> For the complete documentation index, see [llms.txt](https://passion-io.gitbook.io/help/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://passion-io.gitbook.io/help/make-your-ai-answer-well/how-your-ai-decides.md).

# How Your AI Decides What to Say

Understand where your AI's answers come from, so you know which lever to pull when one misses.

> **Your AI** is available on certain plans and app types. If you don't see the AI block in your sidebar, please check our pricing page for more information.

## In This Article

* [The Short Version](#the-short-version)
* [What Your AI Draws On](#what-your-ai-draws-on)
* [How They Weigh Your Material](#how-they-weigh-your-material)
* [What They Won't Do](#what-they-wont-do)
* [When an Answer Misses](#when-an-answer-misses)

## The Short Version

Every reply is shaped by three things you control: the **Resources** you've given your AI, the **role and rules** you set in Behavior, and the **boundaries** you've drawn. Resources are what your AI knows; Behavior is how they use it; boundaries are what they leave alone. Nothing else is guessed from the open web.

## What Your AI Draws On

Your AI answers from the Resources in your library. Each Resource's **Visibility** decides how it's used:

* **Cited** — used in answers and shown to clients as a clickable source.
* **Private** — used silently, never shown.
* **Off** — ignored.

So if your AI doesn't seem to "know" something, the first question is whether a Resource covers it and isn't set to **Off**.

## How They Weigh Your Material

Your **Primary role** in Behavior sets how closely your AI sticks to your Resources versus drawing on broader context:

* **Coaching** keeps the conversation moving while drawing from your material.
* **Reference** stays close to your material and shows where answers came from.
* **Ideation** ranges more widely to brainstorm and make new connections.

On top of the role, your **Custom rules** add fixed always-do / never-do lines that your AI follows even when a Resource doesn't spell out an answer.

## What They Won't Do

Two things hold your AI back, by design: the **Boundaries** you set (topics they decline and steer away from) and the safety defaults that are always on, with no setup from you. Boundaries are how you keep conversations inside your scope; safety defaults are non-negotiable.

## When an Answer Misses

Trace the miss back to the lever that owns it:

* **Wrong or missing facts** → a Resource is absent, set to **Off**, or out of date. Fix it in [Resources](/help/make-your-ai-answer-well/add-resources.md).
* **Right facts, wrong delivery** → tone or approach is set in [Behavior](/help/make-your-ai-answer-well/set-primary-role.md) under Custom rules.
* **Answered something off-limits** → add a **Boundary** in Behavior so they decline it next time.

## Related

* [Add Resources to Your Library](/help/make-your-ai-answer-well/add-resources.md)
* [Set Your AI's Primary Role](/help/make-your-ai-answer-well/set-primary-role.md)
* [Fix Common Issues With Your AI](/help/when-something-goes-wrong/troubleshooting.md)

For more assistance, please contact us.