---
title: "AI Classify"
description: "Add an AI Classify step to your form's workflow. AI answers your questions about each submission, and later steps use the answers to filter spam, qualify leads and route requests."
url: https://forminit.com/docs/ai-classify/
---

# AI Classify

AI Classify is a step in your form's [Workflow](https://forminit.com/docs/workflow/). When a submission arrives, it sends the parts of the submission you choose to AI, asks the questions you set up, and saves one answer per question on the submission. Steps placed after it use those answers to route, filter or enrich the submission.

![AI Classify step settings with What should AI read? and the ready-made question sets, next to a submission showing its AI answers for is_spam, topic, urgency and is_frustrated](https://forminit.com/blog/changelog/2026-11-01/ai-classify-ss.webp)

> **Use Cases**: Filtering spam before anyone reads it, qualifying leads by buying intent and company size, triaging support requests by topic and urgency, routing each request to the right team, spotting refunds, cancellations, bug reports and churn risk.

```txt
              ╭───────────────────────╮
              │  Submission received  │
              ╰───────────┬───────────╯
              ╭───────────▼───────────╮
              │      AI Classify      │
              │     Is this spam?     │
              ╰───────────┬───────────╯
              ╭───────────▼───────────╮
              │       If / else       │
              │  is_spam equals true  │
              ╰───────────┬───────────╯
             ╭────────────┴────────────╮
             │ If                      │ Else
  ╭──────────▼──────────╮   ╭──────────▼──────────╮
  │     Mark as spam    │   │      Send email     │
  ╰─────────────────────╯   │     notification    │
                            ╰─────────────────────╯
```

You decide what AI reads and what it answers. Forminit stores the answers with the submission, shows them in the dashboard and makes them available to every later step as variables.

---

## How AI Classify works

Each AI Classify step has two parts: a text that tells AI what to read, and a list of questions. For each submission, AI reads the text, answers every question and Forminit saves one answer per question under its **Question ID**.

```txt
 What should AI read?       Questions                   Answers
 ────────────────────────   ─────────────────────────   ─────────────────────────────
 Company: Northwind Ltd     Is this submission spam?    ai.classify.is_spam = false
 Message: Can we book a  →  Which team should handle →  ai.classify.team = sales
 demo for a team of 40?     this request?
                            How ready is the sender     ai.classify.buying_intent = 2
                            to buy?
```

A later **If / else** step can then check `ai.classify.team`, an email subject can include `{{ai.classify.buying_intent}}`, and a Google Sheets step can write `is_spam` to a column.

Three things to know before you start:

- **Answers have a type.** A question is a **Choice** (one option from a list), a **Score** (a number on a scale you define) or **Yes / No** (`true` or `false`). The type decides which comparisons later steps offer.
- **Order matters.** Only steps placed after AI Classify on the same path can use its answers. Put it before the **If / else** that should route on them.
- **Nothing happens until you publish.** Like every workflow step, AI Classify runs for new submissions only after you select **Publish**.

---

## Add AI Classify to a workflow

1. In your Forminit dashboard, open your form and select **Workflow** in the form header.

2. Select **+** where the step should run, usually straight beneath **Submission received**, then choose **AI Classify** from the **AI** group of the action list.

3. Select the step to open its settings and fill in **What should AI read?**. Use **Add field** to insert the form fields your questions are about.

4. Select **Add question**. Choose **Start from scratch** for an empty question, or pick a [ready-made question set](#ready-made-question-sets) and edit it.

5. For each question, write the **Question**, pick an **Answer type** and check the **Question ID**.

6. Add the steps that should use the answers after AI Classify: an **If / else** to route on them, an email that includes them, a Google Sheets column that records them.

7. Select **Publish**. Submit your form with a sample message, then open the submission to see the answers.

The rest of this page explains each setting, how to use the answers and what to expect when the step runs.

---

## What should AI read?

This text box tells AI what to look at for each submission. It is the only part of the submission AI receives.

- Select **Add field** to insert a form field as a labelled line, for example `Message: {{message}}`. When the workflow runs, `{{message}}` is replaced with that submission's value.
- Besides form fields, **Add field** offers **Submitter Country**, the two-letter country code of the submitter's IP address, and the answers of any earlier AI Classify step on the same path.
- Type fixed facts AI should know, such as what your company sells, your refund policy or the roles you are hiring for. Keep questions out of this box. They have their own section below.

A lead form might read:

```text
Company: {{company}}
Job title: {{job_title}}
Message: {{message}}
What we sell: Project management software for teams
```

Keep these rules in mind:

- Add only the fields your questions are about. Extra fields make answers less accurate.
- A field that a submission does not include is read as empty text.
- If every field is empty for a submission, AI is not asked. All answers are empty and the result shows **No input**.
- The text can be up to 8,000 characters. A longer text cannot be published. Forminit never shortens it silently.

---

## Questions

Each step has 1 to 10 questions. A question has three parts:

| Part | What it is |
| - | - |
| **Question** | What you want to know, in plain words. AI reads it literally, so ask one clear thing and name exactly what to look at. |
| **Answer type** | **Choice**, **Score** or **Yes / No**. See below. |
| **Question ID** | A short name in `snake_case`, for example `is_spam`. It names the answer for later steps. AI never sees it. |

### Answer types

| Answer type | What AI does | Answer value |
| - | - | - |
| **Choice** | Picks one option from a list of 2 to 10 options. Each option has a name and an optional description of when to pick it. | The name of the chosen option, for example `sales`. |
| **Score** | Rates the submission on 2 to 10 ordered levels, lowest first. Levels are numbered from 0. | A number from 0 to the top level. It can be fractional, for example `1.4`. Empty when AI is less sure than the **Minimum confidence**. |
| **Yes / No** | Decides whether something is true. You can describe what Yes and No mean, both or neither. | `true` or `false`. It is `true` only when AI's confidence reaches the **Yes threshold**. |

#### Choice

Use Choice to sort submissions into categories: which team, which topic, which language.

Give each option a short `snake_case` name and a one-line description of when to pick it. The description is what AI reads, so make it concrete.

| Option | Description |
| - | - |
| `sales` | Pricing, demos, buying or upgrading |
| `support` | Help using the product or a problem with it |
| `billing` | Invoices, payments and refunds |
| `other` | Anything that fits none of the options above |

A Choice question always produces one option, so include an option such as `other` or `not_stated` when a submission may fit none of the others. Without it, AI has to force every submission into a category.

