# Choose a model and response length

URL: https://aihio.ai/en/help/ai-agent/response-settings\
Description: Choose an appropriate model, adjust reply length and write a useful fallback answer when information is missing.

Model selection, response length and the fallback answer solve different problems. Choose a model for the task, allow enough room for the reply and give customers a useful next step when the knowledge base lacks enough information.

## Before you start

Select the correct workspace and agent. You need permission to change the agent's settings. Available models depend on the workspace's subscription. Prepare a few real customer questions and a question the knowledge base cannot answer.

## Change one setting and compare

1. **Choose a model**

   Open the agent's **Overview → Model**. Choose a model available to your
   workspace. For the full settings, open **Settings → AI** and **Show advanced
   model settings**. Check advanced-model availability in this guide's
   **Feature availability** table before switching models.
2. **Choose an appropriate reply length**

   In **Maximum response length**, choose a length preset or **Custom** value.
   The displayed word count is an estimate. Save this section's changes and
   wait for confirmation.
3. **Write a fallback answer**

   Find **Fallback answer** in the same AI view. Explain that sufficient
   information was not found and give a contact method or another real next
   step. Save this section's changes separately.
4. **Test in a new conversation**

   Test the same questions before and after the change. Check correctness,
   length and the fallback answer. Preview test messages consume message
   credits. Change the next setting only after comparing the results.

## Choose length for the task

| Preset  | Example use                                      | Check in the test                                                 |
| ------- | ------------------------------------------------ | ----------------------------------------------------------------- |
| Concise | One clear fact or next step.                     | Is the essential information preserved?                           |
| Short   | A brief support reply.                           | Does the customer get the needed answer without another question? |
| Medium  | Instructions with several steps or a comparison. | Are all steps and qualifications included?                        |
| Long    | A task requiring explanation and detail.         | Is the result still easy to read?                                 |
| Custom  | Your own limit justified by testing.             | Does the answer finish completely?                                |

> **Note — Length does not improve source quality:**
>
> A longer reply or a more advanced model does not replace missing or outdated
> source information. For incorrect content, fix the knowledge base or
> instructions and test again. For a cut-off answer, try a larger length limit.

**Response variability** controls variation in the model's wording. It does not control factual correctness. If the view says this setting has no effect on the selected model, compare results by changing the task instructions or model instead.

## Example fallback answer

This is an editable example. Replace the contact method with your organization's actual details.

```text
I cannot find enough information to answer this. Please send your question to
[your support address] so our customer service team can check it.
```

A fallback answer does not itself create a support request or summon an operator. Use [human support](/en/help/ai-agent/human-support) or [lead collection](/en/help/ai-agent/lead-collection) to capture a request with an Aihio action.

## When information is uncertain

The service assesses the information found before answering. The current service uses balanced handling: insufficient information can produce a fallback answer, while an uncertain reply may include a notice. This does not guarantee correctness or provide a customer-adjustable certainty percentage.

Check the answer against its source and the real situation. [Analytics response quality](/en/help/ai-agent/analytics) helps choose what to fix; even a high score still needs human review.

## Feature availability

Also check availability and any applicable add-on in your workspace. This table gives minimum tiers for these features; it is not a workspace usage limit or a list of every integration.

| Feature         | Minimum tier |
| --------------- | ------------ |
| Advanced models | Pro          |
## Next steps

- [Test your agent before launch](/en/help/ai-agent/testing) — Check answer quality, boundaries and the chat widget before making your agent available to customers.
- [Data sources](/en/help/ai-agent/knowledge) — Add, inspect, and keep your AI agent’s data sources up to date in one view.