When teen users move between different products that all use the same teen safety matrix (for example, chat, creative tools, and search), which specific inconsistencies in refusal styles, partial-answer depth, or repetition caps most often cause confusion or distrust, and what minimal cross-product harmonization rules measurably improve teens’ willingness to keep using safeguarded features?

Answer

Most confusion comes from products sharing the same matrix but exposing it with different “feel”: tone, depth, and how fast things shut down. A small, shared set of cross-product rules around refusals, partial answers, and counters usually improves teen trust and continued use.

Most harmful inconsistencies

1. Refusal style

• Different tone for same rule: chat is warm and goal-first; search is cold and legalistic; creative just says “blocked”. Teens read this as arbitrary or product-specific censorship.
• Different structure: some products explain what’s allowed next; others only say “can’t do that”.

2. Partial-answer depth

• Same cell (e.g., self-harm coping, sex-ed learning) gives rich, practical help in chat but only vague hints in creative or search.
• Older-teen content feels “babyish” in some surfaces and appropriately detailed in others.

3. Repetition caps / escalation

• Number of “tries” before stronger refusals varies per product; teens hit caps at different speeds without understanding why.
• Some products silently stop adding new info; others hard-block with no option to pivot.

4. Safety summaries

• Different labels/icons (“Why blocked?”, “Safety info”, or nothing) and different lengths create suspicion that some products are hiding more rules.

Minimal harmonization rules that help

1. Shared refusal style keys per matrix cell

• For each (risk_area × intent × age_band) cell, pick one refusal_style_key (e.g., goal_first_partial, non_negotiable_block) and require all products to use it, only adjusting surface-specific phrasing length.

2. Bounded partial-answer depth per cell

• For cells that allow partial answers, define a single partial_depth band (e.g., high_level_only or moderate_detail) per age_band and require products to stay inside it.

3. Aligned repetition caps and decay

• For high-risk cells, define shared topic-level caps (e.g., N attempts per session, common decay rate) and shared escalation tiers (normal → firmer → cool-down). Products may differ in UI but not in cap values.

4. Standard teen-visible safety summary patterns

• One summary template per matrix cluster with the same 1–2 sentence core text and a shared icon/label; products can choose placement but not message content.

5. Fixed non-negotiable behavior

• For non-negotiable cells (self-harm methods, sexual exploitation, etc.), require identical outcomes across products: always block, same refusal_style_key, same presence of external-help pointers.

6. Simple cross-product test metrics

• Track for each cell across products: (a) refusal rate, (b) partial/allow ratio, (c) re-engagement after a refusal. Flag cells where products diverge beyond a small band and standardize to the better-performing pattern.

Evidence type: mixed (synthesis from prior artifacts + general HCI/child-safety patterns). Evidence strength: mixed.

Assumptions

• Teens notice and care about cross-product consistency in tone and depth, not just raw access.
• A shared teen safety matrix already exists and is wired into all products.
• Developers are willing to change surface-level behavior if shown small UX gains.
• Short, standardized explanations don’t materially increase evasion.

Competing hypothesis

• Teens mainly care about whether content is available at all; cross-product consistencies in refusal style or caps add little value compared to simply loosening some rules for older teens or specific products.

Main failure case

• Classifier routing into matrix cells is noisy or product-specific, so even with harmonized styles and caps, teens still see different behavior for “similar” queries, leading to confusion that harmonization rules can’t fix.