Steps hand their results forward, and later steps can read them. Anywhere you can write free text or a condition — an AI research prompt, a filter, a route condition, a Slack message — type @ to insert a field from an earlier step. This is how one step builds on another.

For example, a Fetch website step produces the page content, and an AI research prompt can read it: "Summarize what @Page content says about the company's product."

You can only reference fields that a step above this one actually produces. If the menu is empty, there's no earlier step to pull from yet — add a sourcing or intelligence step first. The tip under the field will say so: "Add a step before this one to pick its fields as @-mentions here."

The trigger produces fields too. Any event trigger that hands a record straight in — Deal closed, Deal created, Account created, Account added, Contact created, or Contact added — exposes that company's or person's full details as @-mentions in later steps, not just its domain. A field the lookup didn't find for a particular record is simply absent for that record, not an error.

Each option shows the field's own name, like "Company domain" or "Page content". If two steps produce a field with the same name, the source step is added as a prefix — "Companies search: Company domain". If you've added two of the same step, they're numbered — "Segmentation 1", "Segmentation 2".

An @-mention resolves per record. If the step that produces it didn't run for a given record — because the record dropped earlier, or the step found nothing — the value is empty for that record. When a value might be missing, set a fallback or filter those records out first. See Troubleshooting workflows.