#### Score

Use Score when the answer is a matter of degree: urgency, buying intent, sentiment, severity.

Describe each level as a situation rather than a degree. "It needs action today. The sender is blocked or losing money" gives AI something to check. "Very urgent" does not.

| Level | Description |
| - | - |
| 0 | It can wait. Nothing is blocked for the sender |
| 1 | It should be handled soon. The sender is slowed down |
| 2 | It needs action today. The sender is blocked or losing money |

The answer is a number from 0 to the top level. It can be fractional, such as `1.4`, when a submission sits between two levels. In an **If / else** condition, compare a Score with **greater than or equal**, **less than** or **between** rather than **equals**.

When AI is less sure than the **Minimum confidence** (80% by default), the answer is left empty and the submission shows **Needs review**. Route these submissions with a **does not exist** condition, for example to a human inbox.

#### Yes / No

Use Yes / No for a single check: is this spam, does the sender want a demo, does the report include steps to reproduce.

Phrase the question so that **Yes is the case you want to catch**, and avoid negatives such as "free of". "Is the sender asking for money back?" works. "Is this message free of refund requests?" does not.

You can describe what Yes and No mean:

| | Description |
| - | - |
| **Yes means** | The sender asks for a refund, a chargeback or money back |
| **No means** | Any other request, including questions about the refund policy |

The answer is `true` only when AI's confidence reaches the **Yes threshold** (80% by default). Otherwise it is `false`.

- **Raise the threshold** when a wrong Yes is costly. Marking a real customer's message as spam is worse than reading one more spam message.
- **Lower the threshold** when a missed case is costly. The ready-made **Property maintenance** set uses 50% for possible emergencies, because a false alarm costs a phone call and a missed gas leak costs much more.

### Confidence settings

Two answer types have a confidence slider, and they behave differently:

| Setting | Applies to | Default | When AI's confidence is below it |
| - | - | - | - |
| **Minimum confidence** | Score | 80% | The answer is empty and the submission shows **Needs review** |
| **Yes threshold** | Yes / No | 80% | The answer is `false` |

Choice questions have no confidence setting. AI always picks an option.

### Write questions AI can answer

- **Ask one thing per question.** Split "Is this urgent and is the sender upset?" into two questions.
- **Name what to look at.** "How well does the applicant's experience match the role they applied for?" tells AI to compare two fields. "Is this a good candidate?" leaves it guessing.
- **Give AI the facts it needs.** If the answer depends on your pricing, policy or service area, put that in **What should AI read?** as a fixed line.
- **Describe options and levels as situations.** AI matches the submission against your descriptions, so the better they describe real submissions, the better the answers.
- **Start from a ready-made set.** The sets below are written this way. Edit one rather than starting blank.

---

## Ready-made question sets

**Add question** opens a menu. **Start from scratch** adds one empty question. Every other entry adds a ready-made set of questions that you can edit, reorder or remove afterwards. When you choose a set, the **What should AI read?** placeholder suggests matching fields from your form.

| General | Industry |
| - | - |
| [Spam filter](#spam-filter) | [Orders and returns](#orders-and-returns) |
| [Lead qualification](#lead-qualification) | [Real estate inquiries](#real-estate-inquiries) |
| [Quote requests](#quote-requests) | [Legal intake](#legal-intake) |
| [Support triage](#support-triage) | [Insurance claims](#insurance-claims) |
| [Route to a team](#route-to-a-team) | [Nonprofit inquiries](#nonprofit-inquiries) |
| [Refunds and cancellations](#refunds-and-cancellations) | [Renovation and installation](#renovation-and-installation) |
| [Job applications](#job-applications) | [Print, signage and manufacturing](#print-signage-and-manufacturing) |
| [Customer feedback](#customer-feedback) | [Property maintenance](#property-maintenance) |
| [Churn risk](#churn-risk) | [Catering and events](#catering-and-events) |
| [Bug reports](#bug-reports) | [Language routing](#language-routing) |
| [Business inquiries](#business-inquiries) | [Appointment requests](#appointment-requests) |
| [Agency enquiries](#agency-enquiries) | |
| [Onboarding answers](#onboarding-answers) | |
| [Safety check](#safety-check) | |

A set can be added only when the step has enough free question slots. Each step holds up to 10 questions.

Every set is listed with its questions, options and levels in the [reference at the end of this page](#ready-made-question-sets-reference).

---

## Use the answers in later steps

Each answer is a variable named `ai.classify.<question ID>`. A question with the ID `is_spam` produces `{{ai.classify.is_spam}}`.

Only steps placed **after** the AI Classify step on the same path can use its answers. A step on another branch, or above the AI step, does not see them.

| Step | How it uses the answers |
| - | - |
| **If / else** | AI answers are listed first in the field menu, tagged **AI**. Comparisons depend on the answer type, see the table below. |
| **Send email notification**, **Send autoresponder email** | Insert answers into the subject or body with the variable picker. An answer cannot be used as an email address. |
| **Send webhook** | The payload includes the answers under `ai.classify`, keyed by Question ID. |
| **Send Slack notification**, **Send Discord notification** | Messages can include the answers. |
| **Send to Google Sheets** | Map an answer to a column. An empty answer leaves the cell empty. |
| **AI Classify** (a later step) | Add an earlier step's answers to **What should AI read?** with **Add field**. |

A webhook payload carries the answers like this:

```json
"ai": {
  "classify": {
    "is_spam": false,
    "team": "sales",
    "urgency": 1.4
  }
}
```

### Conditions by answer type

| Answer type | Comparisons in **If / else** |
| - | - |
| **Choice** | Equals, does not equal, is one of, is not one of |
| **Score** | Equals, does not equal, greater than, greater than or equal, less than, less than or equal, between |
| **Yes / No** | Equals `true` or `false` |
| All types | Exists, does not exist |

Use **does not exist** to route answers that are empty: a Score below the **Minimum confidence**, a submission with **No input**, or a step skipped for **Plan required**.

```txt
              ╭────────────────────────────╮
              │        AI Classify         │
              │      urgency (Score)       │
              ╰──────────────┬─────────────╯
              ╭──────────────▼─────────────╮
              │         If / else          │
              ╰──────────────┬─────────────╯
                             │
         ╭───────────────────┼───────────────────╮
         │ If: urgency       │ Else if: urgency  │ Else
         │ does not exist    │ >= 2              │
╭────────▼────────╮ ╭────────▼────────╮ ╭────────▼────────╮
│   Email triage  │ │   Notify Slack  │ │   Email support │
│      inbox      │ │                 │ │                 │
╰─────────────────╯ ╰─────────────────╯ ╰─────────────────╯
```

Put the **does not exist** branch first. Otherwise an empty answer fails every comparison and falls through to **Else**, where it is easy to miss.

