Designed the first AI feature editors actually trust
Designed the first AI feature editors actually trust
Context
Context
In early 2023, generative AI was just hitting the newsroom and trust was fragile. Editors at Business Insider worried about losing creative control or finding AI-generated content slipped into their stories without consent.
In early 2023, generative AI was just hitting the newsroom and trust was fragile. Editors at Business Insider worried about losing creative control or finding AI-generated content slipped into their stories without consent.
Problem
Problem
Tedious metadata tasks—URL slugs, summaries, tags—were time-consuming but critical. AI could clearly help, but no one had defined what "editorial-safe AI" should look like, and getting it wrong risked the trust of an already-skeptical newsroom.
Tedious metadata tasks—URL slugs, summaries, tags—were time-consuming but critical. AI could clearly help, but no one had defined what "editorial-safe AI" should look like, and getting it wrong risked the trust of an already-skeptical newsroom.
What I did
What I did
Designed an opt-in AI assistant during BI's internal Hack-AI-Thon, a sparkle pencil icon next to each field that generated suggestions only on demand.
Designed an opt-in AI assistant during BI's internal Hack-AI-Thon, a sparkle pencil icon next to each field that generated suggestions only on demand.
Hack Week prototype
Editors clicked the sparkle pencil to generate metadata suggestions, always on demand.
Editors clicked the sparkle pencil to generate metadata suggestions, always on demand.
Built one scalable interaction model—one icon, one behavior, one UI across all supported fields—reducing engineering lift and avoiding editor retraining.
Launched quietly given internal tensions, tracking adoption through analytics and trusted-editor feedback rather than a big announcement.
Built one scalable interaction model—one icon, one behavior, one UI across all supported fields—reducing engineering lift and avoiding editor retraining.
Launched quietly given internal tensions, tracking adoption through analytics and trusted-editor feedback rather than a big announcement.
Refined the UI post-launch to align with the evolving design system and clarify consent language.
Refined the UI post-launch to align with the evolving design system and clarify consent language.
Updated design
Clearer consent language, aligned with our evolving CMS design system.
Clearer consent language, aligned with our evolving CMS design system.
Outcome
Outcome
Became the first generative AI feature shipped in BI's CMS
Became the first generative AI feature shipped in BI's CMS
Adopted across 6+ field types and still in production
Adopted across 6+ field types and still in production
A year later
The same opt-in pattern powered story summary bullets, growing without added complexity.
The same opt-in pattern powered story summary bullets, growing without added complexity.