AI Prohibited Items - Admin Operations Guide

Published by Give.Gives · Updated April 2, 2026

AI Prohibited Items - Admin Operations Guide

1. Purpose and Business Context

AI Prohibited Items is the admin page used to maintain the prohibited-item list that feeds the platform's AI item-checking flow during donation upload and item analysis.

This list helps the system decide whether an uploaded item should be treated as:

• a Safety Violation that must be rejected immediately

• an Unsuitable Item that is not appropriate for donation flow and may require user confirmation or skip handling

This page is important because the AI does not work from generic common sense alone. It also relies on the prohibited-item list maintained by admin.

2. The Most Important Operating Rule

Admins must treat this list as a living operational control list.

The platform's default list is only a starting point. It may contain gaps, outdated assumptions, or items that the system fails to catch clearly in real traffic.

That means:

• admins should monitor what is actually appearing on the site

• admins should review missed cases, borderline cases, and new abuse patterns

• admins should update this list whenever platform behavior shows that the current defaults are incomplete

Do not assume the default prohibited-item list is complete.

This is one of the most important admin responsibilities for keeping AI moderation effective.

3. Where This List Is Used

The prohibited-item list is used by the AI analysis flow during item/donation review.

Operationally, that means changes made here affect how the AI is prompted to detect:

• prohibited content

• unsafe item types

• unsuitable donation categories

The system also caches prompt-ready prohibited-item text for AI usage, so this page is not just an internal note list. It is part of the actual AI decision support layer.

4. Two Categories on This Page

Safety Violations

These are items that should be treated as hard-stop problems.

Operational meaning:

• they are considered severe enough for immediate rejection behavior

• they should not be treated as skippable donation issues

Examples in the default logic include:

• adult/sexual content

• violence/gore

• drugs or drug paraphernalia

• real weapons

• stolen goods

Unsuitable Items

These are items that are not appropriate for normal donation flow, but they are not handled the same way as safety violations.

Operational meaning:

• they are unsuitable for the platform's donation model

• they may support skip/confirm behavior instead of hard safety rejection

Examples in the default logic include:

• vehicles

• large furniture

• large appliances

• clothing

• alcohol

• tobacco

• perishables

[Placeholder: Screenshot of the AI Prohibited Items page showing both Safety Violations and Unsuitable Items sections]

5. Why Admin Must Keep Updating This List

This page should not be treated as a “set once and forget forever” configuration.

There are several reasons:

• real user uploads change over time

• new abuse or edge-case patterns appear

• the AI may miss item types that were not clearly described in the current prompt list

• the default initialization list is broad, but it is not guaranteed to cover every operational risk

The correct mindset is:

The default list is the baseline, not the final answer.

Admin should continuously refine the list based on:

• what actually appears on the platform

• repeated moderation misses

• item types that are causing confusion in the upload flow

• legal, safety, or reputational risk changes

6. How to Review This Page Operationally

Admins should revisit this list whenever any of the following happens:

• a problematic item appears on the site and should have been blocked earlier

• the AI fails to recognize a clearly prohibited or unsuitable category

• support, moderation, or operations notices repeated upload mistakes

• the platform expands into new item patterns that were not covered well before

This should be an ongoing operational habit, not a rare maintenance task.

7. Standard Workflow

Step 1: Review Recent Platform Behavior

Before editing the list, review what is happening on the platform:

• what users are uploading

• which bad items slipped through

• which items are repeatedly flagged too late

• which categories confuse the AI or human reviewers

Step 2: Decide the Correct Category

When adding a new item type, first decide:

• is this a Safety Violation

• or is this an Unsuitable Item

Use Safety Violation when the item should be treated as a strong rejection case.

Use Unsuitable Item when the item does not fit the platform donation model but is not in the same severity class as a true safety problem.

Step 3: Add or Edit the Prompt Phrase

The Name field is important because it is the phrase used in AI prompt construction.

Write phrases clearly and concretely so the AI can recognize the category well.

Good examples are:

• specific item families

• grouped item descriptions

• realistic naming users would upload

Avoid vague internal-only shorthand that the AI would not interpret clearly.

Step 4: Save the Item

After saving:

• the prohibited item becomes part of the maintained list

• the system can use the updated active list in prompt generation and cached prompt text flow

Step 5: Observe Whether the Update Solves the Problem

After an update, continue watching platform behavior.

If the same missed case still appears, the list entry may need:

• better wording

• broader wording

• category adjustment

• additional related phrases

8. Initialize Defaults: What It Means

The Initialize Defaults button loads the built-in default prohibited-item list when the database list is empty.

This is useful for setup, but admins should understand:

• it is only a baseline starter set

• it does not guarantee complete protection

• it should not replace active operational maintenance

In other words:

Initialize Defaults is the starting point, not the finished policy.

[Placeholder: Screenshot of the Initialize Defaults button and Add Item button]

9. What Admin Should Pay Attention To

When maintaining this page, pay attention to:

• whether the list reflects what is actually appearing on the site

• whether the wording is clear enough for AI prompts

• whether any important prohibited category is missing

• whether a category should move from unsuitable to safety, or vice versa

• whether a repeated miss indicates the current list is too narrow

This page should be updated whenever the current list is not strong enough to reflect platform reality.

10. Common Cases

Case 1: A problematic item appeared on the site even though it should have been blocked

This is a direct signal that the prohibited-item setup may be incomplete.

Admin should review:

• whether the item type is missing from the list

• whether the wording is too vague

• whether it belongs under the wrong category

Case 2: The default list exists, so the team assumes no update is needed

That assumption is wrong.

The default list is only the initial baseline. Real platform traffic may expose gaps that require admin updates at any time.

Case 3: The AI is inconsistent on a specific item family

This often means the list phrase needs improvement.

Instead of assuming the AI “just knows,” admin should strengthen the prompt-side category wording through this page.

11. Day-to-Day Operations Suggestions

• Review this list regularly, not only when a major issue happens.

• Use missed moderation cases as direct inputs for updating the list.

• Treat the platform's real upload behavior as the source of truth for whether the list is sufficient.

• Do not rely on default entries alone.

• Update the list whenever site activity suggests the AI prompt coverage is incomplete.

The key operational principle is simple:

Admins must keep watching the site and keep this list current, because the system's default prohibited-item coverage may still miss things.