### Several AI Classify steps

On Pro and higher, a workflow can hold up to 5 AI Classify steps, with up to 3 on one path. The Free plan allows one. A later step can read an earlier step's answers in its **What should AI read?** text, for example to ask a follow-up question only about submissions that were not spam.

If two earlier AI Classify steps use the same Question ID, the step that uses the answer asks you to pick which one it means.

### Rename a Question ID

Renaming a Question ID does not update steps that already use the old name. When you select **Publish**, Forminit lists the broken references so you can point each one at the new name.

---

## Example workflows

### Filter spam and route the rest to a team

**Available on all plans.**

Use one contact form for sales, support and general enquiries. This example adds the **Spam filter** and **Route to a team** sets to one AI Classify step, then routes on both answers with a single **If / else**.

1. Add **AI Classify** after **Submission received**.

2. In **What should AI read?**, use **Add field** to insert the sender's name, email and message.

3. Select **Add question**, then **Spam filter**. Select **Add question** again, then **Route to a team**. The step now has two questions: `is_spam` (Yes / No) and `team` (Choice).

4. Add **If / else** after AI Classify and configure its branches using the table.

5. On the **If** branch, add **Mark as spam**. On each other branch, add **Send email notification** and set its recipient.

6. Select **Publish**, then submit one sample enquiry for each route and one obvious spam message.

| Branch | Condition | Action |
| - | - | - |
| If | `is_spam` equals `true` | **Mark as spam** |
| Else if | `team` equals `sales` | Email `sales@example.com` |
| Else if | `team` equals `support` | Email `support@example.com` |
| Else | No earlier branch matched | Email `hello@example.com` |

```txt
                              ╭───────────────────────╮
                              │  Submission received  │
                              ╰───────────┬───────────╯
                              ╭───────────▼───────────╮
                              │      AI Classify      │
                              │    is_spam · team     │
                              ╰───────────┬───────────╯
                              ╭───────────▼───────────╮
                              │       If / else       │
                              ╰───────────┬───────────╯
         ╭─────────────────────┬──────────┴──────────┬─────────────────────╮
         │ If: is_spam         │ Else if: team       │ Else if: team       │ Else
         │ equals true         │ equals sales        │ equals support      │
╭────────▼────────╮   ╭────────▼────────╮   ╭────────▼────────╮   ╭────────▼────────╮
│   Mark as spam  │   │      Email      │   │      Email      │   │      Email      │
╰─────────────────╯   │      sales@     │   │     support@    │   │      hello@     │
                      ╰─────────────────╯   ╰─────────────────╯   ╰─────────────────╯
```

The spam check comes first, so a spam message never reaches a team inbox. Billing questions fall through to **Else**, because the `billing` option has no branch of its own. Add an **Else if** for it when your billing team wants its own inbox.

Forminit's own spam protection runs before the workflow, so this step sees only submissions that passed it. AI Classify adds a second layer for messages that look legitimate to filters but not to a reader: SEO offers, link requests, vendor pitches.

### Qualify leads and alert sales

**Requires Pro or higher** for the Google Sheets and Slack steps.

Use a lead form with company, job title and message fields. This example records every lead in Google Sheets, then alerts sales in Slack only for leads who are ready to buy or ask for a demo.

1. Add **Send to Google Sheets** after **Submission received** and map the lead's name, email and company to columns.

2. Add **AI Classify** after the Google Sheets step. In **What should AI read?**, insert the company, job title and message fields, then add a line describing what you sell.

3. Select **Add question**, then **Lead qualification**. The set adds `buying_intent` (Score, levels 0 to 2), `company_size` (Choice) and `wants_demo` (Yes / No).

4. Add **If / else** after AI Classify. Set the **If** branch to **Match any** with two conditions: `buying_intent` **greater than or equal** `2`, and `wants_demo` **equals** `true`.

5. On the **If** branch, add **Send Slack notification** for your sales channel and include `{{ai.classify.company_size}}` in the message.

6. On the **Else** branch, add **Send email notification** for your general enquiries inbox.

7. Select **Publish** and submit sample leads: one that mentions a budget and timeline, one asking for a demo, and one that is only browsing.

```txt
                ╭───────────────────────╮
                │  Submission received  │
                ╰───────────┬───────────╯
                ╭───────────▼───────────╮
                │ Send to Google Sheets │
                ╰───────────┬───────────╯
                ╭───────────▼───────────╮
                │      AI Classify      │
                │     buying_intent     │
                │       wants_demo      │
                ╰───────────┬───────────╯
                ╭───────────▼───────────╮
                │       If / else       │
                │  buying_intent >= 2   │
                │  or wants_demo true   │
                ╰───────────┬───────────╯
              ╭─────────────┴─────────────╮
              │ If                        │ Else
  ╭───────────▼───────────╮   ╭───────────▼───────────╮
  │      Notify sales     │   │     Email general     │
  │        in Slack       │   │       enquiries       │
  ╰───────────────────────╯   ╰───────────────────────╯
```

A message that mentions a budget and a start date scores `2` and reaches Slack. A message asking about pricing scores around `1` and goes to the general inbox. If AI is less sure than the **Minimum confidence**, `buying_intent` is empty and the lead follows **Else** as well, so nobody is lost.

Google Sheets runs before AI Classify, so every lead is recorded even when classification fails. To record the answers too, place the Google Sheets step after AI Classify and map `buying_intent` and `company_size` to columns.

