AI-powered classroom engagement: what worked in 2025

2025 was the year AI-in-classrooms stopped being a pilot and started being a default. Here's a roundup of the patterns that actually stuck — what worked, what fizzled, and where we're pointing tools in 2026.

What worked in 2025

Anonymous AI-moderated questions

Classroom tools that let students post questions anonymously (routed through a Chrome sidebar with AI filtering for duplicates and spam) saw a 3–5× increase in participation vs. raised-hand Q&A. The key wasn't the AI, it was the anonymity. But AI was what made anonymous at scale tractable without flooding the instructor with garbage.

Voice-triggered polls in lectures

Professors running polls every 10 minutes via voice commands (instead of pre-loaded Slido polls) moved from "we ran 0 polls in the semester" to "we ran 40+ polls per class on average." Formative assessment shot up. Final exam averages improved 4–8 percentage points in classes that ran voice polls vs. control.

Live captions in student's native language

International student cohorts with live captions in their L1 had noticeably higher course completion rates. The effect size was largest for Asian-language cohorts (Chinese, Japanese, Korean) where the projected-English lecture and native-language captions closed what had been a 20–30% comprehension gap.

Screenshot-annotated questions

Students who can attach a photo of the confusing slide — instead of typing "I don't understand slide 14" — give instructors dramatically better signal. The AI can cluster photos, letting the professor see "these 12 students are confused about the same thing" at a glance.

What fizzled

AI-generated personalized study plans

Lots of ed-tech shipped AI coaches that generated per-student study plans. Students rarely used them. The pattern: students wanted AI help during class (questions, translations, clarifications) not in asynchronous coaching mode.

Avatars and gamified participation

Points, leaderboards, virtual classrooms with avatars. Mostly dead on arrival. The pattern here is that students don't want a new app — they want their existing browser to get smarter.

AI-graded essays (for primary work)

AI grading was fine for grammar and structural feedback but not ready to replace human assessment of ideas. Worked well as a first-pass filter before human grading, didn't work as a replacement.

What to try in 2026

Real-time cluster-based re-explanation

When the AI detects 30% of the class is confused about the same topic (based on questions + poll results + attention signals), automatically prompt the instructor to re-explain that section. A dashboard that says "32% of the class didn't follow the part about X" is more useful than "here's the grade distribution on Thursday's quiz."

Cross-class knowledge graphs

Questions from class 1 inform class 2 at scale. If a concept needed three extra minutes in one section, flag it for the parallel section's instructor.

Tighter integration with Chrome-based decks

Most classroom decks are now in Google Slides, Keynote Web, or Notion — all Chrome-based. A sidebar that pairs natively with the active tab (rather than a separate app or second screen) is where the action is.

Voice-first moderation for the instructor

Right now instructors still tap buttons for most moderation actions. Voice-first moderation ("dismiss that last question," "mark that one answered") could free the instructor from the mouse entirely.

What TA pilot is working on in 2026

TA pilot is the Chrome sidebar side of this ecosystem. Our focus for 2026:

• Better voice models for non-English lecture languages
• Instructor dashboard with cluster-based confusion detection
• Cross-session question carry-over (so common confusions don't have

to be re-discovered every semester)

If you teach a live class and want to try an AI sidebar, install TA pilot — it's free for classes of any size.