Should you track AI Overviews and AI Mode as one channel or two?

Why the question matters now

The question matters because both surfaces are now big enough to report on, and they behave differently. Google said at I/O on 19 May 2026 that AI Mode had passed one billion monthly users, with AI Mode queries more than doubling quarter on quarter.¹ That is Google's own statement, without a published method.

AI Mode is also widely available. Google announced in October 2025 that it had expanded AI Mode to more than 200 countries and territories, with more than 35 new languages.² A brand selling across markets will meet it in most of them.

Many tools and agencies still report a single "Google AI" number. Our view: that number is easy to read and hard to act on, because it blends a surface people did not ask for with one they chose to open.

How are AI Overviews and AI Mode different?

They differ in four ways that change what you measure: how an answer is triggered, how long it is, what it cites and who sees it.

Two Google AI surfaces, four differences

AI Overviews

• Trigger: Google decides, on a normal search
• Short summary above the results
• Cites a few pages, often ones that rank
• Seen by everyone who searches that query

AI Mode

• Trigger: the user opens it and asks
• Longer, conversational, with follow-ups
• Cites more widely, including Google's own cards
• Seen by people who chose a conversation

The trigger difference is the one most often missed. An AI Overview shows only on some searches, so the first thing to measure is whether one appears at all. AI Mode answers whenever it is asked, so presence is never the question there.

The audience difference follows from the trigger. Everyone who types a query may see its AI Overview. AI Mode reaches people who opted into a longer conversation, which often means a more considered question.

What the studies show about each surface

The published studies support treating them separately, though none compares the two surfaces on one prompt set.

AI Overviews appear on a minority of searches

Ahrefs, which sells SEO and AI visibility tools, studied 146 million desktop results pages in September 2025, data now close to a year old. AI Overviews appeared on 20.5% of them, on 44.1% of medical queries and on 57.9% of queries phrased as questions.³ Ahrefs did not state the geography.

The useful point is the spread, not the exact rate, which changes as Google adjusts triggers. Question-shaped and health queries carry AI Overviews far more often than the average. A prompt set heavy with questions will see more of them than your real traffic does.

Engines cite very different numbers of sources

Citation counts per answer vary widely between AI engines. Semrush, which sells an AI visibility toolkit, analysed 126 million US prompts from January to April 2026. It found ChatGPT cites about 15 sources per response against about 3 for Gemini.⁴ AI Mode was not split out in that figure.

The implication for tracking is that the share of citations you hold depends on how many slots an answer has. On a surface that cites a handful of pages, one citation is a large share. On one that cites many, it is a small one. That alone is a reason to keep denominators separate.

AI Mode leans on Google's own cards

Profound, which sells AI visibility software, reported that google.com's share of AI Mode citations grew 8.4 times between 15 April and 30 June 2026, driven by Business Profile and product knowledge panel cards.⁵ That pattern was reported for AI Mode, and a brand's Google-hosted listings matter more there than on a page-citing surface.

How to split a prompt set without double counting

Use one prompt set, run it on both surfaces and tag every result by surface. Writing two prompt sets makes the surfaces impossible to compare, and adding the results counts one buyer question twice.

1. One prompt list. Write the buyer questions once, grouped by intent: category, comparison, problem and brand.
2. Run on both surfaces. For AI Overviews, run each prompt as a normal search and record whether an Overview appears. For AI Mode, ask the same prompt in the conversation.
3. Repeat across days. Answers vary from run to run. Our audits run each prompt on three separate days before reporting a share.
4. Tag and report separately. Each result carries its surface, date, country and the brands and URLs it names.

Our reporting split. Illustrative; adapt the intents to your category.

Country belongs in the tag as well. AI Mode is available in many markets and languages, and answers to the same prompt can differ by country. If you sell in several markets, run the prompt set from each one and keep the results apart. Example prompts should name the buyer's real context, such as "best payroll software for a 50-person company in Germany", rather than a generic query.

The denominator row is the one to protect. A brand named in half of the AI Overviews that appeared has not been named on half of all searches, if Overviews appeared on only some of them. Our buyer's spec for an AI visibility score covers how to report a share with its margin of error.

What goes in the board report

Show the two surfaces side by side, each with its own share and trend, and never a total. A combined "Google AI" figure moves when the mix of prompts changes, even if nothing about your visibility did.

Pair each share with one outcome measure, such as clicks from pages that were cited. Our guide to AI search numbers in a board report sets out which numbers belong there and which belong in a working dashboard.

Our view: if a vendor can only give you one Google AI number, ask for the raw answers tagged by surface. If it cannot give you those either, it is not measuring the thing you will be asked about.