Suggested prompts for AIs in WhatsApp groups
Trust was working — activation wasn't.
After launching Meta AI in WhatsApp group chats across Mexico, Chile, and South Africa, the trust and transparency work I'd led was succeeding — 95.6% ToS acceptance, 79.1% day-14 return rate. But engagement data told a different story:
- 33.6% removal rate — one in three groups removed AI
- 78% of removals were by the same user who added AI — a "tried and rejected" pattern
- 82% of AI groups sat dormant on any given day
- Groups that did engage retained at 2× the baseline
The AI worked. People just couldn't get started. The welcome message I'd designed introduced Meta AI and explained how to interact — but at 50 words, it was passive text with no call to action. Users faced a blank composer and had to invent a prompt from scratch, which is too much cognitive load in a chat they opened to talk to friends.
No amount of text can replace a tappable action.
Not on the roadmap — I owned it end-to-end.
This wasn't assigned to me. As the solutions owner, I identified the activation gap in the data, proposed the solution, and drove the entire process — from hypothesis to spec to PRD to working prototype built with Cursor.
From signal to chips to screen real estate.
The 78% same-user-removal stat was the sharpest signal. Users who were curious enough to add AI — who accepted the ToS and read the welcome message — still gave up within a median of 4.8 minutes. The gap wasn't awareness. It was action.
I ruled out a welcome message A/B test (server-side plumbing, can't gate via ABProp, doesn't work in encrypted groups) and proposed something different: tappable suggestion chips below the welcome message that give users a one-tap path to their first AI interaction.
Research surfaced a validated pattern: WhatsApp's own null state suggestion pills (shipped January 2025) found that first-person utilitarian phrasing ("I need help with homework") drove +1.04% pDAU. I adapted this from first-person singular to first-person plural for groups — "Help us…" instead of "I need…"
I stress-tested all chip candidates against WhatsApp's voice and tone principles (7th-grade readability, active voice, casual and friendly). The stress test caught issues early — "Can you help us decide something?" failed for hedging; revised to "Help us decide something."
Recommended set:
- "Plan something fun for us" — action-oriented; planning is the #1 group use case
- "What should we do this weekend?" — question triggers reflexive response
- "Help us decide something" — positions AI as group facilitator
Content strategy · Chip copy & voice stress test
I built static mockups to visualize the chips in context. They immediately revealed something I hadn't anticipated: the 50-word welcome message plus 3 chips consumed the entire screen. No group conversation visible, no familiar chat context — violating WhatsApp's "simple" and "reduce cognitive load" principles.
This expanded the scope from "add chips" to "shorten the welcome message AND add chips." I proposed cutting the message from 50 to 25 words — because if chips are demonstrating capabilities below the message, the message doesn't need to list them. The chips are the capability demo.
Interactive prototype
Proposal — value in the process.
This is a proposal — not yet in production. The value is in the process: a content designer reading engagement data, identifying a product gap that wasn't on anyone's roadmap, and driving the solution from hypothesis through content strategy, prototyping, and a full PRD.
What prototyping revealed: The screen real estate problem — something no spec or deck would have surfaced. Without the mockups, I'd have proposed chips alongside a 50-word message and the combined experience would have overwhelmed the screen.
What it demonstrates: Content designers are positioned to find product gaps that others miss. We think about the moment-to-moment user experience — and sometimes the answer isn't better text. It's a tappable action.