More and more advertisers are doing the same thing when their ads stop working: they export their data from Ads Manager, paste it into ChatGPT or Claude, and ask what is wrong.
It is a reasonable idea. These models can read a table, spot patterns and explain them in plain language. But anyone who has tried it knows the answers are often disappointing: generic advice about testing creatives and refining audiences, confident statements based on ads with almost no spend, and different conclusions every time you ask.
The problem is not the AI. It is what the AI receives.
The three ways people do it
1. Exporting a report and pasting it
The most common approach. You export a table from Ads Manager and paste it, or upload the file, into the chat.
2. Screenshots
Quick, but the AI only sees what fits on the screen, and numbers can be misread.
3. Connecting the AI to the ad account
Some AI apps can now connect directly to Meta through connectors based on MCP (Model Context Protocol). The AI can pull data from your account on its own when you ask a question.
All three can work for simple questions. All three struggle with the question you actually care about: why is performance changing, and what should you do about it?
Why the answers are often generic
The AI does not know your targets
A $35 cost per purchase is great for one store and a disaster for another. Unless you tell the AI your target CPA, it has to guess what “good” means, and it will usually fall back on vague benchmarks.
It sees one period, not a change
A single export shows what happened, not what changed. Without a comparison period, the AI cannot tell you what got worse, only what looks low.
It has no rule for what counts as a result
Is an ad with $12 of spend and no purchases a failure? Without a spend threshold, the AI might tell you to pause ads that simply have not had a chance yet, or praise ads that got lucky with one sale.
It is missing context
Raw exports often lack the ad copy, the relationship between ads, ad sets and campaigns, or the funnel steps. A direct connection gives the AI access to data, but it still receives lists of IDs and numbers, and has to invent its own logic for what matters. That is why answers change from one question to the next.
How to get better answers if you do it yourself
If you want to use ChatGPT or Claude with an export, you can improve the results a lot:
- Export two equal periods (for example the last 7 days and the previous 7) at ad level, and tell the AI which is which.
- State your targets explicitly: target CPA, and the minimum spend an ad needs before it can be judged.
- Include the structure: campaign and ad set names next to each ad, so the AI knows what belongs together.
- Include the funnel columns: landing page views, adds to cart, checkouts and purchases, not just clicks and spend.
- Ask specific questions. “Which ads above my spend threshold have a CPA over $30, and what changed for them compared to last week?” works far better than “audit my ads.”
It takes preparation every time, but it turns generic advice into something you can use.
What changes when the data is prepared first
The preparation above is exactly what makes an AI audit useful: targets, comparisons, classification and context. When that work is done before the AI reads anything, the AI can reason about your account the way an experienced analyst would, because it starts from the same structured picture.
That is how our AI Analyst works. Our Meta Ads Analyzer connects to your ad account with read-only access and, before the AI answers anything:
- Classifies every ad against your CPA target and test spend threshold, so winners, underperformers and scaling candidates are already identified
- Calculates period-over-period comparisons at campaign, ad set and ad level
- Ranks audiences by efficiency
- Bundles the ad copy with each ad’s performance, so the AI can connect what the ad says with how it performs
You ask in plain language and get answers grounded in your actual account. We wrote more about this shift in AI Is Changing How Meta Advertisers Get Answers.
Chat with an AI that already understands your account. Connect your Meta ad account, set your targets, and ask our AI Analyst anything. Start your 3-day free trial
Already using Claude or Gemini? Use it with better data
If you prefer working in your own AI assistant, our Meta Ads Analyzer includes an MCP connection. Instead of giving Claude, Gemini or another compatible agent raw numbers from Meta, it passes the fully enriched account context: the classifications, comparisons and rankings described above. Your assistant works with the same structured picture our AI Analyst uses.
Frequently asked questions
Can ChatGPT audit my Facebook ads?
Yes, if you give it the data. You can upload an export from Ads Manager, or use an AI app that connects to Meta. The quality of the answer depends heavily on whether the AI also knows your targets, has a comparison period and has enough context about your account structure.
Is it safe to connect AI to my Meta ad account?
Check what access you are granting. Our Meta Ads Analyzer uses read-only access, so it can read your data but can never change campaigns, budgets or ads.
Can AI make changes to my Meta campaigns?
Some tools can. Ours deliberately does not. Our AI Analyst tells you what to change and why, and you decide what to do in Ads Manager.
Which is better for a Meta ads audit, ChatGPT or Claude?
Both are capable. The bigger difference comes from the data they receive. A well-prepared dataset with targets and comparisons will give better answers in either model than a raw export will.
Do I still need to understand Meta ads metrics?
It helps, but you do not need to be an expert. A good AI audit explains what happened and what to do in plain language. If you want the basics, see Meta Ads Metrics Explained.
Get AI answers based on your real account, not generic advice. 3 days free, then $59 per ad account per month. Try our Meta Ads Analyzer or see what’s included.
Related: Why Your Meta Ads Stopped Performing (and How to Find the Real Cause)