> ## Documentation Index
> Fetch the complete documentation index at: https://docs.growthxai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Routing

> Describe each path in plain words and let AI sort leads down the right one, using the whole profile.

**AI routing** is a branching step where you describe each path in ordinary language instead of writing rules.

> **Founders and C-level** · **Agencies, not product companies** · **Everything else**

Add it from the **AI** group in the step palette.

## Setting it up

Each route has a **label** (the branch name on the canvas) and a **description** (what the AI is matching on). Be concrete:

* Good: *"Agencies and consultancies that sell services to other businesses."*
* Weak: *"Good fit."*

An **everything else** branch is always present and cannot be removed. Every lead ends up somewhere.

## What the AI sees

The whole picture, not a handful of fields:

* Lead fields: name, title, company, headline, location
* **Custom fields**
* **[Enriched profile data](/ai-and-data/profile-enrichment)**: About, past roles, education, skills, follower count
* **Recent posts**, when the profile has them

## The decision

* Made **once per lead**, when they reach the step, and then stored.
* Stored with a short **reason** and the **facts it relied on**, both visible on the lead's timeline.
* A lead re-entering the sequence later is decided again from whatever is known then.

While the decision is being made the enrollment shows **waiting** — no teammate needs to do anything.

## Test before you publish

**Test on 20 leads** runs the routing on a sample and shows the split and the reasons, without changing anything. Do this before publishing. A route that catches 19 of 20 leads is a route description that needs narrowing.

## Why this one doesn't need approval

Everything else AI-written needs a person to approve it, because it is text that goes to a prospect. Routing chooses a **path**, not words. The text on each path is your text, reviewed as usual. The test run is the review.

## Tips

* Keep routes **mutually exclusive**. Overlapping descriptions make the split unstable.
* Three or four routes is plenty. Ten route descriptions are hard to keep distinct.
* Put the most specific route first and let *everything else* catch the long tail.
* Routing on data you have is more reliable than routing on inference. If you can write it as a **Condition** step, do that instead — it is instant and free.

## Related articles

* [Sequence Nodes](/sequences/creating-managing-sequences/sequence-nodes)
* [AI Variables and the Review Table](/ai-and-data/ai-variables-and-review)
* [Profile Enrichment](/ai-and-data/profile-enrichment)