How I Used ChatGPT Work as My Data Analyst to Fix My Content Strategy (The 6 Exact Prompts)

Every content creator I know has the same problem. You post, you set up a comment-to-DM keyword, people type the word, the automation fires, and then... you never look at the numbers again. This post is about how I fixed that with ChatGPT Work. If you don't have it yet, download it first, because the whole workflow depends on it being able to see your screen.

I had 40+ CTA keywords sitting in my DM automation platform. Some were pulling hundreds of emails. Some were doing nothing. A few were duplicates I set up at 1am and forgot about. And I had no idea which posts were actually worth repeating.

So I gave the whole thing to ChatGPT Work and asked it to act as my data analyst. Forty minutes later I had a branded dashboard telling me exactly what to make more of, what to stop, and what to build a paid product behind.

Here's the exact workflow, all six prompts, word for word.

Why I did this in ChatGPT Work instead of a spreadsheet

Two reasons.

It can see the platform. ChatGPT Work connects to my browser, so it could open my DM automation platform, scroll through every flow, read the run counts, capture rates and status, and map each CTA keyword back to the actual post thumbnail. I didn't export a CSV. I didn't screenshot 40 rows. It pulled the data itself.

It can use judgment. A spreadsheet will tell you a CTA has zero email captures. It won't tell you that the CTA was for a sponsored post that was never designed to capture emails, so zero is fine. That distinction is the whole difference between a useful analysis and a misleading one.

The six prompts

Run these in order, in the same conversation. Each one builds on the last.

Prompt 1: Pull the data

Do not skip to recommendations. Get a clean table first.

Act as a content conversion analyst. Review my DM automation platform and collect data for each active CTA/post.

For each row, capture:
- CTA keyword
- Automation/flow name
- Post thumbnail or post identifier
- Post topic/visual summary
- Runs or trigger count
- CTR or engagement rate
- Email captures
- Email capture rate
- Status
- Last modified date

Do not make recommendations yet. First create a clean table that maps each CTA to the actual post.

Prompt 2: Separate sponsored from organic

This is the step most people miss, and it poisons everything downstream. A sponsored post with a brand-name keyword and no email capture is not a failed post. It's a different kind of post. Get it out of the dataset before you judge your own content.

Before analyzing performance, classify each CTA as organic/owned, sponsored/paid, duplicate/test, or unclear.

Use judgment, not just email capture:
- Company or brand-name CTAs are likely sponsored
- Paid ad/resource delivery paths are sponsored
- Duplicate stopped flows are setup noise, not content performance
- Zero email capture alone does not mean bad content
- Sponsored posts should be excluded from organic content strategy decisions because they may not be designed to capture email

Show me what you excluded and why before building recommendations.

Prompt 3: The three decisions

Now the real analysis. Every piece of organic content lands in exactly one bucket.

Analyze only organic/owned content.

Create three decision buckets:

1. Make More
Content I should repeat because it has strong evidence of first-time engagement, a clear audience, strong promise/resource fit, and repeatable creative structure.

2. Stop or Rework
Content I should stop publishing as-is because the package is weak, vague, redundant, attracts low-intent people, or underperforms relative to better content. This is not the same as productize.

3. Productize
Content themes that show strong demand and should have a paid digital product, offer, workshop, template, or system built behind them. Do not include anything from Stop/Rework here.

For every recommendation, include the numbers and the strategic reason.

Prompt 4: Guardrails

The first time I ran this, the AI put the same post in "Stop" and "Productize." That is a contradiction. So I added rules.

Apply these rules strictly:

- Do not put the same post in both Make More and Stop.
- Do not put Stop/Rework content in Productize.
- Productize means "this demand is strong enough to build a paid offer behind."
- Stop means "do not repeat this content package as-is."
- If none of my content has paid products attached, Productize should rank the best positive demand signals, not punish posts for lacking products.
- Use qualitative judgment alongside metrics.

That last bullet matters. If you have no paid products yet, a naive analysis will say "nothing is productized, so nothing should be productized." The guardrail flips it: rank the strongest demand signals and tell me where to build.

Prompt 5: Build the dashboard

A table in a chat window is not something you'll open again. A dashboard is.

Build me a clean branded dashboard from this analysis.

The dashboard should include:
- A sidebar with three filters: Make More, Stop/Rework, Productize
- Post thumbnails tied to each CTA
- Metrics for runs, CTR, emails, and capture rate
- A detail panel explaining the strategic read
- Clear rationale for why each item is in its category
- Sponsored/ad rows excluded from the decision views
- No overlap between Stop and Productize

Use my brand colors, typography, and tone from my website. Make it feel like an internal strategy dashboard, not a generic analytics report.

Give it your website URL. It will pull your colors and fonts and the result actually looks like yours.

Prompt 6: QA the logic

Never ship an AI analysis without making the AI audit itself.

Review the finished dashboard for logic errors.

Check:
- Does each filter show only its own content?
- Are hidden tabs actually hidden?
- Are sponsored/ad rows excluded?
- Are Stop and Productize clearly different?
- Is any post appearing in contradictory categories?
- Are recommendations based on both data and strategic judgment?
- Does every CTA map back to the correct post?

If anything is ambiguous, flag it before finalizing.

What I actually learned from mine

I won't share every row, but the shape of the results surprised me:

• My best-converting posts were not my highest-reach posts. The ones that went semi-viral pulled lots of comments and almost no emails. The ones that quietly did the work had a very specific promise ("comment X and I'll send you the exact template") and a very specific audience.
• Half my "underperformers" were sponsored. Once those were excluded, my organic hit rate looked completely different. I had been beating myself up over numbers that were never mine to hit.
• Productize was obvious once it was isolated. Two themes had strong demand, repeat engagement and no offer behind them. That's not a content problem. That's a missing product.

How to run this for yourself

1. Open ChatGPT Work and connect it to your browser (or to wherever your automation data lives).
2. Paste Prompt 1 and let it pull everything. Check the table. Fix any CTA that got mapped to the wrong post before moving on.
3. Run Prompts 2 through 4 in order. Read the exclusions in Prompt 2 carefully. This is where the analysis gets honest.
4. Run Prompt 5 with your website URL.
5. Run Prompt 6 and actually read the flags.

The whole thing takes under an hour. Then you have a strategy document instead of a feeling.

If you try this and it tells you something surprising about your content, I want to hear about it. Reply to any of my emails or DM me. And if you want more workflows like this, the prompt packs in my Skool community are exactly this kind of thing: the exact prompts, the guardrails, and the reasons behind them.