AI Overviews vs. AI Mode: Two Different Optimization Jobs

AI Mode is autoplay. AI Overviews is the Netflix browse. That single sentence, lifted from a new clickstream study of 846,000 Google sessions published by Kevin Indig on May 25, 2026, explains why most AI visibility playbooks are about to break. Users do not behave the same way in both surfaces. They behave in opposite ways. And if your measurement stack treats them as one thing, you are measuring the wrong thing.

TL;DR

The study, run by Eric Van Buskirk at Clickstream Solutions using anonymized data from Surfer SEO, found four behavioral shifts when an AI Overview is on the page:

• Cursor positions spread across 83% of the viewport versus 66% without an AIO

• Half of all scrolling now goes backward (47.5% reverse scroll versus 27% without an AIO)

• Search intent stops predicting time-on-page (the 20-point spread collapses to 6)

• Branded searches lose their shortcut (cursor activity up 40% on navigational queries)

Meanwhile, a separate April 2026 study of 185 high-stakes purchase decisions showed that inside AI Mode itself, 88% of users take the shortlist as-is, 74% pick the first result, and 64% click nothing at all.

Same person, opposite jobs. That is the whole story.

Two surfaces, two completely different jobs

Most SEO and GEO tools still treat AI Mode and AI Overviews as a single "AI surface" you optimize for. The data says they are two distinct problems.

AI Mode is a closed loop. The user reads the answer, picks something from inside the answer, and moves on. If your brand is not in the model's shortlist, you do not exist for that query. That is a model-layer problem. You solve it by building authority signals the underlying model can actually see and trust.

AI Overviews is a comparison environment. The AIO box sits at the top of a real SERP. The user reads it, pauses, scrolls back up to reread, weighs the options against the blue links below, and only then decides whether to click. Your listing gets two or three impressions per session, not one. That is a SERP-layer problem. You solve it by surviving the second look.

Same content strategy will not solve both.

What the numbers actually say

Here is the comparison view nobody else is going to put in one table.

Sources: cursor and scroll data from the 846K-session Clickstream Solutions / Surfer SEO study; AI Mode acceptance rates from the April 2026 high-stakes purchases study.

The numbers point in opposite directions. In AI Mode, the user is short-circuiting deliberation. In AI Overviews, the user is doing more deliberation than they would on a classic SERP, not less.

Four measurement implications most tools are missing

If you accept that AIO and AI Mode are different jobs, four things should change about how you measure performance.

1. AIO is a comparison environment, so track who you appear next to

When half of scroll movement on an AIO SERP goes backward, the user is comparing. They are not just checking whether you appear. They are checking who appears with you and whether the others look stronger.

A presence-only metric (did we get cited, yes or no) tells you almost nothing in that environment. The richer signal is your co-citation set: which sources show up alongside you in the AIO box, in what order, with what framing. That is the data that tells you whether the user kept reading or kept scrolling.

2. Half of scrolling is reversal, so measure share of voice across the full SERP

The old model was simple. Win the AIO box, claim the impression, move on. The new model is messier. The user reads the AIO box, scrolls down through the organic listings, and then scrolls back up to compare.

That means your share of voice has to be measured across the full visible SERP, not just inside the AIO box. If you are cited in the AIO but your competitor owns positions one through three of the organic results below, the comparison you lose happens during the reverse scroll, not during the first read.

3. Intent no longer predicts time-on-page, so stop segmenting by intent

This is the finding that breaks the most reporting templates. Without an AIO, time-on-page varied by 20 percentage points across intent types. With an AIO, that spread collapses to 6 points. Navigational, informational, local, transactional, video: all five cluster between 41.9% and 48.5% still-active at the 21-second mark.

Intent still matters for what you write. It barely matters for how long users stay on the SERP. If your dashboards segment AIO performance by intent type, you are slicing on a dimension that no longer predicts behavior. Topic cluster, query difficulty, and competitive density are now better axes.

4. Branded searches lost their shortcut, so brand monitoring is table stakes

Cursor activity on navigational queries jumps 40% with an AIO present. A user who types your brand name into Google is no longer taking the express lane to your homepage. They are scanning the AIO content first, evaluating what Google says about you before deciding whether to click through.

That is a defensive measurement problem. You need to know, for every branded query that triggers an AIO, what Google is saying about you, who is being cited as a source for those claims, and whether any of those sources are competitors or hostile reviews. If you only monitor non-branded queries, you are blind to the most consequential surface of all.

What changes in your workflow this quarter

Three things to do this quarter, in order of leverage.

First, audit your tracked queries and split the report. For each monitored prompt or keyword, log whether it triggers AIO, AI Mode, or both, and report performance separately. A dashboard that averages across both surfaces is hiding the signal.

Second, add reverse-scroll context to your meta descriptions. The old assumption was that a user reads your snippet once and decides. The new reality is that they read it, scroll past, and read it again on the way back up. Snippets that worked on the first impression because they were punchy can fail on the second impression because they have no substance. Write for the reread, not the glance.

Third, start tracking citation co-occurrence. Who appears next to you in the AIO box is now part of your competitive set, even if they are not a competitor in any traditional sense. A Wikipedia article, a Reddit thread, and a niche industry blog can collectively decide whether the user clicks through to you or to someone else. You cannot influence what you cannot see.

The bigger picture

For two decades, SEO measurement worked because the SERP worked the same way for everyone. You appeared, you got an impression, you got a click or you did not. The AIO study is the clearest evidence yet that the equation has changed.

AI Mode took users out of the SERP entirely for a class of decisions. AI Overviews kept users on the SERP but changed what they do there. Treating both as variations of "Google traffic" misses that they require different content, different signals, and different metrics.

The tools that win the next two years will be the ones that measure each surface on its own terms. The ones that bolted on AI tracking as a column in an SEO dashboard will keep producing reports that look right and predict nothing.

See how your brand is showing up in both

AI Sightline tracks AI Overviews and AI Mode separately, with citation-level visibility, co-citation context, and sentiment scoring across six platforms including ChatGPT, Perplexity, Gemini, Google AIO, AI Mode, and Copilot. If you want to see what users are actually seeing when they search for you, start free. No credit card, two minutes to first scan.

Data sources: Users behave differently in AI Overviews vs. AI Mode by Kevin Indig (Growth Memo, May 25, 2026); Clickstream Solutions methodology; How consumers navigate high-stakes purchases (Growth Memo, April 2026).