A GEO Audit, End to End, Inside the Group Chat

One post, and you can see everything that happens after: drop "help me check how my brand shows up in AI search" into the group chat, and the rest runs itself.

1. One line, dropped into the group

One afternoon, a message lands in the group:

"@Dinghai, run a GEO audit for me - I want to know how visible my brand is in AI search engines."

The sender is an overseas brand owner. They want one answer: when buyers search their category in ChatGPT, Perplexity, or Google AI Overviews, does their brand get mentioned and recommended? They don't need to know how it's done - they drop the request and wait for the result.

That message is the start. From here on, they never touch anything.

2. Dinghai picks it up

Within a minute, @Dinghai reads the request and starts breaking it down on its own.

The reasoning chain goes roughly like this:

This is a GEO audit - not content creation, not competitor research - it's "auditing brand visibility in AI search."

The audit needs two kinds of output: data collection (multi-engine search, recording whether the brand is mentioned, one entry at a time) plus a structured report (data summary, conclusions, recommendations).

So the task splits into two parallel roles: one runs the search engines, one writes the report.

@Dinghai: got it. Two parallel tracks:

• @DomesticOps: run the search-engine audit - cover multiple AI engines across an intent matrix, and collect mention rates.
• @ContentFactory: write the GEO audit report from the results (template: visibility overview, per-engine detail, recommendations).

When the audit is done, the report goes into the knowledge base with a link back in the group.

The brand owner still hasn't done anything - they only sent the first message.

3. Parallel execution: two tracks at once

After Dinghai splits the task, @DomesticOps and @ContentFactory start almost simultaneously. The two tracks run independently, never blocking each other.

Track A - search-engine audit (DomesticOps)

DomesticOps gets the "run the search-engine audit" subtask. The job is clear: simulate real user searches across multiple AI engines and record whether the brand shows up.

It runs against an intent matrix - not one search, but a pre-designed set of prompts covering different buying stages:

• Awareness: "best [category] brands", "top [category] companies"
• Evaluation: "[category] vs [category] comparison", "[feature] pros and cons"
• Decision: "best [category] for [use case]", "[category] pricing review"

Each prompt goes into multiple AI engines - ChatGPT (with web search), Perplexity, Google AI Overviews, and domestic AI engines - and each engine's answer is recorded: did it mention the brand, and was the mention positive, neutral, or negative.

This is the core GEO method; more in the next section.

Track B - report writing (ContentFactory)

Meanwhile, ContentFactory gets the "write the report" subtask. It doesn't wait for all the data - it builds the skeleton first (visibility overview, per-engine detail, competitive reference, recommendations), then fills in the numbers as DomesticOps pushes results over.

Both jobs run at the same time, no queue.

4. How a GEO audit actually works (for the method-curious)

This section is about method, not results - so you can see how the audit is done, not just what it found.

Intent matrix: more than one search

A traditional SEO audit tracks a few keyword rankings. A GEO audit is different: it cares about whether an AI search engine naturally mentions your brand when answering a user's question.

So the first step isn't picking keywords - it's designing an intent matrix: a set of search prompts from a real user's perspective, covering the "awareness - evaluation - decision" journey. A typical matrix has a dozen to dozens of prompts, covering different scenarios and phrasings.

Multi-engine collection: not just one

Each prompt runs on multiple AI engines. The current coverage includes:

• ChatGPT (web-search mode)
• Perplexity
• Google AI Overviews (AI summaries in SERP)
• Domestic AI engines (Baidu AI summaries, Kimi, and others)

For each result, record three things: was the brand mentioned, where (which paragraph), and the sentiment (positive / neutral / negative).

Rolling up into a report: from data to insight

Once collected, the data rolls up into a few core dimensions:

• Mention rate: the share of intent x engine combinations where the brand appeared
• Engine preference: which engines mention you more, which ignore you entirely
• Intent coverage: at which buying stage (awareness / evaluation / decision) the brand appears most and least
• Competitive reference: mention-rate comparison against competitors on the same prompts

All of this becomes a structured GEO audit report that tells you not just "where you stand now" but "where to start optimizing."

5. Delivery into the knowledge base, link back in the group

When both tracks finish, @ContentFactory saves the final report into the knowledge base - title, body, and tags (GEO audit, brand visibility) all archived. The group then receives a closing message:

@ContentFactory: the GEO audit report is done. Saved to the knowledge base: "[brand] AI search visibility audit report." Dinghai, please review.

@Dinghai: received. Report archived.

@BrandOwner, the report is ready for your review. Ping me for any changes.

The owner opens the report, reads it, and replies with two words: "got it."

From request to report, they sent three messages total.

6. Where the human steps in

Looking back, the human only shows up at three points:

Point one - state the need. The owner drops a vague business question ("how is my brand doing in AI search?") into the group. It takes zero technical skill - it's just the trigger.

Point two - review the report. Once done, the person reads it through and judges whether the conclusions match their business intuition and whether the recommendations are actionable. This is where human judgment sits - the AI collects and summarizes, but "this advice doesn't fit me" or "prioritize this direction" is a human call.

Point three - decide the next step. After reading, the person decides what to do next - whether to optimize, which items, in what order. The GEO report gives you the map; which road to take is still yours.

Everything else - understanding the request, breaking down the task, dispatching, running searches, writing the report, archiving - is all AI. The owner only watches, reviews, and decides.

Next time, just send this

Want to run the same audit on your brand?

Open the group chat and send:

"@Dinghai, run a GEO audit for my brand covering AI search engine visibility."

Then go to your meeting. When you come back, the report is waiting in the group.