Your AI is a pretty good copywriter with one bad habit. Ask it to fix an ad and it grabs the most expected move every single time, which happens to be the exact move your readers learned to scroll past years ago.
So you get "make it punchier." You get three more adjectives. You get a headline that any of your competitors could run without changing a word.
The Persuasion Taxonomy MCP fixes that habit, and it's free.
What it is
It's a free add-on for the AI you already use. You paste one address into Claude, ChatGPT or Gemini, and from then on your AI can look things up in The Persuasion Taxonomy, our catalog of 767 persuasion techniques, all documented from real ads.
You don't need an account and you don't need an API key. It runs inside your own AI, so it adds nothing to your bill.
What it does
Once it's added, you ask for help the way you normally would, and your AI has 5 new jobs it can do properly:
- Plan a piece of copy before you write it. You tell it what the reader has to do, and it tells you which questions the copy has to answer to get them there.
- Work out why something isn't converting. You start from what you can see, like people clicking and bouncing, and it works back to what the copy is missing.
- Check headlines and subject lines. It compares your options and tells you which one is doing its job and what to fix on the others.
- Check your claims. It catches the "best" and "#1" lines that readers discount on sight.
- Plan an A/B test and read the result. It does the math, so you know how long the test has to run before you start it.
Why it's useful
Everyone who reads your ad is silently asking 9 questions. Why am I even reading this? Is this written for me? Can I trust the claim? What's stopping me from saying yes? Why should I act right now? And 4 more.
Copy sells when it answers the questions its job needs. Copy flops when it skips one, and you can't tell which one by squinting at it.
That's the part this tool hands your AI. Now your AI can say "your page never answers 'Can I trust the claim?', and here are 2 named ways real ads have answered it," with a link to each technique so you can see the real examples. You get something you can act on in the next 10 minutes.
How to add it
The address is the same everywhere:
https://taxonomy.coppica.com/mcp
Claude. Open Customize, then Connectors. Choose Add, then Add custom connector. Name it Persuasion Taxonomy, paste the address, then choose Continue and Add. Claude's free plan allows 1 custom connector. In Claude Code, run claude mcp add --transport http persuasion-taxonomy https://taxonomy.coppica.com/mcp.
ChatGPT. On the web, open Plugins at chatgpt.com/plugins, choose the plus button, then Add custom MCP server. Name it, paste the address as the connection URL, leave authentication off, tick the acknowledgement and choose Create. OpenAI offers this on its Business, Enterprise and Edu plans, and on Pro with read-only access.
Gemini. In Gemini CLI, run gemini mcp add --transport http persuasion-taxonomy https://taxonomy.coppica.com/mcp. In the Gemini app, custom MCP servers are added by a Gemini Enterprise admin.
The steps and each company's own guide are on our Free Tools page, and we keep them current there.
3 things to ask it
We ran all 3 of these against the live tool. What you see under each one is what it sent back, trimmed for length. Your AI will put it in its own words.
1. "People read my landing page and don't buy. Why?"
Before you touch the copy, rule out the things copy can't fix. Check the offer itself first, meaning the price, the terms and the fit, and make sure the call to action is easy to find and easy to use.
If those check out, the problem is most likely a question the copy leaves unanswered. Here are the suspects, most likely first.
1. "Can I trust the claim?" They hear the promise, but they don't believe it. To check, go through every big claim and look for the evidence sitting right next to it: a number, a name, a demonstration.
2. "What's stopping me from saying yes?" An objection nobody answered, like the price, the time or "will it work for me?", is standing in the way of yes.
It goes on to name 2 techniques for each of those, lists 2 more suspects, and then tells you how to test whether it guessed right.
2. "Which of these 3 subject lines should I send?"
We gave it a made-up coffee subscription and 3 subject lines. It didn't flatter any of them:
None of these does its job yet. B comes closest, so apply the fix below and check again.
A. "Our biggest sale of the year starts now" doesn't do its job. Any competitor could send this word for word, because nothing in it belongs to this product, this reader, this offer or your proof.
C. "Half off your first bag, roasted this week" doesn't do its job. The words that answer "Can I trust the claim?" ("roasted this week") fall past the cutoff, and since only about 40 characters show in a mobile inbox, most readers never see them. What readers actually see is "Half off your first bag, roasted this we…".
And then it says the quiet part out loud: "Nobody can pick the winner from the words alone, and that includes this check." It wants you to test.
3. "I get 400 visitors a day and convert 2.5%. How long do I need to run a test?"
To detect a 20% lift on a 2.50% conversion rate, which means going from 2.50% to 3.00%, at 95% confidence and 80% power, you need 16,792 visitors for each version, or 33,584 in total.
At 400 visitors a day, that takes about 84 days.
That is more than eight weeks, which is a long time to wait, and you have three ways to shorten it. You can test a bigger difference, meaning two genuinely different answers to a reader question instead of a change in wording.
That last line is worth the whole install. Most tests fail because the 2 versions were never different enough to tell apart.