---
title: NBT Bot data handling
canonical: https://nbt.bot/data-handling/
---

# NBT Bot data handling

NBT keeps enough to show what was tested and explain the result. We avoid larger or more sensitive captures when a smaller piece of evidence will do.

## What we keep

Our content-use level is **reference**. We keep the smallest useful evidence we can: status codes, timings, hashes, parsed values, short excerpts and other diagnostics needed for authorised QA. Screenshots or larger captures are kept only when they help explain what happened.

We do not run a public copy of the sites we test and we do not use crawled customer content to train models.

If a customer turns on a feature that analyses content with a model, we only do that where the site's Content Signals allow it. `ai-input=no` blocks that use. `use=immediate` means the content should not be kept for later reuse.

Authenticated, session-bound or private customer responses are never shared between customers.