---

## Where results appear

| Where | What you see |
| - | - |
| Submission detail | One block per question with its answer, above the sender details. While AI is working it shows **Classifying...**. |
| Submission **Workflow** section | The AI Classify step with its status and answers. When it failed, a **Retry** button. |
| **Settings** > **Logs** | Entries labelled **AI Classify**: completed, no input, skipped, retry scheduled or held. |
| Form **Analytics** | For each AI Classify question, how many of the form's submissions got each answer. Pro and higher. |

The submissions table does not show AI columns. Open a submission to see its answers, or open the form's **Analytics** page for totals across all of its submissions, such as how many were spam or how many leads asked for a demo.

A submission's AI block shows one of these states:

| State | Meaning |
| - | - |
| **Classifying...** | AI is working on this submission. Later steps wait for it. |
| **Needs review** | A Score answer was below the **Minimum confidence**. That answer is empty. |
| **No input** | Every field in **What should AI read?** was empty for this submission. AI was not asked. |
| **Failed** | AI could not be reached or returned an error. The workflow stopped at this step. See [failures and retries](#failures-and-retries). |
| **Plan required** | The step was skipped because the form's plan does not include it. See [Plans](#plans). |

---

## How it runs

AI Classify runs when a submission arrives, before any step placed after it. The workflow waits for the answers, so later steps always see them.

### Test is a dry run

The workflow's **Test** button never calls AI. In a test, the AI Classify step shows **AI not run**, and branches that depend on AI answers are not predicted. AI runs only for real submissions. To check your questions, publish the workflow and submit a few sample messages.

### Failures and retries

If AI fails, for example because the AI service is busy or unreachable, the submission is kept and the workflow stops at the AI Classify step. Steps before it have already run. Steps after it have not.

```txt
╭─────────────────────────╮
│  Send to Google Sheets  │  ✓ ran before AI Classify, not run again
╰────────────┬────────────╯
┏━━━━━━━━━━━━▼━━━━━━━━━━━━┓
┃       AI Classify       ┃  ✗ failed, the workflow stops here
┗━━━━━━━━━━━━┯━━━━━━━━━━━━┛  ↻ Retry continues from this step
             │
╭────────────▼────────────╮
│ Send email notification │  runs once AI Classify succeeds
╰─────────────────────────╯
```

Forminit retries automatically when the failure is temporary. You can also select **Retry** in the submission's **Workflow** section. The workflow then continues from the AI step: it classifies the submission and runs the steps after it. Steps that already ran, such as an email that was already sent, do not run again.

### Old submissions

Submissions received before you added the step are never classified automatically, including after an upgrade. Only an explicit **Retry** reruns a failed step, and only for that submission.

---

## What is sent to AI

For each submission, Forminit sends the text produced by **What should AI read?**, together with your questions, options and descriptions, to an AI service, and stores the answers with the submission.

Nothing outside that text is sent. A form field you did not add is not included, and neither are Question IDs or submissions caught by Forminit's spam protection, which never reach the workflow. To keep a field such as the sender's email out of AI's view, leave it out of **What should AI read?**.

AI Classify is off until you add it. Forms whose workflow has no AI Classify step send nothing to the AI service, and a new step runs only for submissions received after you publish it. The AI service processes the text in the United States and does not use it to train models. It is listed with the other [subprocessors](https://forminit.com/gdpr/#subprocessors) on the GDPR page.

---

## Limits

| Limit | Value |
| - | - |
| AI Classify steps per workflow | 5 on Pro and higher, 1 on Free |
| AI Classify steps on one path | 3 |
| Questions per step | 1 to 10 |
| Options per Choice question | 2 to 10 |
| Levels per Score question | 2 to 10 |
| **What should AI read?** text | 8,000 characters |
| All questions, options and descriptions in one step | 8,000 characters |

A workflow that exceeds a limit cannot be published. Forminit points to the step that needs fixing.

---

## Plans

AI Classify is available on **all plans**, including Free. The Free plan allows one AI Classify step per workflow. Adding another asks you to upgrade to Pro, which allows up to 5. For forms in a workspace, the workspace owner's plan applies. See [pricing](https://forminit.com/pricing/) for plan details.

Answer counts on the form's **Analytics** page need Pro or higher, like the rest of Analytics. On Free, the answers still appear on each submission and work in later steps.

After a downgrade, AI Classify steps that your new plan does not include stay in the workflow but are skipped. Their answers are empty, the rest of the workflow still runs, and the submission shows **Plan required**. Conditions on their answers then behave as if the answer **does not exist**. The steps run again for new submissions after an upgrade.

---

## Troubleshooting

| Problem | What to check |
| - | - |
| The answers do not appear in an **If / else** field menu | The AI Classify step must be before the **If / else** on the same path. Move it, then publish. |
| A submission shows **No input** | Every field in **What should AI read?** was empty for it. Check that the inserted fields match the names your form sends, and add the fields your questions are about. |
| A Score answer is empty or shows **Needs review** | AI was less sure than the **Minimum confidence**. Route empty answers with **does not exist**, make the level descriptions more distinct, or lower the slider. |
| A Yes / No answer is `false` when it should be `true` | Phrase the question so Yes is the case you want to catch, describe what Yes means, and lower the **Yes threshold** if missed cases cost more than false alarms. |
| A Choice answer is often wrong | Add a one-line description to each option, include an `other` option, and remove fields the question is not about from **What should AI read?**. |
| **Publish** reports the text is too long | **What should AI read?**, and all questions with their options and descriptions, must each stay within 8,000 characters. Remove fields or shorten fixed facts. |
| **Publish** lists broken references | A Question ID was renamed. Point each listed step at the new name. |
| You cannot add another AI Classify step | The Free plan allows one AI Classify step per workflow. Pro and higher allow up to 5, with up to 3 on one path. |
| **Test** shows **AI not run** | Expected. Test is a dry run. Publish and submit a sample to see real answers. |
| The step shows **Failed** | Open the submission's **Workflow** section for the error. Forminit retries temporary failures automatically, or select **Retry**. |
| The step shows **Plan required** | The form's plan, or the workspace owner's plan, does not include this step, for example a second AI Classify step on the Free plan. Upgrade to run it for new submissions. |
| Old submissions have no answers | Submissions received before the step was published are never classified automatically. |

---

## FAQ

### Which plans include AI Classify?

AI Classify is available on all plans, including Free. The Free plan allows one AI Classify step per workflow, and Pro and higher allow up to 5. For forms in a workspace, the workspace owner's plan applies.

