> For the complete documentation index, see [llms.txt](https://doc.aissist.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.aissist.io/capabilities/insight-assets.md).

# Insight Assets

Insight Assets are the QA guidelines, policies and procedures that Agent Insight, Evaluation, Session Insight and Analyze commands check conversations against. They are never used to answer customers.

*Last updated: November 6, 2026*

Insight Assets are the knowledge your insight tasks score and analyze conversations against: your QA guidelines, policies, procedures and internal documentation. With them, a score can say whether the agent gave the **right** answer, not only whether it sounded helpful.

They belong to the project, not to a workspace, and they are **never used to answer customers**. Your AI agent keeps answering from its [assets](https://doc.aissist.io/capabilities/assets.md); Insight Assets are only read when a conversation is evaluated or analyzed.

### Why it matters

* **Accuracy, not just tone.** Without a reference, a grader can only judge how an answer reads. With your policies at hand, it can tell that a refund was promised outside the refund window, or that a step of the procedure was skipped.
* **One standard for people and AI.** Human agents and the AI agent are checked against the same documents.
* **Internal documents stay internal.** A QA scorecard or an internal procedure can guide scoring without ever reaching a customer.
* **Set up once.** Every insight task of the project uses them. There is nothing to choose per evaluation set.

### Who uses them

Every insight task of the project uses every Insight Asset that is on:

| Task                                                                                                            | What the assets are used for                                        |
| --------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------- |
| [Agent Insight](https://doc.aissist.io/capabilities/agent-insight.md)                                           | Scoring each agent - your people and the AI - on real conversations |
| [Evaluation](https://doc.aissist.io/capabilities/evaluation.md)                                                 | Scoring simulated conversations in an evaluation run                |
| [Session Insight](https://doc.aissist.io/capabilities/session-insight.md)                                       | Extracting information from conversations as they close             |
| [In-note commands](https://doc.aissist.io/capabilities/in-note-command.md) of type **Analyze** and **Evaluate** | Analyzing or scoring one conversation on demand                     |

Replies to customers, and in-note commands that write for a customer (Summarize, Translate, Rephrase), never use them.

### How it works

The AI does not read every document for every conversation. For each conversation it evaluates:

1. It reads the conversation and the measurements (or, for Session Insight and Analyze, what the commands extract), and works out what it needs to know that the conversation doesn't say. For example: "what is the refund window for annual plans?"
2. It searches your Insight Assets for those questions, the same way a reply searches its assets: up to 8 searches for scoring, up to 4 for Session Insight and Analyze.
3. Each passage found is checked for relevance to the conversation. Weak matches are dropped.
4. At most 20 passages (10 for Session Insight and Analyze) go to the grader, alongside the conversation and the measurements.

The questions are worked out once per conversation and shared by every evaluation set that scores it. If nothing relevant is found, or the project has no Insight Asset that is on, the conversation is scored exactly as it would be without them.

### Add an Insight Asset

1. Go to **Insight → Insight Assets**.
2. Click **Add Asset** and choose:
   * **A website** - a site, or part of one, or only the pages you pick. You can leave out pages you don't want used.
   * **From Google Drive** or **From OneDrive** - pick documents, or follow a folder. This needs a Google Drive or OneDrive [connector](https://doc.aissist.io/capabilities/connectors.md) that is allowed for knowledge.
3. Review what was found and create it.

The asset is read and kept up to date automatically, like any website or drive asset. It is used within about 5 minutes of being added. Insight Assets don't count against your plan's asset limit.

The list shows each asset's **Last synced** time and **Status**: **Synced**, **Waiting for its first sync**, **Sync failed**, or how many documents can't be read or are not used. Open an asset to see its pages or files.

To stop using an asset, turn its **Enabled** switch off. To remove it for good, delete it from **Actions**.

### See what was used

In **Agent Insight → Dashboard → Explore**, open a scored conversation. Each agent's result shows its **References**: how many passages were used, and from which documents. "References: none found" means nothing relevant was found for that conversation.

### What makes a good Insight Asset

* **Write down what a good answer is.** QA guidelines, policies with their exceptions, step-by-step procedures, eligibility rules and limits.
* **One topic per section, with headings**, so each passage makes sense on its own.
* **Keep it current.** An outdated policy marks a correct answer as wrong. Turn an asset off while it is being rewritten.
* **Keep the measurements as they are.** What every conversation must be judged on belongs in the measurement itself. Insight Assets add the facts a measurement needs to be checked.
* **Don't duplicate your workspace assets for no reason.** Add what a grader needs and the customer-facing knowledge may not hold, such as internal rules and QA criteria.

### Related

* [Agent Insight](https://doc.aissist.io/capabilities/agent-insight.md)
* [Evaluation](https://doc.aissist.io/capabilities/evaluation.md)
* [In-note Commands](https://doc.aissist.io/capabilities/in-note-command.md)
* [Assets](https://doc.aissist.io/capabilities/assets.md)
