An 8,000-Token Prompt That Interviews You Before It Writes
It forces any LLM to interview you about your reader before writing a word, then audits its own output for AI tells.

What it does
ai-copywriter is not an app; it is a roughly 8,000-token Markdown skill file that you feed into an LLM harness. It instructs the model to write conversion copy—headlines, microcopy, blog posts—by first interrogating you about the reader’s emotional state and the simplest possible explanation, then running the result through a 33-pattern audit to strip out AI-generated prose tics. Think of it as a very opinionated style guide that happens to do the writing.
The interesting bit The skill treats laziness as a feature, not a bug: if you hand it a weak product story or a vague demographic, it keeps asking follow-up questions instead of generating generic text. It also borrows a humanizer framework from another project to catch Wikipedia-catalogued “signs of AI writing,” so the copywriting and cleanup phases share the same underlying rules rather than fighting each other.
Key highlights
- Single file, zero dependencies:
SKILL.mdruns anywhere you can paste text, from Claude Code to ChatGPT custom GPTs. - Built on two distinct sources: enso.bot communication research for copy strategy, and blader’s Humanizer for the 33 detectable AI-writing patterns.
- Refuses to hallucinate product facts; if the best headline needs a number, it demands the real one from you.
- Includes a dedicated strategic-blog template with a strict intake phase that blocks trend pieces built on vapor.
- Supports voice calibration by analyzing your own writing samples and applying their rhythm to the rewrite.
Caveats
- It is essentially a very long, structured prompt, not executable code; you still need an LLM and a compatible harness to run it.
- The interview process is aggressive: expect to be grilled on your ICP’s 11 p.m. search queries and your category’s table stakes before seeing any copy.
- At ~8,000 tokens, it eats a meaningful slice of context window, which could matter on smaller models.
Verdict Worth a look if you treat LLMs as draft generators but hate the default “Unlock the Ultimate Guide” voice. Skip it if you are looking for a self-hosted API or a tool that writes copy without asking you questions first.
Frequently asked
- What is mikiarlo3/ai-copywriter?
- It forces any LLM to interview you about your reader before writing a word, then audits its own output for AI tells.
- Is ai-copywriter open source?
- Yes — mikiarlo3/ai-copywriter is open source, released under the MIT license.
- What language is ai-copywriter written in?
- mikiarlo3/ai-copywriter is primarily written in Python.
- How popular is ai-copywriter?
- mikiarlo3/ai-copywriter has 1k stars on GitHub.
- Where can I find ai-copywriter?
- mikiarlo3/ai-copywriter is on GitHub at https://github.com/mikiarlo3/ai-copywriter.