### Does AI Classify run on submissions I already received?

No. The step runs only for submissions that arrive after you publish it. Old submissions are never classified automatically, and only an explicit **Retry** reruns a failed step.

### What does AI see?

Only the text produced by **What should AI read?**, together with your questions, options and descriptions. Question IDs are never sent, and neither is any form field you did not add to that text.

### Can I test AI Classify without a real submission?

No. The workflow's **Test** button is a dry run that never calls AI. The AI step shows **AI not run**, and branches that depend on its answers are not predicted. Publish the workflow and submit a sample to see real answers.

### Why is a Score answer empty?

AI was less sure than the question's **Minimum confidence**, so the answer was left empty and the submission shows **Needs review**. Route these submissions with a **does not exist** condition, or lower the slider if you prefer an answer over a review.

### Can a Score answer be a decimal?

Yes. A Score is a number from 0 to the top level and can be fractional, such as `1.4`, when a submission sits between two levels. Compare it with **greater than or equal**, **less than** or **between** rather than **equals**.

### Where can I use the answers?

In every step placed after AI Classify on the same path: **If / else** conditions, email notification and autoresponder subjects and bodies, webhook payloads, Slack and Discord messages, Google Sheets columns, and the **What should AI read?** text of a later AI Classify step. An answer cannot be used as an email address.

### Can I see how many submissions got each answer?

Yes, on Pro and higher. The form's **Analytics** page counts the answers to each AI Classify question across all submissions in the form, for example how many submissions were spam or how many leads asked for a demo.

### What happens if AI fails?

The submission is kept and the workflow stops at the AI Classify step. Forminit retries automatically when the failure is temporary, and you can select **Retry** in the submission's **Workflow** section. The workflow then continues from the AI step, and steps that already ran do not run again.

---

## Ready-made question sets reference

Every set as it ships in the product, with its questions, answer types, Question IDs, options and levels. Add a set from the **Add question** menu, then edit, reorder or remove its questions like any other.

- The suggested **What should AI read?** text uses placeholders such as `{{message}}`. When you add a set, the placeholder suggests matching fields from your form. Select **Add field** to insert your own fields in their place, and keep any fixed facts the set relies on, such as what you sell.
- Every **Score** question uses a **Minimum confidence** of 80% and every **Yes / No** question uses a **Yes threshold** of 80%, except where noted.
- A set can be added only when the step has enough free question slots. Each step holds up to 10 questions.

| Set | Use it to | Question IDs |
| - | - | - |
| [Spam filter](#spam-filter) | Catch ads, scams and bot messages before anyone reads them | `is_spam` |
| [Lead qualification](#lead-qualification) | Rate buying intent and company size, and spot demo requests | `buying_intent`, `company_size`, `wants_demo` |
| [Quote requests](#quote-requests) | Size the project and check for a budget and timeline | `project_size`, `budget_stated`, `timeline` |
| [Support triage](#support-triage) | Sort requests by topic and urgency, and flag upset customers | `topic`, `urgency`, `is_frustrated` |
| [Route to a team](#route-to-a-team) | Send each submission to sales, support or billing | `team` |
| [Refunds and cancellations](#refunds-and-cancellations) | Spot refund and cancellation requests so you can act fast | `refund_requested`, `cancellation_requested` |
| [Job applications](#job-applications) | Rate role fit and seniority, and check for work samples | `role_fit`, `seniority`, `has_work_samples` |
| [Customer feedback](#customer-feedback) | Measure sentiment, sort feedback and find who wants a reply | `sentiment`, `feedback_type`, `wants_reply` |
| [Churn risk](#churn-risk) | Find unhappy customers who may leave, and why | `churn_risk`, `mentions_competitor`, `main_reason` |
| [Bug reports](#bug-reports) | Rate severity and check that the report can be reproduced | `severity`, `has_steps` |
| [Business inquiries](#business-inquiries) | Separate sales leads from partners, press, recruiters and vendors | `inquiry_type` |
| [Agency enquiries](#agency-enquiries) | Tell new projects apart from client support, partners and sales pitches | `enquiry_type` |
| [Onboarding answers](#onboarding-answers) | Check that a client's written answer really answers your question | `answer_complete` |
| [Safety check](#safety-check) | Flag abusive messages and sensitive personal data | `is_abusive`, `has_sensitive_data` |
| [Orders and returns](#orders-and-returns) | Sort shop messages into order status, returns, damage and cancellations | `order_request`, `has_order_number` |
| [Real estate inquiries](#real-estate-inquiries) | Tell buyers, sellers and renters apart and see how soon they move | `client_type`, `move_timeline`, `financing_ready` |
| [Legal intake](#legal-intake) | Find the practice area and flag matters with a close deadline | `practice_area`, `is_time_sensitive` |
| [Insurance claims](#insurance-claims) | Sort claims by type and catch reports with missing details | `claim_type`, `details_missing` |
| [Nonprofit inquiries](#nonprofit-inquiries) | Route donors, volunteers, event guests and people asking for help | `contact_reason` |
| [Renovation and installation](#renovation-and-installation) | Sort jobs by service and intent, and spot descriptions that need more detail | `service_category`, `request_intent`, `description_ready` |
| [Print, signage and manufacturing](#print-signage-and-manufacturing) | Find the product, spot who needs design help and check the job can be priced | `product`, `needs_design_help`, `spec_clear` |
| [Property maintenance](#property-maintenance) | Send repairs to the right trade and flag possible emergencies for review | `trade`, `possibly_urgent` |
| [Catering and events](#catering-and-events) | Route by event type and tell catering-only requests from full service | `event_type`, `service_level` |
| [Language routing](#language-routing) | Detect the language so the right person replies | `language` |
| [Appointment requests](#appointment-requests) | Tell new bookings apart from changes, cancellations and questions | `request_type` |

The first 14 sets are listed under **General** in the menu, the rest under **Industry**.

