# Free AI classifiers

> Use free AI Classifier steps to score leads, sort messages, detect intent or spot messages that need a reply, then test them in one click.

Canonical: https://help.spacebrain.ai/automations/ai-classifiers/
Last modified: 2026-09-25T00:00:00.000Z

An AI Classifier step reads a lead's details or an incoming message and answers a question about it, such as "Is this a sales or support message?" or "How urgent is this?". Your automation then follows a different path depending on the answer. Classifiers are free: they use no AI credits, including when you test them. This guide is for anyone building automations that need to sort or prioritize leads and messages.

## Before you begin [#before-you-begin]

* **Trigger:** classifiers work with triggers that involve a lead, such as **Form Submitted**, **Lead Created** or **Message Received**. If your trigger isn't lead-based, the classifier doesn't appear in the step library.
* **Permission:** you need editing permission to add and test steps.
* **Know your question:** decide what you want to find out and what should happen for each answer.

## Where to find it [#where-to-find-it]

Open an automation in the editor (**Automations → All Automations**, then open the automation). Then either:

* Select **Add classifier** (marked **Free**) in the toolbar, or
* Select **Add step** and look under **Conditions**.

## Choose a preset [#choose-a-preset]

Each preset is a ready-made classifier you can adjust:

| Preset                  | What it answers                                            | Answer type                                                | Reads by default                  |
| ----------------------- | ---------------------------------------------------------- | ---------------------------------------------------------- | --------------------------------- |
| **Lead Score**          | How well does this lead fit?                               | Score, from no evidence of need to a clear need and intent | Lead source details, Company name |
| **Email Classifier**    | Is this a sales, support or other message?                 | Category                                                   | Inbound message text              |
| **Intent Detector**     | Does the person want to buy, learn more or something else? | Category                                                   | Lead source details               |
| **Priority Classifier** | How urgent is this message?                                | Score, from routine to an immediate deadline               | Inbound message text              |
| **Needs Reply**         | Does this message ask a question or need a response?       | Yes / No                                                   | Inbound message text              |
| **Custom Classifier**   | A yes/no question you write yourself                       | Yes / No                                                   | Lead source details               |

## Set up the classifier [#set-up-the-classifier]

Select the classifier on the canvas to open its settings.

1. **Input fields:** tick what the classifier should read. Choices include **Lead source details**, **Company name**, **Lead source**, **Lead status**, **Inbound message text** and your custom lead fields.
2. For each **Answer**:
   * **Name:** a short label, such as "Message type".
   * **Answer type:** **Yes / No**, **Category** or **Score**.
   * **What should it assess?:** the question in plain language.
   * For **Category**, list up to eight **Categories**, each with an ID (lowercase letters, numbers and underscores) and a description.
   * For **Score**, list the **Score levels, lowest to highest**.
   * **Acceptance threshold:** how confident the AI must be, from 0.51 to 1. The default is 0.8.
3. Select **Add answer** to ask up to four questions in one step.
4. Save the step and the automation.

## Connect the two paths [#connect-the-two-paths]

Every classifier has two exits:

* **Ready:** every answer met its threshold. Continue your automation here.
* **Needs review:** at least one answer was uncertain, or the input was missing. Connect this to a safe fallback, such as creating a task for a person to check.

Later steps can use the classifier's answers as variables, under **Classifier result**. Score answers also give a 0–100 value.

## Worked example: route sales and support messages [#worked-example-route-sales-and-support-messages]

1. Create an automation with the **Message Received** trigger.
2. Select **Add classifier** and choose **Email Classifier**.
3. After **Ready**, add a **Condition**. In **Field**, choose the classifier's category answer from **Classifier result**, and check that it equals *sales*.
4. On the true path, add **Assign Owner** and **Create Task** for your sales team.
5. On the false path, add **Add Tag** "support" and **Notify a Teammate**.
6. Connect **Needs review** to **Create Task** so someone reads the message.
7. Test the classifier, then run a dry test of the whole automation before publishing.

## Test the classifier [#test-the-classifier]

Testing checks the classifier on sample text. It's free, runs no later steps and changes no records.

1. Save the step and the automation.
2. If the automation is active, pause it first. Otherwise you'll see &#x2A;*Save the step and automation, and pause the automation before testing.**
3. Under **Test classifier**, type sample text into each input field, for example "Hi, can I get a price for 20 seats?".
4. Select **Test classifier · Free**.

## What you should see [#what-you-should-see]

The result shows **Ready** or **Needs review**, a short reason, and each answer's value. For a score it also shows a lead score out of 100. The result is labelled **0 credits** and has a test receipt number. The button changes to **Test recorded**. Change the sample or settings to test again.

## If something goes wrong [#if-something-goes-wrong]

| What you see                                            | What to do                                                                                                      |
| ------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------- |
| **Add classifier** isn't shown                          | Your trigger isn't lead-based. Choose a lead trigger such as **Form Submitted** or **Message Received**.        |
| The test button is greyed out                           | Save the step and automation, pause the automation if it's active, and enter sample text in at least one field. |
| **Category IDs must be unique**                         | Give each category a different ID using lowercase letters, numbers and underscores.                             |
| Too many results go to **Needs review**                 | Make the question more specific, add better input fields, or lower the threshold slightly.                      |
| **This classifier must be free. Reload and try again.** | Reload the page and test again.                                                                                 |
| Only part of an email is classified                     | Emails that contain only HTML may supply just their subject line.                                               |

## FAQ [#faq]

### Do classifiers really cost nothing? [#do-classifiers-really-cost-nothing]

Yes. Classifier steps and their tests use no AI credits. Other AI steps in the same automation, such as AI calls or AI replies, may still use credits.

### Does a classifier stop an AI reply that's already drafted? [#does-a-classifier-stop-an-ai-reply-thats-already-drafted]

No. A classifier only decides which path the automation takes. It doesn't cancel an existing AI draft reply.

### Can I see classifier signals outside automations? [#can-i-see-classifier-signals-outside-automations]

Yes. Lead and inbox conversation pages can show current signals and offer shortcuts to add a free classifier. The exact controls depend on your workspace.

## Next step [#next-step]

Finish your automation in [the editor](/automations/workflows/), then check its first runs in [Runs and troubleshooting](/automations/runs-and-troubleshooting/).
