How to Version Control Prompts in a Small Marketing Team
How to version control prompts in a small marketing team using git, spreadsheets, or dedicated tools — with the metadata and rollback process that actually work.
Topic
Prompting
6 articles.
How to Test One Prompt Against 20 Inputs at Once (Cheap Prompt Regression)
Run one prompt across 20 inputs at once with a CSV loop, promptfoo, or n8n. Concrete setup, grading methods, and cost math for cheap prompt regression.
Why ChatGPT Forgets Your Instructions Mid-Chat, and How to Re-Anchor Them
How to stop ChatGPT from forgetting instructions in a long conversation: why instruction drift happens inside the context window, and how to re-anchor rules so they hold.
How to Write a System Prompt That Forces JSON-Only Output (With Test Cases)
A system prompt alone can't guarantee valid JSON, but the right one gets you close. How to write a system prompt that forces JSON-only output, the test cases that break weak prompts, and the schema modes to fall back on when prompting isn't enough.
How to Stop ChatGPT From Using Em Dashes (Prompts That Actually Hold)
Every trick to stop ChatGPT from using em dashes in its output, and why most of them leak. Replacement rules, custom instructions, logit_bias at the API level, and the one post-processing step that is actually deterministic.
Context Length Exceeded? How to Summarize a Long Transcript Without Breaking the Model
The context length exceeded error when summarizing a long transcript is an arithmetic problem, not a model problem. Here's how input, output and reserved tokens are counted, and the four mechanisms that bring a transcript back under the limit.