### General

#### Spam filter

Catch ads, scams and bot messages before anyone reads them.

Suggested **What should AI read?** text:

```text
Name: {{name}}
Email: {{email}}
Message: {{message}}
```

1. **Is this submission spam?** Yes / No, `is_spam`
   - Yes means: Unsolicited advertising, scams, SEO or link offers, or automated junk
   - No means: A real person writing about something relevant to us

Forminit's own spam protection runs before the workflow, so this set sees only submissions that passed it. Use it for messages that pass filters but a reader would still bin.

#### Lead qualification

Rate buying intent and company size, and spot demo requests.

Suggested **What should AI read?** text:

```text
Company: {{company}}
Job title: {{job_title}}
Message: {{message}}
What we sell: Project management software for teams
```

Replace the last line with what your company sells. AI uses it to judge whether the sender is a fit.

1. **How ready is the sender to buy?** Score, `buying_intent`
   - 0: Just browsing or researching. No plans or needs mentioned
   - 1: Interested. Asks about features, pricing or fit
   - 2: Ready to buy. Mentions a budget, a timeline or a purchase decision
2. **How big is the sender's company, based on what they wrote?** Choice, `company_size`
   - `individual`: A freelancer or a single person
   - `small`: A small team of up to about 50 people
   - `mid_market`: A company of about 50 to 1,000 people
   - `enterprise`: A large organization with more than 1,000 people
   - `not_stated`: The submission doesn't say
3. **Is the sender asking for a demo or a sales call?** Yes / No, `wants_demo`

#### Quote requests

Size the project and check for a budget and timeline.

Suggested **What should AI read?** text:

```text
Service needed: {{service}}
Project details: {{details}}
Start date: {{start_date}}
```

1. **How big is the project the sender describes?** Score, `project_size`
   - 0: A small one-off task or a quick fix
   - 1: A defined project of a few days or weeks
   - 2: A large or ongoing project with several parts
2. **Does the sender mention a budget or price range?** Yes / No, `budget_stated`
3. **When does the sender want the work to start?** Choice, `timeline`
   - `asap`: Right away or within two weeks
   - `this_quarter`: Within the next three months
   - `later`: More than three months away
   - `not_stated`: The submission doesn't say

#### Support triage

Sort requests by topic and urgency, and flag upset customers.

Suggested **What should AI read?** text:

```text
Subject: {{subject}}
Message: {{message}}
Plan: {{plan}}
```

1. **What is the request mainly about?** Choice, `topic`
   - `account`: Signing in, passwords or account settings
   - `billing`: Invoices, payments, plans or refunds
   - `bug`: Something isn't working as expected
   - `how_to`: A question about how to use the product
   - `feature_request`: A request for something the product doesn't do yet
   - `other`: Anything that fits none of the options above
2. **How urgent is this request?** Score, `urgency`
   - 0: It can wait. Nothing is blocked for the sender
   - 1: It should be handled soon. The sender is slowed down
   - 2: It needs action today. The sender is blocked or losing money
3. **Is the sender upset or frustrated?** Yes / No, `is_frustrated`

#### Route to a team

Send each submission to sales, support or billing.

Suggested **What should AI read?** text:

```text
Subject: {{subject}}
Message: {{message}}
```

1. **Which team should handle this request?** Choice, `team`
   - `sales`: Pricing, demos, buying or upgrading
   - `support`: Help using the product or a problem with it
   - `billing`: Invoices, payments and refunds
   - `other`: Anything that fits none of the options above

#### Refunds and cancellations

Spot refund and cancellation requests so you can act fast.

Suggested **What should AI read?** text:

```text
Order number: {{order_number}}
Message: {{message}}
Refund policy: Refunds are possible within 30 days of purchase.
```

Replace the policy line with your own. It helps AI tell a refund request from a question about the policy.

1. **Is the sender asking for money back?** Yes / No, `refund_requested`
   - Yes means: The sender asks for a refund, a chargeback or money back
   - No means: Any other request, including questions about the refund policy
2. **Does the sender want to cancel a subscription, order or booking?** Yes / No, `cancellation_requested`

#### Job applications

Rate role fit and seniority, and check for work samples.

Suggested **What should AI read?** text:

```text
Role applied for: {{position}}
Experience: {{experience}}
Portfolio: {{portfolio_url}}
Role requirements: 3+ years of React and TypeScript
```

Replace the requirements line with the role's own. The role fit score compares the application against it.

1. **How well does the applicant's experience match the role they applied for?** Score, `role_fit`
   - 0: No relevant experience is mentioned
   - 1: Some related experience, but key skills are missing
   - 2: Relevant experience that covers most key skills
   - 3: Strong, directly relevant experience across all key skills
2. **What seniority level does the applicant describe?** Choice, `seniority`
   - `entry`: Student, intern or less than 2 years of experience
   - `mid`: About 2 to 5 years of experience
   - `senior`: More than 5 years, or a lead or management role
   - `not_stated`: The application doesn't say
3. **Does the application include a link to a portfolio, GitHub or work samples?** Yes / No, `has_work_samples`

#### Customer feedback

Measure sentiment, sort feedback and find who wants a reply.

Suggested **What should AI read?** text:

```text
Rating: {{rating}}
Feedback: {{feedback}}
```

1. **How does the sender feel about their experience?** Score, `sentiment`
   - 0: Very unhappy. Angry or complaining
   - 1: Disappointed or somewhat unhappy
   - 2: Neutral or mixed
   - 3: Satisfied
   - 4: Delighted. Full of praise
2. **What kind of feedback is this?** Choice, `feedback_type`
   - `praise`: Positive comments about the product or service
   - `complaint`: A problem or something the sender is unhappy with
   - `suggestion`: An idea for an improvement
   - `question`: The sender mainly wants an answer
   - `other`: Anything that fits none of the options above
3. **Does the sender ask for a reply or a follow-up?** Yes / No, `wants_reply`

#### Churn risk

Find unhappy customers who may leave, and why.

