Genre missed the reason
A viewer chose Sinners for its Southern Gothic qualities, then received horror recommendations despite not broadly enjoying horror.
Original product research · Streaming discovery
Intent Layer
Recommendation systems know what you watched. What if they understood why?
An independent exploratory study of how mood, attention, social context, time, and story qualities shape what a viewer wants from a particular streaming occasion.
consumer researchintent modelingpersonalizationUX strategy
intent 01
“I want to feel understood”
intent 02
“I need something comforting”
intent 03
“I loved their dynamic”
occasion
“I only have 40 minutes”
viewing mode
“I want something in the background”
54
survey starts
44
eligible viewers
27
submitted responses
8–10
interviews planned
method · August 2026
Exploratory, directional evidence
I designed an anonymous survey around real viewing occasions—not general taste—to examine where recommendation context breaks down.
Question-level sample sizes range from 24–43 because valid partial responses were retained. Of respondents reporting age, 96% were 18–24, so these findings primarily reflect Gen Z behavior and should be treated as exploratory.
what the survey surfaced
Useful recommendations can still miss the moment.
Percentages use valid responses for each question; sample sizes are shown individually.
97%
n=34
Repeated recommendations
frequently saw the same titles repeatedly
83%
n=24
One-title overreaction
received unwanted recs after one title
79%
n=34
Browsing friction
sometimes browsed longer than they wanted
62%
n=37
Situational choice
said mood or viewing mode shaped selection
62%
n=24
Pre-watch preference
preferred clarifying intent while browsing
52%
n=33
Misunderstood intent
recalled a platform misreading why they watched
the numbers in practice
Three moments where similarity misunderstood the viewer.
A shared account distorted taste
After someone else watched a war movie on one respondent’s account, war titles began filling their recommendations.
Tone mattered within genre
A viewer wanted a particular kind of rom-com, while recommendations grouped titles with very different levels of romance, humor, and sincerity.
the product opportunity
Similarity asks what resembles the last title. Intent asks what fits what the viewer wants right now.
the intent layer
Capture what matters for this viewing occasion.
Intent supplements watch history and content similarity. It does not replace them; it gives those systems a human reason to start from.
pre-watch
Shape what fits now
Choose a few occasion and story signals while browsing.
post-watch
Learn what worked
Clarify the relationship, tone, setting, or dynamic that mattered.
chosen familybackground viewingdark comedysomething comfortingshort on timesomething new
01
Optional
Support indecision without interrupting viewers who already know what they want.
02
Low effort
Select only a few signals, then skip, remove, or refine without restarting.
03
Compositional
Combine mood, viewing mode, narrative qualities, time, and social context.
04
Explainable
Show which selected qualities each result matches.
living product deck
Research in motion
research complete · prototype in development
start small · build toward the layer
The path from signal to product
01
Research
Study when context breaks down.
02
Taxonomy
Turn motivations into usable language.
03
Pre-watch MVP
Prototype the moment of indecision.
04
Learning loop
Test lightweight post-watch feedback.
key takeaway
Watch history explains the past. Intent can shape what comes next.
The opportunity is not more recommendation noise. It is a lightweight signal that the platform cannot reliably infer on its own—and an experience that lets viewers correct the system without managing it.