Feature entitlement: aiContentGuard · Plan: 👑 Add-on — not in a standard Device Plan

Prompt optimizer

Content Guard analyzes screenshots from your devices, and the accuracy of detection depends heavily on how you write your prompts. You have three ways to get a reliable prompt:

• Start from a predefined prompt — pick a tested, ready-to-use category. Fastest and recommended.
• Refine it with the prompt corrector — let Content Guard suggest improvements to a prompt you typed.
• Write your own — full control, following our best practices.

How prompts work

Content Guard evaluates each of your active prompt-based items against the screenshot and returns a match/no-match result for each one. Your prompt acts as a content description to detect — Content Guard looks at the screenshot and decides: "Is the content described in this prompt visually present?"

Key behaviors:

• When uncertain, Content Guard leans toward "not detected" to minimize false alarms.
• Each prompt must be self-contained — they are evaluated independently.
• Content Guard pays special attention to "does NOT qualify" exclusion clauses in your prompt.

Start from a predefined prompt

The fastest way to create a reliable prompt-based item is to start from one of our predefined prompts — a curated set of common detection categories that we have already tested and tuned for accuracy. Each one already follows the three-part structure described below, including the all-important "does NOT qualify" exclusions.

How to use them:

1. Open the "Add Item" drawer and choose the "Prompt" option.
2. Switch the mode toggle from "Type prompt" to "Predefined prompt".
3. Open the "Select a predefined prompt" dropdown and pick a category (e.g. Nudity Detection).
4. The Prompt field is automatically populated with the optimized description. Optionally adjust it for your environment, then Save.

tip

Predefined prompts are already optimized, so no AI suggestion is shown while one is selected. If you want to write your own prompt from scratch, switch to "Type prompt" mode — anything you typed there is preserved when you switch back.

Available predefined prompts

Content Safety

Rule	Prompt
Nudity Detection	"The screen must display fully exposed human breasts, genitalia, or buttocks with no clothing or covering present. Exposed body parts covered by clothing, artistic representations in paintings, or individuals wearing swimwear does NOT qualify."
Violence or Gore	"The screen must show explicit, real-life violence or gore with visible blood, injuries, or physical assault. Cartoon violence, action movie posters, or text descriptions of violent events do NOT qualify."
Profanity Text	"Visible words on the screen that are clearly and fully spelled-out obscene or vulgar terms. Common abbreviations, censored words with asterisks, or foreign language text that is not profane does NOT qualify."
Gambling Content	"Gambling-related content is visible: casino imagery, slot machines, poker chips, playing cards, dice, roulette wheels, betting odds, sports betting displays, lottery tickets, or gambling-related promotional text. The content can appear alongside other material."

Compliance

Rule	Prompt
Religious Symbols	"The screen must display a clear and prominent religious symbol such as a cross, crescent, Star of David, Buddha statue, or Hindu deity. Generic architectural features, decorative patterns, or artwork lacking identifiable religious iconography does NOT qualify."
Political Symbols	"The screen must clearly show a recognized political party logo, campaign sign, or election-related flag as the central visual element. Generic symbols, decorative flags, or non-political slogans does NOT qualify."
Legal Disclaimer	"Legal text must clearly display terms and conditions, privacy policy, or copyright notices with specific language like "©" or "All rights reserved." Generic small print, footers with unrelated information, or decorative symbols does NOT qualify."

Screen Health & Operations

Rule	Prompt
Error Messages	"An error message clearly displays words such as Error, Exception, Fatal, or Crash with a distinct background or alert icon. General notifications, warnings, or informational pop-ups without explicit error terminology do NOT qualify."
QR Code	"The screen must clearly display a square QR code with three distinct corner squares and a grid of black and white modules. Images containing only partial QR patterns, decorative squares, or regular barcodes does NOT qualify."
Countdown Timer	"The screen must display a countdown timer with distinct numerical values labeled as days, hours, minutes, or seconds in a decreasing sequence. A regular analog or digital clock showing the current time does NOT qualify."

Refine a prompt with the prompt corrector

When you write your own prompt in "Type prompt" mode, Content Guard automatically reviews it in the background and, if it can be improved, proposes a more precise version — for example by adding a missing "does NOT qualify" exclusion or making a vague description more specific.

How it works:

1. Type your prompt in the Prompt field. A couple of seconds after you stop typing, Content Guard analyzes the text and a short "Analyzing your prompt…" bar appears below the field.
2. If the prompt is already good, the bar simply disappears and you can save as usual.
3. If it can be improved, an AI Suggestion panel appears with the proposed rewrite and an "Accept Suggestion" button.
4. Click "Accept Suggestion" to replace your text with the improved version, or keep editing — every change triggers a fresh analysis.

If you click Save while a suggestion is still showing and unaccepted, Content Guard asks you to confirm with a "Suboptimal Prompt Detected" dialog:

• "Go Back & Optimize" returns you to the form so you can accept or refine the suggestion.
• "Yes, Save Anyway" saves your original prompt unchanged.

note

The corrector only runs on prompts you type yourself — it is not shown for predefined prompts, which are already optimized. If the analysis can't be reached, it is skipped silently and you can save without any interruption.

Writing your own prompt

You can always write your own prompt from scratch in "Type prompt" mode — whether or not you use the corrector's suggestions. These are the practices that make a prompt reliable. The single most impactful technique for reducing false positives is adding "does NOT qualify" exclusions.

Structure every prompt in three parts

1. What to detect — describe the visual content clearly.
2. What qualifies — give concrete examples or thresholds.
3. What does NOT qualify — list common false triggers.

Before and after

❌ Without exclusions (high false positives)	✅ With exclusions (accurate)
"Food on screen"	"PHOTOGRAPHS or realistic IMAGES of actual food or drinks are visible: meals, snacks, beverages, coffee, desserts. A text-based menu listing dish names without food photographs does NOT qualify."
"Red screen"	"The entire screen is a solid red color with NOTHING else visible — no text, no menus, no icons. A red-themed advertisement or red background with content does NOT qualify."

Techniques that work

• Describe appearance, not concepts — "Topless imagery, exposed body parts" beats "Inappropriate content". Content Guard analyzes what is visible.
• Give examples and thresholds — "THREE OR MORE consecutive words in ALL UPPERCASE letters" is far more reliable than "lots of caps".
• Distinguish photos from text — say "PHOTOGRAPHS of actual food, not a text menu listing dish names". Content Guard can confuse text about something with an image of it.
• Define exact screen states — for color/blank/error screens, set boundaries: "UNIFORM solid blue with NOTHING on it — if you can read ANY words, it is NOT a blue screen."

Common pitfalls

Pitfall	Problem	Fix
No exclusions	"QR code" triggers on any square pattern	Add: "A QR code with three corner squares. Map icons and decorative patterns do NOT qualify."
Too vague	"Text on screen" matches everything	Add threshold: "Dense paragraphs covering most of the screen. Menus or scattered labels do NOT qualify."
Instruction-style	"Check if there's a logo"	Description-style: "A company logo or brand mark is visible."
Ambiguous state	"Error screen"	Specific: "System error message: crash dumps, blue screen of death, 'Application not responding'."

Quick Tip

Check out our interactive Onboarding Tour directly in the app (click the question mark How does it work?). It will walk you through creating your first AI Prompt step-by-step!