GTMcraft Plays · The Train-On-You Play

AI trained on the internet sounds like the internet; train it on your best work.

Play visual

claude-train-on-you-signal.html

The 90-second read

You use Claude to write faster, and the drafts come out clean but hollow. They read like everyone else because the model learned from the open web, not from you. The job is to make AI sound like your best self, not the internet average. That breaks the moment you skip your own data.

• Train Claude on your top performers; weak inputs give weak output even from 400 posts.
• Examples teach voice; instructions alone cannot.
• Ruben Hassid extracts his posts, builds a report, then saves a personal skill.
• You cannot automate what you cannot articulate; own the voice brief.
• Keep the last 20% and the send call human, always.

Read time: about 5 minutes. First step runnable Monday morning.

The #1 move this week

Pull your 30 best-performing posts into one file, ranked by engagement.

Not your latest posts. Your genuine top performers. Hand that sheet to Claude, have it study what actually worked for you, then save the patterns as a reusable skill you invoke every time.

Start here: extract your posts with a scraper, rank by real engagement, and keep only the true winners. You need a spreadsheet with text, format, and engagement columns to feed the model. Aim for 30 or more posts you read back and recognize as you.

Ignore this week: chasing volume to hit a post count. Fewer strong posts beat many weak ones, even against 400.