Everything here works in a browser tab. Run manual loops that don't drift, set up reusable projects, loop over documents, use tools, and converge multi-turn.
Welcome to Running Loops in ChatGPT and Claude. You now have the three foundations — you know what a loop is, you can write a goal it can finish, and you can build a check that catches a false “done”. This module puts those to work in the tools most people already have open: ChatGPT and Claude, right in a browser tab. There’s no command line here and no code required.
You’ll learn the practical craft of running a loop by hand without it drifting off course — because the biggest failure mode in a long chat is the AI slowly forgetting what you actually asked for. You’ll set up Projects and custom instructions so a loop’s goal, brief, and check don’t have to be re-typed every time. You’ll run loops over documents — long inputs, uploads, and material too big to paste in one go — by breaking the work into passes. You’ll put the built-in tools to work, like web search and file analysis, so a loop can go fetch what it’s missing. And you’ll practice the multi-turn patterns — refine, expand, critique, converge — that turn a messy conversation into a deliberate loop.
A caution up front, and a promise: exact menus and buttons in these tools change often, so every tool-specific step here is marked and the loop ideas underneath it will outlast the current interface. The module project is a full research-and-report loop that produces Harborlight’s competitor brief. Start with Lesson 1, on keeping a manual loop from drifting.
Complete all 5 lessons to finish the Running Loops in ChatGPT and Claude module.