n8n AI Transform nodeConfigure AI Transform in n8n.

First scenario, tidying contact data before it is stored. A form or an import gives you firstname and lastname in separate fields, and the sheet you feed expects one column. The placeholder in the node says it plainly: merge firstname and lastname into a field details.name and sort by email. Generate the code once, and every run of your Google Sheets append gets the same clean payload.

Second scenario, flattening an API response. A call made with HTTP Request comes back with values buried two levels down. Instructions use dot notation to reach them, the way the node's own description asks: refer to a nested field as address.street. The generated snippet pulls those values up to the top level, and the nodes after it can read them with a plain expression such as {{ $json.street }}, the n8n syntax for reusing a field from the item at hand.

Third scenario, preparing a message body. Before an Gmail send, you want the subject built from two fields and the empty ones dropped. Writing that by hand is five minutes of fiddling with string concatenation. Describing it is one sentence, and the result is readable code you can inspect before the workflow goes live.

When to prefer the alternative. The Code node is the better choice as soon as you want to keep editing the logic: the snippet AI Transform produces is read-only inside the node, and the only ways forward are rewriting the prompt or copying the code into a Code node. Loops you tune by hand, helper functions, anything you expect to revisit in three months, start there.

The limits are worth knowing before you build on it. The prompt stays in plain English and under 500 characters, which rules out a full specification and rewards one clear instruction at a time. The node is at version 1, with those two parameters and nothing else to set. And it runs downstream only: a Schedule Trigger, a Webhook or the trigger of a tool has to open the workflow first.