Suggested **What should AI read?** text:

```text
Plan: {{plan}}
How likely to recommend us (0 to 10): {{score}}
Comment: {{comment}}
```

1. **How likely is the sender to stop using our product or service?** Score, `churn_risk`
   - 0: Happy and plans to keep using it
   - 1: Has complaints but no sign of leaving
   - 2: Is considering leaving or downgrading
   - 3: Says they are leaving or have already cancelled
2. **Does the sender mention switching to or comparing us with another company?** Yes / No, `mentions_competitor`
3. **What is the main reason for the sender's feedback?** Choice, `main_reason`
   - `price`: Cost, pricing or value for money
   - `quality`: Bugs, reliability or product quality
   - `support`: The help or service they received
   - `ease_of_use`: The product is hard to use or learn
   - `missing_feature`: Something they need is missing
   - `other`: Anything that fits none of the options above

#### Bug reports

Rate severity and check that the report can be reproduced.

Suggested **What should AI read?** text:

```text
Page or feature: {{page}}
What happened: {{description}}
Steps to reproduce: {{steps}}
```

1. **How severe is the reported issue?** Score, `severity`
   - 0: Cosmetic. No impact on how things work
   - 1: A feature is broken or slow, but there is a workaround
   - 2: Blocking. There is no workaround
2. **Does the report explain the steps to reproduce the problem?** Yes / No, `has_steps`

#### Business inquiries

Separate sales leads from partners, press, recruiters and vendors.

Suggested **What should AI read?** text:

```text
Company: {{company}}
Message: {{message}}
```

1. **What kind of inquiry is this?** Choice, `inquiry_type`
   - `sales`: A potential customer interested in buying
   - `partnership`: A proposal to partner, integrate or resell
   - `press`: A journalist, podcast or media request
   - `job_seeker`: Someone asking about jobs or sending a CV
   - `vendor_pitch`: Someone trying to sell us a product or service
   - `other`: Anything that fits none of the options above

#### Agency enquiries

Tell new projects apart from client support, partners and sales pitches.

Suggested **What should AI read?** text:

```text
Company: {{company}}
Message: {{message}}
About us: We design, build and maintain websites for small businesses
```

Replace the last line with a sentence about your agency.

1. **What is the sender getting in touch about?** Choice, `enquiry_type`
   - `new_project`: A potential client who wants a new website, app or design work
   - `client_support`: An existing client asking for changes, fixes or help with their site
   - `partnership`: A proposal to partner, refer clients or white-label work
   - `sales_pitch`: Someone trying to sell us a product, service or leads
   - `job_seeker`: Someone asking about jobs or offering to freelance
   - `other`: Anything that fits none of the options above

#### Onboarding answers

Check that a client's written answer really answers your question.

Suggested **What should AI read?** text:

```text
Question asked: What makes your business different from competitors?
Answer: {{answer}}
```

Replace the first line with the question your form asks. Use one AI Classify step per question you want checked, up to the [limits](#limits) above.

1. **How well does the answer respond to the question asked?** Score, `answer_complete`
   - 0: It doesn't answer the question or is off topic
   - 1: It partly answers it. Key points are missing or vague
   - 2: It answers the question clearly and specifically

#### Safety check

Flag abusive messages and sensitive personal data.

Suggested **What should AI read?** text:

```text
Message: {{message}}
```

1. **Is the message abusive toward a person?** Yes / No, `is_abusive`
   - Yes means: Insults, threats or harassment aimed at someone
   - No means: Polite, neutral or critical messages that don't attack anyone
2. **Does the submission include sensitive personal data, like ID numbers, card numbers or health details?** Yes / No, `has_sensitive_data`

### Industry

#### Orders and returns

Sort shop messages into order status, returns, damage and cancellations.

Suggested **What should AI read?** text:

```text
Order number: {{order_number}}
Message: {{message}}
```

1. **What does the customer want help with?** Choice, `order_request`
   - `order_status`: Where an order is, or when it will arrive
   - `return_or_exchange`: Sending an item back, or swapping it for another
   - `damaged_or_wrong`: An item arrived damaged, broken, wrong or missing
   - `cancel_order`: Cancelling an order before it ships
   - `product_question`: A question about a product before buying
   - `other`: Anything that fits none of the options above
2. **Does the message include an order number?** Yes / No, `has_order_number`

#### Real estate inquiries

Tell buyers, sellers and renters apart and see how soon they move.

Suggested **What should AI read?** text:

```text
Message: {{message}}
Preferred area: {{area}}
Budget: {{budget}}
```

1. **What does the sender want to do with a property?** Choice, `client_type`
   - `buy`: Buy a home or property
   - `sell`: Sell a home or property they own
   - `rent`: Rent a place to live or work
   - `landlord`: Rent out a property they own
   - `other`: Anything that fits none of the options above
2. **When does the sender want to buy, sell or move?** Choice, `move_timeline`
   - `within_30_days`: Within the next month
   - `within_3_months`: In one to three months
   - `later`: More than three months away
   - `not_stated`: The submission doesn't say
3. **Does the sender say they have mortgage pre-approval or cash ready?** Yes / No, `financing_ready`

#### Legal intake

Find the practice area and flag matters with a close deadline.

Suggested **What should AI read?** text:

```text
Your matter: {{message}}
Our practice areas: Family, employment and immigration law
```

Replace the last line with the areas your firm handles.

1. **Which area of law is the sender's matter about?** Choice, `practice_area`
   - `family`: Divorce, custody, child support or adoption
   - `criminal`: Arrests, charges or criminal defense
   - `personal_injury`: Accidents, injuries or medical negligence
   - `employment`: Work disputes, dismissal or discrimination
   - `immigration`: Visas, residency or citizenship
   - `business`: Contracts, companies or commercial disputes
   - `property`: Real estate, leases or landlord disputes
   - `other`: Anything that fits none of the options above
2. **Does the sender mention a deadline, court date or arrest in the coming days?** Yes / No, `is_time_sensitive`

#### Insurance claims

Sort claims by type and catch reports with missing details.

Suggested **What should AI read?** text:

```text
Policy number: {{policy_number}}
Date of incident: {{incident_date}}
What happened: {{description}}
```

1. **What kind of insurance claim is this?** Choice, `claim_type`
   - `vehicle`: Car, motorcycle or other vehicle damage
   - `property`: Damage to a home, building or belongings
   - `health`: Medical treatment or health costs
   - `travel`: Trip cancellation, delays or lost luggage
   - `liability`: Damage or injury the sender caused to someone else
   - `other`: Anything that fits none of the options above
2. **Is the report missing when, where or how the incident happened?** Yes / No, `details_missing`

#### Nonprofit inquiries

Route donors, volunteers, event guests and people asking for help.

