---
title: Content Screening
description: How Preservated automatically flags sensitive imagery for review, and how curators approve or clear those flags.
section: guides
order: 12
updated: 2026-07-17
verified: 2026-07-17
related:
  [
    guides/content-warnings,
    guides/ai-analysis,
    guides/importing-photos,
    guides/working-with-artifacts,
  ]
features: [content-screening]
---

# Content Screening

You'll learn how Preservated screens images for sensitive content as they're ingested, and how to review what it flags from **Admin → Operations → Screening**.

Archives hold difficult material — historical nudity in art and ethnographic photography, medical imagery, photographs of war and disaster. Preservated's guiding principle is the museum one: **preserve, don't censor.** Screening exists to *label and contextualize* sensitive material so you can give visitors appropriate warnings — not to delete it.

## How screening works

Every image is screened automatically when it's imported, before AI metadata analysis runs. Screening has two stages:

1. **A fast moderation sweep** runs on every image. Most images are clearly unproblematic and are marked `general` with no further work.
2. **A museum-context review** runs only on the small minority the first stage flags. This step is told it's looking at historical archive material, so it distinguishes fine-art or ethnographic nudity and historical violence (core archival content) from genuinely explicit material, and assigns a content rating with an explanation.

Each image receives a **content rating**:

- **general** — nothing notable.
- **caution** — sensitivity or non-graphic content; intended for an inline advisory.
- **restricted** — nudity, explicit, or graphic material; intended to be shown with care.

:::info
Screening **never** assigns the highest "blocked" tier. Automated tools cannot reliably judge whether material is illegal, so anything that might be is routed to a person instead (see below). Blocked is only ever set by a human decision.
:::

Ratings are recorded on the asset and rolled up to its artifact, with the source marked as AI. A curator decision always overrides the AI — once you approve or clear a rating, re-screening never changes it back.

## Review what's flagged

Open **Admin → Operations → Screening**. Two lists appear.

### Flagged for review

AI-flagged images (caution or restricted) awaiting a curator decision. Each shows a thumbnail, the rating, the suggested warning labels, and the reviewer's reasoning. For each item you can:

- **Approve & keep** — confirm the rating and its labels. This locks the decision as curator-made.
- **Clear (false positive)** — reset the image to `general`. Use this when the flag is wrong.

Most flagged museum material is legitimate archival content (a classical nude, a battlefield photograph). The expected workflow is to approve and keep, so the right warning travels with the item.

You can also inspect any single item without leaving its page: in admin mode the artifact page shows an insights control in the action row (between **Edit** and **Stats**) whose hover card carries both the editorial "visual interest" scores and the screening result — rating, warnings, and the reviewer's reasoning.

### Flagged for human review

A separate, deliberately sparse list for material the review step judged it could not responsibly rate — content that may be illegal or has no legitimate archival presentation. These entries show **only a pointer and a short reason — never the image.**

:::danger
For anything in this list: **report, don't investigate.** Do not download, copy, forward, or re-open the material to "confirm" it. Follow your institution's illegal-content runbook and report through the proper channel. Over-handling such material can itself be unlawful.
:::

## Screen existing images

Screening runs automatically on new imports. To screen images that were imported before screening existed, use **Screen unscreened images**. It queues screening for every unscreened image in the institution, skipping anything a curator or community has already decided. Progress runs through the background job queue.

## Costs

The first-stage sweep is free. The museum-context review calls a vision model, but only for the flagged minority — so for a typical collection the per-image cost is near zero.

## What visitors see

Ratings drive the visitor-facing treatments — inline advisory banners, blur-until-click on restricted media, the sensitive-content statement, child-safe browsing, and the visitor report queue. See [Content Advisories & Safe Browsing](/docs/en-US/guides/content-warnings) for what visitors experience and the settings that control it.