> ## Documentation Index
> Fetch the complete documentation index at: https://simpai.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Understanding the hallucination risk score

> S.I.M.P. scores every prompt from 1 to 10 for hallucination risk. Learn what Low, Moderate, High, and Critical ratings mean and how to lower yours.

Every prompt you submit to S.I.M.P. receives a hallucination risk score from 1 to 10. Hallucination is when an AI confidently makes up information that isn't true — it's not lying, it genuinely doesn't know what it doesn't know. A higher score means your prompt is more likely to trigger this behavior because it leaves too much for the AI to infer, invent, or assume. The score tells you how much room your prompt gives an AI to go off the rails.

## The four risk levels

| Level | Score | What it means |
| - | - | - |
| **Low** | 1–3 | Your prompt is mostly clear. There may be minor improvements, but the AI has enough to work with. Hallucination risk is minimal. |
| **Moderate** | 4–6 | The prompt has gaps — missing context, vague scope, or implicit assumptions the AI will need to fill in. Worth reviewing before use. |
| **High** | 7–8 | Multiple ambiguities are present and there are no constraints to guide the AI. The AI is likely to invent details to compensate. |
| **Critical** | 9–10 | The prompt is extremely vague or is missing fundamental context. The AI has almost nothing reliable to work from. |

## How the score is displayed

S.I.M.P. shows your risk score as a color-coded progress bar beneath the results header. The color shifts from green (Low) through yellow (Moderate) and orange (High) to red (Critical), giving you an at-a-glance read of where your prompt stands before you dig into the issues list.

## How to lower your score

Most high scores come from a handful of common patterns. To reduce your hallucination risk:

* **Add context.** Tell the AI who you are, what you're building, and what the output is for. The more relevant background you provide, the less the AI needs to guess.
* **Define scope.** Specify what's in bounds and what's out. If there are topics, formats, or approaches you don't want, say so explicitly.
* **Specify the output format.** Should the response be a bullet list? A paragraph? A JSON object? If you don't say, the AI decides — and it may not decide well.
* **Avoid implicit assumptions.** If something is important enough to affect the answer, it's important enough to write down. Don't rely on the AI to "know" your conventions, your codebase, or your preferences.

After S.I.M.P. rewrites your prompt, you can use the "Run Again" button to submit the sanitized version and see how much the score improves.

<Info>
  A lower score doesn't guarantee perfect AI output — it means your prompt gives the AI the best possible foundation. Model quality, knowledge cutoffs, and task complexity all affect the final result.
</Info>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.