On this page
TLDR:
- ChatGPT cannot open the Meta Ad Library; its browsing returns articles. Connect Dovey and it can search 100,000+ ads by meaning and open any advertiser's account.
- Ask questions with a shape: a category for a Dovey grid, "which angles perform best" for a chart, a brand or URL for an account.
- Read days running and versions. Dovey tells ChatGPT to state sample sizes and never to invent performance numbers.
- Turn winners into briefs: structure first, then Dovey's remix writes a version for your product with three hooks.
- Every ad cited through Dovey links to its Ad Library page.
What ChatGPT gets wrong without data
Ask it for the best supplement ads and it writes a confident list from training data with no dates and no way to check. The library changes daily. The connector turns the same question into a live search with start dates and version counts.
How do I connect ChatGPT to the ads data?
ChatGPT's browsing returns pages about ads, not the ads themselves, and it cannot open the Meta Ad Library. A connector fixes that: Dovey's server speaks MCP, and ChatGPT's custom connectors use it.
- Open ChatGPT, then Settings, then Connectors (on some plans this sits under Apps and Connectors, and Developer mode has to be on).
- Choose Create, name it Dovey, and paste the address:
https://api.makedovey.com/mcp - Sign in to Dovey when asked. Search is free; the agent and creator lists are on a plan.
- In a new chat, enable the Dovey connector from the tools menu. ChatGPT then uses it when a question needs the data.
The prompts that work
A search:
supplement ads that lead with a discount
A question over the data, answered with a chart and a sample size:
Which angle performs best for supplement ads: offer, proof or problem-solution? Give the median days and the count for each.
A competitor:
Open gruns.co's ads, top performers first. What do they lead with, and how much of it is video?
DTC Mode2 variations
One ad, in full:
Give me the script of the top-performing one, its hook, and the sibling versions and what changed between them.
A brief:
Take that ad's structure (hook, angle, offer, proof, CTA) and write it for a magnesium gummy. Three hook options, one in the customer's words.
What to believe, and what not to
Believe the days and the versions. An ad at 300 days with 12 versions has been paying for itself for ten months and the brand keeps scaling it. Believe "of the index" and "3 of 5 ads" when ChatGPT says them; the instructions require the sample size under 20.
Do not believe a CTR, a CPA or a spend for an ordinary ad, from ChatGPT or anyone. Meta does not publish them.
A Monday routine
What started in my category this week with more than three versions?
Compare my three competitors: active ads, share of video, the angle each leads with.
Three hooks from this week's top performers, rewritten for my product.
FAQ
Can ChatGPT search the Meta Ad Library?
Not directly. With a connector to an index of the library it can search by meaning, open accounts and count across the ads, with dates and links back to the library.
Does ChatGPT know which ads perform best?
It knows run time and version counts, the public signals. Meta publishes no performance for ordinary ads and ChatGPT is instructed not to invent them.
See every ad that works. Then make yours.
Search 100,000+ Meta ads by what they mean, with the top performers first.
Search the ads