Google AI Overviews: Metrics, experiments and plays for publishers and business leaders

Has AI made Google Search better, or worse, for publishers and businesses?

Look at the top of a Google results page lately and the familiar list of links is often not the first thing you see. Instead you may get an “AI Overview”: a short, generative answer, sometimes with cited links, sitting above the classic organic results. Google pitches it as a faster, more helpful experience. Publishers see it as a new filter between users and their content.

Both perspectives can be true. Google’s Liz Reid, vice‑president of search, has called this change “the most significant upgrade of the Google Search experience ever.” Google also says, “Total organic click volume from Google Search to websites has been relatively stable year‑over‑year.” Independent researchers, publishers, and vendors report something different at the page and query level. When an AI Overview appears, users are less likely to click through, and some pages show measurable drops in pageviews. That’s not a contradiction so much as a measurement problem, aggregate traffic can stay flat while attention shifts within the web.

What independent studies and vendors are actually finding

“AI Overview” is Google’s product label for the generative summary card that appears above organic links (also related to what Google has called the Search Generative Experience). It’s not universal: an academic audit published on arXiv in 2026 found AI Overviews activating on a meaningful minority of queries, about 13.7% overall and roughly 64.7% of question‑form queries, and used SERP scraping plus difference‑in‑differences comparisons (with Wikipedia pages as a control) to estimate traffic effects for pages that triggered the card. That same audit reported single‑digit to low‑double‑digit declines in pageviews for affected pages.

Pew Research Center (2025) reported lower click‑through rates when an AI summary is present, roughly 8% of visits resulted in clicks when an AI Overview appeared versus about 15% when it did not, and noted citation links inside the AI summary themselves generate only a small share of clicks (around 1% of visits). Those figures come from Pew’s user‑behavior measurements rather than platform logs.

Vendor analyses paint a mixed picture and should be read with vendor caveats. Adobe’s digital insights (2025) found large growth in AI‑sourced visits to U.S. retail sites (about a 1, 200% increase between July 2024 and February 2025) and reported those visitors on average spent more time and converted better. Microsoft Clarity reported higher conversion rates for AI‑assistant visitors in its datasets (roughly three times higher in the contexts it measured). By contrast, Ahrefs (2025) found AI‑referred visitors were on average shallower, fewer pages per session and a slightly higher bounce rate. In short: AI referrals are heterogeneous, commerce users can behave very differently from readers seeking encyclopedic answers.

Why this matters to the bottom line

For many publishers, ad revenue depends on pageviews and session volume. If the search experience satisfies users inside Google’s AI layer and they don’t click through, the economics shift. That’s why at least one major publisher, Penske Media Group, has filed suit against Alphabet/Google alleging AI summaries rely on publishers’ work and divert traffic and revenue.

At the same time, business analysts point to large opportunity: McKinsey (2025) estimated AI‑powered search could mediate roughly $750 billion in U.S. consumer revenue by 2028. The net effect depends on whether your content is the kind that gets cited (and benefits from higher‑quality referrals) or the kind that historically relied on high‑volume, low‑value clicks. In short: winners and losers will be determined by query type, content format, and how often your pages are cited inside AI Overviews.

How to measure the true impact on your business

Google’s “total organic click volume” metric is useful but opaque and aggregate. To understand the business impact, track metrics that tie search UI changes to revenue and engagement at the page and query level. Make these operational:

Activation rate, Definition: percent of your tracked keyword set that returns an AI Overview. How to measure: scrape SERPs daily for your top 500-1, 000 queries and compute the share with the AI card present.
Presence / citation rate, Definition: percent of AI Overviews that cite your domain for queries you rank for. How to measure: scrape the AI card and record cited domains. Compare against your organic rank positions for the same queries.
Citation click‑through rate (citation‑CTR), Definition: clicks on links inside the AI Overview versus clicks to the first page of organic results. How to measure: combine server logs or UTM‑tag experiments with SERP audits. Note that referral patterns may not always pass clean referrers from generative cards, so instrument test pages with unique identifiers.
Per‑visit economics, Definition: revenue per session (ad revenue, conversions, or lifetime value) segmented by referral type. How to measure: attribute conversions and ad revenue to sessions via analytics and compare AI‑referred sessions to other channels.
Query‑level lift/loss (diff‑in‑diff), Definition: the traffic delta for pages that trigger AI Overviews versus matched controls over time. How to measure: run a difference‑in‑differences experiment comparing affected pages to similar unaffected pages across a 60-90 day window before vs after AI rollout or activation.

Concrete experiments you can run this quarter

Don’t guess, test. Here are practical experiments analytics teams can set up quickly.