Suggested **What should AI read?** text:

```text
Name: {{name}}
Message: {{message}}
```

1. **What is the sender reaching out about?** Choice, `contact_reason`
   - `donation`: Giving money, goods or a recurring donation
   - `volunteering`: Offering their time or skills
   - `event`: Attending, hosting or asking about an event
   - `request_help`: Asking for support from our programs
   - `partnership`: A company or group proposing to work together
   - `other`: Anything that fits none of the options above

#### Renovation and installation

Sort jobs by service and intent, and spot descriptions that need more detail.

Suggested **What should AI read?** text:

```text
Project details: {{message}}
Postcode: {{postcode}}
Our services: Kitchens, bathrooms, flooring, windows, doors and roofing
```

Replace the last line with the services you offer.

1. **Which kind of work is the customer asking about?** Choice, `service_category`
   - `kitchen`: Kitchen cabinets, worktops or a full kitchen
   - `bathroom`: Bathrooms, showers, tiling or toilets
   - `flooring`: Floors, carpets or laminate
   - `windows_doors`: Windows, doors or glazing
   - `roofing`: Roofs, gutters or chimneys
   - `extension`: Extensions, loft conversions or structural changes
   - `other`: Anything that fits none of the options above
2. **What does the customer want from us next?** Choice, `request_intent`
   - `quote`: A price or an estimate
   - `site_visit`: Someone to visit and look at the job
   - `question`: An answer to a general question
   - `other`: Anything that fits none of the options above
3. **Does the description say what work is needed and where in the property?** Yes / No, `description_ready`

#### Print, signage and manufacturing

Find the product, spot who needs design help and check the job can be priced.

Suggested **What should AI read?** text:

```text
What you need: {{message}}
Quantity: {{quantity}}
Our products: Business cards, flyers, signs, banners and packaging
```

Replace the last line with the products you make.

1. **Which product or service is the customer asking for?** Choice, `product`
   - `business_cards`: Business cards or stationery
   - `flyers_brochures`: Flyers, leaflets, brochures or posters
   - `signs_banners`: Signs, banners, displays or window graphics
   - `vehicle_graphics`: Vehicle wraps, decals or lettering
   - `packaging`: Boxes, labels or custom packaging
   - `custom_manufacturing`: A custom part, product or prototype
   - `other`: Anything that fits none of the options above
2. **Does the customer need help creating the design or artwork?** Yes / No, `needs_design_help`
   - Yes means: They have no artwork yet or ask us to design it
   - No means: They have print-ready artwork or only need printing
3. **Does the request give enough detail to price the job, such as size, quantity or material?** Yes / No, `spec_clear`

#### Property maintenance

Send repairs to the right trade and flag possible emergencies for review.

Suggested **What should AI read?** text:

```text
Property address: {{address}}
Problem: {{message}}
```

1. **Which trade is needed to fix the reported problem?** Choice, `trade`
   - `plumbing`: Leaks, blocked drains, taps, toilets or water pressure
   - `electrical`: Power, sockets, lights, switches or the fuse box
   - `heating`: Boilers, radiators, hot water or air conditioning
   - `structural`: Roofs, walls, ceilings, windows, doors or damp
   - `appliances`: Ovens, fridges, washing machines or other appliances
   - `pests`: Insects, rodents or other pests
   - `other`: Anything that fits none of the options above
2. **Does the report describe a possible safety risk or damage that is happening now?** Yes / No, `possibly_urgent`
   - Yes means: For example a leak, a gas smell, sparks, exposed wiring, no heating in cold weather or a door that won't lock
   - No means: A problem that can wait for a normal appointment
   - **Yes threshold: 50%.** This question is tuned to catch possible emergencies, so it accepts more false alarms than the other sets. Route a `true` answer to someone who can check quickly, not to an automatic callout.

#### Catering and events

Route by event type and tell catering-only requests from full service.

Suggested **What should AI read?** text:

```text
Event date: {{event_date}}
Number of guests: {{guests}}
About your event: {{message}}
```

1. **What kind of event is the customer planning?** Choice, `event_type`
   - `wedding`: A wedding, engagement or rehearsal dinner
   - `corporate`: A business meeting, conference, launch or office party
   - `private_party`: A birthday, anniversary or other private celebration
   - `other`: Anything that fits none of the options above
2. **How much does the customer want us to provide?** Choice, `service_level`
   - `catering_only`: Food and drinks only
   - `full_service`: Food plus staff, setup, decoration or event planning
   - `not_stated`: The submission doesn't say

#### Language routing

Detect the language so the right person replies.

Suggested **What should AI read?** text:

```text
Message: {{message}}
```

1. **Which language is the submission written in?** Choice, `language`
   - `english`
   - `spanish`
   - `german`
   - `french`
   - `portuguese`
   - `turkish`
   - `other`: Any other language

Add or rename options to match the languages your team answers in. Pair it with **Submitter Country** in **What should AI read?** when the message is too short to tell.

#### Appointment requests

Tell new bookings apart from changes, cancellations and questions.

Suggested **What should AI read?** text:

```text
Preferred date: {{date}}
Message: {{message}}
Opening hours: Monday to Friday, 9:00 to 17:00
```

Replace the last line with your own opening hours.

1. **What does the sender want to do about an appointment?** Choice, `request_type`
   - `new_booking`: Book a new appointment
   - `reschedule`: Move an existing appointment to another time
   - `cancel`: Cancel an existing appointment
   - `question`: Ask a question without changing a booking
   - `other`: Anything that fits none of the options above

---

## Related

- [Workflow](https://forminit.com/docs/workflow/) - Build the workflow, route with If / else, publish and retry
- [Email Notifications](https://forminit.com/docs/email-notifications/) - Include answers in notification subjects and bodies
- [Webhooks](https://forminit.com/docs/webhooks/) - Receive answers in the webhook payload
- [Google Sheets](https://forminit.com/docs/google-sheets/) - Record answers as spreadsheet columns