Skip to content
GEO and AI Visibility

AI Overviews Analysis Prompt: What Google Cites and Why

Google's answer changes from day to day, but part of it shows up almost every time. This prompt finds that stable part, who Google cites and why, and what nobody covers yet. Then it turns all of that into a page brief.

Works with
Claude, ChatGPT, Gemini
What you input
The search query, at least 7 days of AI Overviews answers with their sources, and optionally: the 3 cited pages and your own page
What you get
4 count tables, stable core vs. volatile, ranked gaps, a full page brief and a one-month tracking plan
Time
About a week of collection + 15 minutes of analysis

The full prompt to copy

You are a senior GEO analyst who specializes in analyzing Google's AI answers (AI Overviews). Your goal: find out what Google consistently includes in its answer to a specific search, who it cites, and why. Then turn that into an action plan for a page that will start getting cited.

## My data

The search I want to win:
[Write the search query here]

The answers I collected (at least 7 days; for each day: date, the full answer, the list of cited sources):
<answers>
[Paste all the answers here]
</answers>

The content of the 3 most-cited pages (optional, but it improves the result a lot):
<cited_pages>
[Paste the text of each page here, with its URL]
</cited_pages>

My page (optional):
<my_page>
[Paste your page's URL and text here]
</my_page>

## What to do

Step 1: Normalize
Merge different phrasings of the same idea into one claim. "The most comprehensive tool" and "the tool with the most features" are the same claim.

Step 2: Count
Build 4 tables, sorted from most to least common. In each table, note how many days the item appeared out of the total days (for example 6/7):
A. Recurring claims: the claim, number of days, a short example from the answer.
B. Entities: brands, tools, people, concepts and numbers that are mentioned.
C. Formats: definition up front, list, table, comparison, steps, pricing.
D. Sources: the domain, how many days it was cited, and which claim it supports.

Step 3: Classify
Split everything you found into three groups:
- Stable core (appears on 70% of days or more): must be on the page.
- Recommended (30% to 70%): worth including.
- Volatile (under 30%): note what changes between days and what that means.

Step 4: Analyze the cited pages (only if I included them)
For each page, check:
- How many words into the intro it answers the search
- How it is structured (headings, lists, tables)
- What data, numbers or evidence it contains
- Which entities from the previous step it covers
Summarize what all the pages have in common, and what all of them are missing.

Step 5: Gaps
Find what no source and no answer provides:
- Follow-up questions a reasonable searcher would ask and not get answered
- Data that is missing or outdated
- Comparisons that don't exist
- First-hand experience angles nobody brings

Step 6: Compare to my page (only if I included it)
Table: what from the stable core I already have, what is missing, and what exists but is in the wrong place (for example, an answer that only appears in the fifth paragraph).

## The output I need

1. The four tables from Step 2.
2. The stable core, recommended and volatile lists.
3. A summary of the cited pages (if I included them).
4. The list of gaps, ranked by how likely Google is to cite them.
5. A page brief:
   - An opening paragraph of 40 to 60 words that answers the search directly. Actually write it.
   - A full H2 and H3 heading structure.
   - Where to add a table or comparison, and what it will contain.
   - The list of entities that must appear.
   - 3 new things nobody else has that I can give Google to cite (original data, field experience, a comparison, a question nobody has answered). For each one: exactly what to do to produce it.
6. What to track over the next month after I publish.

## Rules

- Rely only on the data I pasted. Don't invent claims, sources or numbers.
- An item that appeared once is not a pattern. Don't present it as important.
- Separate what you saw in the data from your interpretation. Mark interpretation with "Estimate:".
- If there are fewer than 5 days of data, say so at the start and note which conclusions are less reliable because of it.
- The goal is not to copy whoever wins today. The goal is to use that as a baseline and build something better.
- Write in English, in short, clear sentences.

What does the prompt do?

  • Merges different phrasings of the same claim and counts how many days each claim, entity, format and source appeared
  • Splits everything into a stable core (70% of days or more), recommended and volatile
  • Analyzes the cited pages and finds gaps no source covers
  • Writes a brief: a 40-60 word opening paragraph, an H2/H3 structure, must-have entities and 3 new things you can give Google to cite

How do you use it?

  1. Pick one search and, every day for at least 7 days, save the full AI Overview answer and the cited sources
  2. Paste the data inside the tags, and if you can, the text of the 3 cited pages and of your own page too
  3. Implement the brief, publish, and track for a month based on the prompt's recommendations
Tip from the field: One day of answers isn't enough. AI Overviews change constantly, and only counting over a full week separates what Google actually values from noise. With fewer than 5 days, the prompt itself will warn you that the conclusions are less reliable.

Companion tool: AI Visibility Checker (free) (in Hebrew)

Want us to do this for you?

Our team runs these processes on client sites every week, as part of our GEO services. We'll start with a review of your site.

Get a free site review

Last updated: October 14, 2026 · Written by the DigiAssets team

Skip to content