SERP scraping + traffic matching, Scrape SERPs for your top 500 queries daily. Tag each query by whether an AI Overview appears and which domains are cited. Then match those flags to daily pageviews for the landing pages that typically rank for those queries.
UTM‑tagged control pages, Publish near‑duplicate pages with UTM variants (or short canonical experiments) and monitor whether AI cards cite one type more often. This helps evaluate which page formats are more “citeable.”
Diff‑in‑diff on similar content, Identify two sets of pages with similar traffic histories. If one set begins to trigger AI Overviews more frequently, compare their traffic deltas to the control set to estimate impact.
Per‑visit value A/B, Route a portion of AI‑targeted landing traffic to alternative conversion flows (different CTAs, first‑time offers) and measure whether AI referrals convert at higher or lower rates than other channels.

Practical plays for publishers, product teams, and marketing leaders

No single tactic will fully reverse any referral losses. But combined, these moves reduce risk and capture upside.

Diversify acquisition, Accelerate owned channels: newsletter, direct app experiences, memberships, and social. Owning the user relationship reduces vulnerability to referral shifts.
Optimize for being cited, Make content easier for generative models to summarize: clear headings, concise lead paragraphs, structured data (FAQ, QAPage schema), and explicit summaries near the top of articles. That increases your presence/citation rate and brand visibility inside citations.
Instrument attribution, Use UTM experiments, server logs, and controlled scraping to detect AI referrals. Don’t rely solely on raw sessions. Tie traffic to revenue per session and downstream value.
Productize unique assets, Turn exclusive reporting, proprietary data, and long‑form analysis into paid products or APIs. Short AI summaries can’t replace subscription models or proprietary datasets.
Explore licensing and partnerships, The Penske Media complaint signals one route. Others may be negotiated licensing or revenue‑share pilots with platforms that use publisher content.
Prepare for reputation risk, Generative summaries can be wrong or misattribute sources. Implement monitoring for misattribution and be ready to use feedback channels, corrections, and public escalation if the AI output harms your brand.

The legal and policy angle

Penske Media Group has sued Alphabet/Google alleging that Google’s AI summaries use publishers’ reporting and reduce traffic and revenue. That complaint brings the debate into the courts and raises questions about attribution, licensing, and compensation for content used to train or feed generative systems. Expect more legal and regulatory attention: courts and regulators may force more transparent source attribution or open the door to licensing frameworks, but litigation timelines are long and outcomes uncertain.

Liz Reid (Google VP of Search): “AI is driving the most significant upgrade of the Google Search experience ever.”

Google (on aggregate clicks): “Total organic click volume from Google Search to websites has been relatively stable year‑over‑year.”

Prioritized checklist for leaders

Immediate (next 30 days)
- Run a SERP audit for your top 200 commercial and informational queries to measure AI Activation Rate.
- Instrument revenue per session and conversion rate for any pages that rank for question‑style queries.
Medium (30-90 days)
- Run a diff‑in‑diff analysis comparing pages that trigger AI Overviews against matched controls over a 60-90 day window.
- Test content formats (short summaries, structured data) to increase citation probability and measure citation‑CTR.
- Start conversations about licensing or partnership pilots for high‑value content verticals.
Strategic (3-12 months)
- Prioritize productization of unique assets (subscriptions, data APIs) to reduce reliance on ad‑supported pageviews.
- Build owned channels (email, apps) and invest in LTV per user rather than raw acquisition volume.
- Develop an ongoing audit process for AI‑generated misattribution and a remediation playbook (feedback, corrections, legal escalation pathways).

Key questions you should be asking now

Are AI Overviews reducing my site’s search referrals?

Measure it directly: scrape SERPs for your top queries, flag AI‑activated results, and run a diff‑in‑diff comparing traffic to affected pages versus matched controls. Academic audits using this approach report single‑digit to low‑double‑digit declines on affected pages.
Is the traffic I still get from AI valuable?

Don’t judge by visits alone. Segment sessions by referral source and track revenue per session, conversion rate, pages per visit, and downstream retention. Vendor reports (Adobe, Microsoft Clarity) show higher per‑visit value in some retail contexts; other studies (Ahrefs) report shallower sessions for informational traffic, your mileage will vary.
Should we pursue legal action or licensing?

Penske Media’s lawsuit shows litigation is an option; licensing or negotiated partnerships are a commercial alternative. Both routes carry costs and uncertainty. Begin by quantifying the economic harm and exploring commercial pilots before committing to litigation.
What metrics should the C‑suite be tracking now?

Move beyond raw organic session counts. Track activation rate, presence/citation rate in AI Overviews, citation‑CTR, revenue per session, and conversion by referral source, these reveal economic impact faster and more accurately than session volume alone.

Where this is likely to go next

Expect iterative change. Google will refine when AI Overviews appear, how they cite sources, and how much they link back. Publishers will optimize formats that are more likely to be cited, explore licensing and partnerships, and accelerate owned‑audience strategies. Regulators and courts may intervene on source attribution or compensation. The near term will be an era of experiments, by platforms, publishers, and advertisers, tested against concrete metrics.

The essential strategic choice for leaders isn’t whether Google Search changed, clearly it has. The question is whether your organization treats that change as a distribution problem to adapt to (optimize for citation, measure per‑visit value) or as an existential threat to be litigated away. Preparing metrics, running experiments, and diversifying revenue and audience channels buys you optionality either way.