When to use this prompt
Run this when you are tracking a specific set of priority commercial queries and need to know whether Google AI Overviews are appearing on them, who is cited there and in AI Mode, and how the cited content is structured. Pair it with the citation share audit for a complete view of AI search visibility.
This is also the right prompt to run after Google rolls out an AI Overview update. Coverage and citations shift after each update, and being late to notice usually means losing share before competitors.
You collect the results first, because no chat model or research agent can reliably render a live Google results page. Two ways to do it: search each query yourself in a logged-out incognito window and copy the AI Overview text, its cited sources, and the AI Mode answer’s cited sources; or pull the same data from a SERP API provider such as SerpApi or DataForSEO and paste the relevant fields. Note the date, location, device, and method, because AI Overviews vary by all four. The model then does the analysis and tabulation.
The prompt
<role>Search analyst analyzing Google AI Overview and AI Mode observations that the user captured. You work only from the pasted observations. Do not search the web and do not use training data to fill gaps.</role>
<task>For each query, turn the pasted observations into a structured record: whether an AI Overview appeared, who it cited, what shape it took, who AI Mode cited, and how the cited pages are structured. Then summarize the patterns.</task>
<inputs>
<target_domain>[YOUR DOMAIN]</target_domain>
<collection_context>
Capture date: [YYYY-MM-DD]
Location: [COUNTRY / CITY]
Device: [Desktop or Mobile]
Method: [Manual logged-out incognito check, or SERP API name]
</collection_context>
<observations>
Query 1: [QUERY]
AI Overview: [Yes / No]
AI Overview text: [PASTE THE OVERVIEW TEXT, or "not captured"]
AI Overview cited sources: [PASTE DOMAINS AND PAGE TITLES IN ORDER SHOWN]
AI Mode cited sources: [PASTE DOMAINS AND PAGE TITLES, or "not checked"]
Query 2: [REPEAT THE SAME FIELDS FOR EACH QUERY]
</observations>
</inputs>
<instructions>
1. Use only the data in <observations>. Do not run searches, and do not infer missing values from training data.
2. For each query, record:
- AI_OVERVIEW_PRESENT: Yes or No, exactly as observed.
- CITED_DOMAINS: comma-separated domains cited in the AI Overview, in the order given.
- CITED_TITLES: a one-line topic summary for each cited page, in the same order, based on the pasted titles.
- ANSWER_SHAPE: exactly one of [Definition, Comparison, List, How-To, Recommendation], judged from the pasted overview text. If no text was pasted, write "not provided".
- WORD_COUNT_ESTIMATE: approximate word count of the pasted overview text. If no text was pasted, write "not provided".
- AI_MODE_CITED_DOMAINS: comma-separated domains cited in the AI Mode answer, or "not checked".
- TARGET_CITED: where the target domain appears: AI Overview, AI Mode, Both, or Neither.
3. If a query did not produce an AI Overview, mark the AI Overview fields "n/a" but still record the AI Mode fields.
4. After the table, output a summary block per the format below. N is the number of queries in <observations>.
5. Do not fabricate. If a field is missing from the observations, write "not provided" rather than guessing.
</instructions>
<output_format>
Markdown table with columns:
Query | AI Overview | Cited Domains | Cited Titles | Answer Shape | Word Count | AI Mode Cited Domains | Target Cited
After the table, output exactly:
Summary (captured [date], [location], [device], via [method]):
- AI Overview present on: X/N queries
- Target domain cited in AI Overviews: X/X queries with overviews
- Target domain cited in AI Mode: X/X queries where AI Mode was checked
- Overlap: domains cited in both the AI Overview and AI Mode for the same query, or "none"
- Most common answer shape: [Definition / Comparison / List / How-To / Recommendation]
- Pattern in cited page types: [1-2 sentences naming the dominant page type, e.g. "Comparison pages and definition pages dominate; pillar pages cited more than blog posts."]
</output_format>How it works
The prompt separates data collection from analysis on purpose. Chat models and research agents cannot reliably load a live Google results page, and when asked to “check Google” they tend to invent plausible overviews and citations. Capturing the results yourself, by hand or through a SERP API, and telling the model to use only the pasted observations removes that failure mode. The collection context (date, location, device, method) is there because AI Overviews vary by all four; without it, week-over-week comparisons are noise.
AI Mode gets its own column because it behaves differently from AI Overviews. AI Overviews appear only on some queries, while AI Mode answers almost anything you ask it, so “did AI Mode answer” is not a useful signal. Who AI Mode cites, and whether those citations overlap with the AI Overview, is. A domain cited in both surfaces is one Google trusts for that topic.
The ANSWER_SHAPE field is the key insight. If 80% of your priority queries trigger comparison-style AI Overviews, the highest-leverage content investment is comparison pages, not blog posts. Most teams skip this step and write whatever they were going to write anyway.
The CITED_TITLES field tells you what type of page is winning. If Reddit threads dominate, your problem is not content structure, it is content surface area. You may need to seed discussion in places you cannot directly control.
Example output
| Query | AI Overview | Cited Domains | Cited Titles | Answer Shape | Word Count | AI Mode Cited Domains | Target Cited |
|---|---|---|---|---|---|---|---|
| best product analytics tool 2026 | Yes | g2.com, capterra.com, mixpanel.com | Top analytics list, Analytics buyer guide, Mixpanel pillar page | Comparison | ~180 | g2.com, mixpanel.com, reddit.com | Both |
| what is product analytics | Yes | amplitude.com, gartner.com, heap.io | Product analytics definition, Gartner definition, Heap explainer | Definition | ~120 | amplitude.com, heap.io | Neither |
| mixpanel vs amplitude | No | n/a | n/a | n/a | n/a | mixpanel.com, amplitude.com, reddit.com | AI Mode |
Summary (captured 2026-10-05, United States, Desktop, via manual incognito check): 4/5 queries had an AI Overview present. Target domain cited in 1/4 AI Overviews and 2/5 AI Mode answers. Overlap: g2.com and mixpanel.com on “best product analytics tool 2026”. Most common answer shape: Comparison (3/4). Pattern: Comparison pages and definition pages dominate citations. Pillar pages are cited more than blog posts.
Variations
- Position tracking: Add a CITATION_POSITION column to track whether your domain is the first, second, or third citation. Position matters because the first citation gets the most click-through.
- Page-type focus: Limit the panel to one page type (definitions, comparisons, how-tos) to deeply optimize for one shape at a time.
- Weekly diff: Capture the same queries weekly with the same location, device, and method, paste last week’s table alongside, and add a “previous overview present” column to flag queries that gained or lost AI Overview coverage.