01
Ask
Every week it asks 20 to 50 questions your buyers would really type, across ChatGPT, Gemini, Perplexity and Claude, with web search on. Where it cannot reach an engine, it asks you to paste the answer in. It does not guess.
Weekly AI Search Audit Skill · Version 1.0
A weekly AI search audit,
in one skill file.Thank you for watching.
This is the agent from the video. Each week it asks the questions your buyers ask AI assistants, logs who gets mentioned and cited, reads the pages that beat you, and drafts one fix for you to approve.
Twenty-five files of plain Markdown and CSV. It runs in ChatGPT, Claude, Codex, or any app that takes project instructions.
GPT-6 Astra or Grok Bot: where I landed.
Build your SEO and AEO agent on GPT-6 Astra, and point the citation work at ChatGPT first. Grok Bot is the better packaged always-on product, and it is worth running as its own lane if your buyers are on X.
The two are not the same kind of thing. Astra is a model. Grok Bot is a product built on Grok models. So the useful question is which model does the thinking, and separately, which product runs the loop. That is my read as of September 30, 2026. This moves monthly. The numbers behind it are in the video with their sources, and they should be rechecked before you lean on them three months from now.
You have ChatGPT Plus and you are not technical
Run the agent in ChatGPT Work with Astra. No code. Paste the skill in as the project instructions.
You are technical and want it on a schedule
Run it through Codex or the API, with a scheduler.
You run this for several clients
Use the API. One folder per client, each with its own brand file, question set and logs.
You want no setup and something that runs all day
Grok Bot is the easier product. I would still put the gap analysis and the drafts through Astra, because that is where accuracy counts.
Your buyers really do live on X
Run this agent for the website work, and a separate X posting plan beside it. Different game, different rules.
One agent, one weekly loop.
It does not need to be a swarm of twelve. The skill is one set of instructions, the detail it loads when it needs it, and the files it fills in.
02
Log
For each answer it records three separate facts: whether you were mentioned, whether you were recommended, and whether a page of yours was cited. Plus who was named instead, and which domains keep winning.
03
Read
It opens every page that beat you and writes down why, in one line. If the reason does not fit in a line, the page has not been understood yet.
04
Draft
It ranks the gaps, picks one, and drafts one answer-first page for it. The draft has to pass nine gates: evidence, consistency, originality, voice, and a reviewer that did not write it, among others.
05
Stop
It puts the draft in an approval queue and stops. You approve, edit or reject. You publish. It never does.
Once a month it also checks the boring things: robots.txt, whether your content is visible without JavaScript, and whether your sitemap is current. It reads and reports. It changes nothing.
Take the skill file.
One email address, and I will send the skill file straight to your inbox. The download also appears on this page, so you are not left waiting.
weekly-ai-search-audit-skill-1.0.0.zip
- Files
- 25
- Zip
- 31 KB
- Unpacked
- 56 KB
Would rather not give an address? The seven-role pack from my first Grok Bot video is still free and ungated at the GrokBot page.
Every file, before you decide.
Plain Markdown and CSV. Nothing compiled, nothing to install, no account to connect. The examples are filled in for a made-up business, and say so.
weekly-ai-search-audit-v1.0/18 Markdown · 7 CSV
| START-HERE.md | what this is, and the first run | 4.1 KB |
| SKILL.md | the skill itself: the agent's instructions | 7.8 KB |
| DECISION-GUIDE.md | where to run it, from the video | 2.3 KB |
| SOURCES.md | where the method came from | 2.0 KB |
| CHANGELOG.md | version history | 0.5 KB |
| LICENSE.md | yours to use, not to resell | 0.5 KB |
| references/ | detail the agent loads when it needs it | 22.3 KB |
| weekly-loop.md | every step, with the judgment calls | 5.6 KB |
| gates.md | the nine gates a draft has to pass | 4.8 KB |
| measurement.md | mentioned, recommended, cited | 3.9 KB |
| monthly-technical-check.md | robots.txt, no-JavaScript, sitemap | 2.9 KB |
| answer-first-draft.md | the spec for the one weekly draft | 2.5 KB |
| rules.md | what it never does, and why | 2.4 KB |
| templates/ | the files you and the agent fill in | 6.0 KB |
| brand-context.md | the one file you fill in first | 2.2 KB |
| question-set.csv | 20 rows, typed and ready | 0.8 KB |
| observation-log.csv | one row per question, per engine | 0.2 KB |
| gap-ledger.csv | one row per gap, with its status | 0.3 KB |
| technical-check.csv | the monthly check | 0.1 KB |
| approval-queue.md | drafts waiting for you | 1.3 KB |
| weekly-report.md | the one-page report | 1.2 KB |
| examples/ | the same files, filled in for a made-up business | 10.2 KB |
| README.md | what is fictional, which is all of it | 0.8 KB |
| brand-context-EXAMPLE.md | a completed brand file | 2.2 KB |
| question-set-EXAMPLE.csv | twenty approved questions | 2.1 KB |
| observation-log-EXAMPLE.csv | one question across four engines | 1.7 KB |
| gap-ledger-EXAMPLE.csv | two gaps, one drafted | 1.2 KB |
| weekly-report-EXAMPLE.md | the report for that week | 2.2 KB |
Your first run.
No tools to connect. Use the AI app you already pay for.
01
Fill in one file
Open templates/brand-context.md. Enter your brand, category, market and three competitors, plus the handful of facts a page about you is allowed to state.
02
Write 20 questions
Put them in templates/question-set.csv, in your buyers' words. If you have none yet, the skill proposes a set from your brand file and waits for you to approve it.
03
Run it once by hand
Use the AI app you already pay for. Read the gap ledger it writes, then approve or reject the one draft it proposes. Do that three weeks running before you trust the numbers.
Before you run it.
Files you can read and change.
Open the Markdown in any text editor and the CSVs in a spreadsheet. The skill is meant to be edited to fit your business before you run it.
Nothing here installs or connects anything.
These files do not set up bots, connect accounts or start schedules. Your AI app does that, with your accounts and your approval. Subscriptions and usage costs are yours.
No promise about the results.
The skill helps you collect evidence and decide what to do next. It cannot guarantee that an AI assistant will cite you, recommend you or send you traffic. Anyone who promises that is guessing.
Not affiliated with the platforms it names.
GPT-6 Astra, ChatGPT, Codex, Grok Bot, Gemini, Perplexity and Claude are named so you know where the skill can run. This page and the files are not affiliated with, endorsed by or sponsored by OpenAI, xAI, Google, Perplexity or Anthropic.
Weekly AI Search Audit Skill · Version 1.0 · Updated September 30, 2026
Want the larger build? The four-agent team from the earlier video is at the agent team page.
Back to the assets