notes, is content marketing and communications teams already produce.

Hill also warns against forcing lurkers to engage before they're ready, which usually annoys them and can cost a buyer already on their way to you. For ready prospects, she recommends personalized retargeting or a relevant email offer. For everyone else, keep showing up with credible content where they already look.

Her closing point lands on the CFO problem. Most measurement stacks were built to count clicks because clicks were countable, not because they told the whole story. The right leading indicators differ for every brand, and the risk is rotating away from what works in favor of what's easiest to measure.