What an Agent Should Do When Google AI Mode and AI Overviews Diverge
Promptwatch citation share data from February to June 2026 shows Google AI Mode and AI Overviews citing different domains. The agent move is to track each surface separately.
Google has two AI search surfaces, and they do not cite the same domains. Google AI Overviews and Google AI Mode each have their own citation share chart in Promptwatch's data, and across the first half of 2026 neither pulled away from the other. For anyone who runs an agent on their AI visibility, the practical question is what an agent should do automatically when one vendor runs two surfaces that cite different domains.
The citation share reports (AI Overviews and AI Mode, published June 28, 2026) cover the top domains each surface cited in June 2026, with a May comparison. The top AI Overviews sources report rounds out the picture.
The signal
Promptwatch's monthly citation share reports list the top domains each surface cited, with their percentage share of total citations. AI Overviews shows the top 16 domains in June 2026. AI Mode shows the top 17. Each report includes a May comparison chart, so the month over month move is visible alongside the level.
The pattern across the first half of 2026 is divergence without a winner. AI Overviews and AI Mode cite overlapping but not identical sets of domains. Some domains gain share in one surface and lose it in the other over the same month. Neither surface holds a steady lead across the six months of data Promptwatch has published for the pair.
What an agent should do automatically
The first move is to track each surface separately, not as one Google AI visibility score. An agent that tracks visibility should split its view by surface, so a citation in AI overviews is not counted the same as a citation in ai mode. A single score would hide the divergence, and a domain that loses share in one surface while it gains in the other would look flat when it is not.
The second move is to build for the query type each surface answers. AI Overviews lean toward research and informational queries, where a reference source or a deep guide is the most useful. AI Mode leans toward transactional and conversational queries, where a product page or a direct answer is the most useful. A Content Agent should pick the format based on which surface cites you, not a default template.
The third move is to watch the two citation share reports over months, not on a single month. An agent should track your domain's share in each surface on a schedule and flag a sustained step in one surface that does not show up in the other. A divergence that holds across months is the signal that the two surfaces need separate work, not the same work.
How to do it with Promptwatch
The measurement that turns the population report into a prompt level action is the citation analytics view in Promptwatch. It tracks page level, domain level, Reddit citations, YouTube citations, offsite mentions, and citation type breakdowns, and it can be read per AI surface.
An agent that runs on visibility should wire the two citation share reports into a Unified Action. When your domain's share moves in one surface but not the other, the action is to pull the citation view split by surface, identify which surface moved, and flag the prompts where the shift happened. That is the automated move that turns a surface split into a fix, not a dashboard you read and forget.
What the report does not tell you
The reports are population level views. They tell you what each surface is doing across the whole web. They do not tell you what each surface is doing for your prompts. For that you need your own citation data joined to your own visibility data, split by surface. A domain can lead at the population level and not be cited for your prompts, or be cited for your competitors.
The reports also do not say why the two surfaces cite different domains. The split is consistent with the two surfaces answering different queries, not with one surface ranking the same domains higher. AI Overviews lean toward research and informational. AI Mode leans toward transactional and conversational. The page that earns a citation in one is the page that answers the query type that surface serves.
The broader pattern
The surface split is one instance of a wider change in AI search. Google is not the only vendor with more than one AI surface. OpenAI has ChatGPT Search and the citation types and shopping surfaces around it. The pattern of one vendor running two surfaces that cite different domains is becoming more common, not less.
For an agent that runs on visibility, that means the number of retrieval surfaces to optimize for is growing, and the surfaces are not interchangeable. A page that earns a citation in AI overviews is not the same as a page that earns a citation in ai mode, and the two need separate work. The practical response is to treat each AI surface as a separate target.
How to read a divergence versus a lead
The two surfaces can move in two ways, and an agent should tell them apart. A divergence is when the two surfaces cite different domains, which is consistent with them answering different queries. A lead is when one surface cites a domain more than the other, which is consistent with a ranking difference. The two call for different responses. A divergence is a reason to build for both surfaces. A lead is a reason to build for the surface where you are behind.
The Google data is the divergence shape. The two surfaces cite overlapping but not identical sets of domains, and neither holds a steady lead. That is the kind of change an agent should treat as two targets, not one. A divergence that holds across months is a signal that the two surfaces need separate work, and an agent that tracks them as one score will miss it.
What to watch next
The reports are snapshots through June 2026. The open questions are whether either surface pulls away in July, whether the top domains keep diverging, and whether a new domain enters the top of either. Those are exactly the questions a per surface citation view answers for your own domain.
The dataset Promptwatch publishes is aggregated and non identifiable, and it is refreshed constantly. That makes it useful for spotting population level shifts. It does not replace the need to track your own brand, your own prompts, and your own citations split by surface. The two work together: the public report tells you what is happening in the field, and your own tracking tells you whether it is happening to you.