AI Overviews at the One-Year Mark: Presence, Size, and What They’re Citing

Summary:

This report examines how often Google shows AI Overviews (AIOs) in search, how much screen space they take up, and whether the sites they cite are the same ones that rank well in normal organic search.

It was published by BrightEdge, an enterprise SEO software company that sells tools for search and AI search tracking.

BrightEdge says it used its own AI Catalyst and Generative Parser tools to track a set of keywords over time. It tracked AI Overview appearance and average pixel height daily from February 2025 to February 2026. It also checked citation overlap weekly to see whether AI-cited pages also ranked in the organic top 10 or top 100.

For industry results, it compared year-apart snapshots across nine sectors. The source does not say how many keywords were tracked or how that keyword set was chosen, which is an important gap.

The main findings:
• AI Overviews grew from about 30% of tracked queries to about 48% in a year. They now appear on nearly half of the keywords BrightEdge followed.
• Even so, about 52% of queries still showed no AI Overview at all, which means regular organic search still makes up most of the experience.
• When AI Overviews do appear, they are big. Average height rose from about 1,050 pixels to about 1,200 pixels, with a peak monthly average of about 1,340 pixels. BrightEdge says that’s taller than a standard desktop screen, so the first organic result is pushed below the fold.

The report also says AI Overview citations are usually not the same as page-one organic winners. Only about 17% of cited sources also ranked in the organic top 10, and that stayed mostly flat over the months tracked.

About half of cited sources ranked somewhere in the top 100, and that share rose from roughly 49% to 53%, suggesting slow movement toward more overlap.

The big catch is the lack of methodology detail. This is based on BrightEdge’s own tracked keyword set and proprietary tools, but the report does not disclose the keyword count, sampling rules, weighting, device mix, or how representative the data is. That makes it hard to judge how generalizable the numbers are.