Explain or audit a GoHighLevel workflow with ChatGPT or Claude

For one workflow in an account you can open, you may not need an outside assistant at all. In the workflow builder, HighLevel’s Workflow AI (opens in a new tab) has a Describe This Workflow button that explains the open workflow from start to finish, and workflows that came in through a Snapshot get an Explain This Workflow button at the top. Its Chat mode also answers from live data (opens in a new tab): whether the trigger is firing, why contacts were rejected, where they drop off.

An outside assistant, working on exported files, is the better tool when:

• you don’t have access to the account, and the client sent you the exports instead;
• you want to compare a workflow with an older version than HighLevel’s Version History keeps, which is the last 10 saved versions or 30 days;
• you are working across sub-accounts, which HighLevel’s assistant doesn’t reach;
• you want your own checklist, your own format, or a document you can hand to a client.

GoHighLevel has no export button. The GHL Workflow Backup & Audit extension exports a workflow, or a whole folder of them, as JSON: the triggers and their filters, every step with its settings, and the workflow’s own settings. A screenshot shows the assistant the shape of a workflow; the file shows it every setting.

Set your own Private Integration Token in the extension’s Options before you export. With it, tags, pipelines, custom fields, calendars and users come through as names in references, and the assistant can say “adds the tag Hot Lead” instead of quoting an id. Without it, the assistant sees only ids and has to guess what they are.

Exporting the workflow, step by step, is in how to export a GoHighLevel workflow.

An export holds the workflow, not your contacts. It is not redacted, though: whatever a step holds is in the file. Before it goes into any chat, open it in a text editor and search for:

• Secrets: Authorization, Bearer, apiKey, token and secret. Webhook headers and custom code steps are where API keys usually sit. Replace each value with something like REMOVED.
• Personal details typed into steps or filters, such as a staff member’s phone number or a specific customer’s email.

Then check your AI plan’s data settings: whether chats are used to train models, and how long they are kept. For a client’s account, their agreement with you may decide this for you.

Attach the file, then paste one of these. Each starts with the same two sentences telling the assistant what the file is, so it reads the structure instead of guessing.

The audit prompt is a short version of the GoHighLevel workflow audit checklist. For a whole account, attach several exports at once and ask which workflows add or filter on a given tag, or which ones send from a given number.

• It can’t see the account. A file shows what the workflow says, not whether the tag it filters on is still in use, the calendar still has slots, or the webhook endpoint still answers.
• It has no run data. Why contacts stopped at step 4 is in the Execution Logs, not the file.
• It fills gaps confidently. An id it can’t name, or a step type it doesn’t know, may get a plausible explanation anyway. The prompts above ask it to say what it couldn’t work out; take that list seriously.

Treat every finding as a lead. Open the step in the builder and confirm it before you change anything.

You can ask an assistant to change the JSON, for example to rewrite every message in a sequence, and import the result back with the extension. The importer refuses a file whose steps no longer have the shape of workflow steps, but it can’t tell whether an edit makes sense. So:

1. Keep the original export.
2. Import the edited file into a new workflow, or one set to Draft, never over the live one.
3. Open every changed step in the builder and test it with a test contact.
4. Publish only once it checks out.

The details of importing, including what gets overwritten, are in how to import a workflow into GoHighLevel.