Essential metadata
Present in every context
- What is this destination called?
- What could I do there?
What I did
- Reframed the success metric: I proposed measuring state of intent, a confident yes or a confident no before stepping through, instead of raw portal entries.
- Wrote the earliest shared artifacts: the jobs to be done, the stages of travel, and the design principles the travel v-team made decisions from, carried to their final versions in the leads review deck.
- Set which piece of information each distance had earned the right to show.
Content design owned the argument: the distance model and the metadata strategy, defended piece by piece against real pushback. Our team's product designer and art director worked out the physical design: where each element sat, and the portal's proportions relative to human height and door frames.
Problem: the tracked metric rewarded blind arrivals
In Horizon, a portal is a floating oval that works like a doorway between worlds. Portal entries were the tracked success metric, and they rewarded exactly the pattern causing the problem: an uninformed arrival counted as a win even when the person looked around and left immediately.
Solution: a portal reveals more as you approach
A body standing in a room has no menu to defer to, so the only variable deciding what someone sees is how close they've chosen to walk:
- Six meters or more
- A title and nothing else: enough to register that something is there, not enough to weigh it.
- Pointing at mid-range
- The one available action appears, because aiming at something is already a small commitment of attention.
- Up close
- The full picture arrives. That's where the decision actually gets made, so it's the only distance that earns it.
UXR testing decided what filled each distance. I standardized those answers into one model:
Challenge: adoption was negotiation as much as design
- Localization on a clock: world creation had just expanded its metadata from one field to three, and stacked that deep the layout pushed what would physically fit. I worked with localization experts so the portal held together in our longest languages before the design settled.
- One matrix for every portal: the home portal, the doorway into someone's created world, the portal a friend spawns beside you. A matrix decided which fields each context earned, and it set the precedent for what metadata belongs on every travel surface. I worked the structure through with the platform and OS teams whose surfaces would inherit it, and carried it before design leadership.
Not everyone agreed. The two objections I heard most: surfacing all of this would decrease portal entries, and portals should stay mysterious. The mystery camp had the early momentum. A design leader's review comments about how people actually decide where to travel became my argument inside the travel team, and the entry objection dissolved under the state-of-intent frame: a confident pass that saves someone's time is a win, not a loss on the entry count.
The piece I fought hardest to keep was "View Details," the second action for the person one beat from committing who wanted a last look first. Against the objection that it added friction, my argument: it's the only thing standing between a guess and an actual yes.
Proof: the portal went on stage at Meta Connect 2024
Mark Zuckerberg introduced the portal at the Meta Connect 2024 keynote as "a new kind of primitive" he expected to become "kind of like the equivalent of a hyperlink in the Metaverse." By then it was already established internally as Horizon's canonical entry point for embodied travel.
The experiment that followed tested the portal against a control with no portal at all:
+7–8%Horizon monthly and weekly actives rose roughly 7 to 8 percent
+6%daily actives about 6 percent
+4%time spent about 4 percent
+3 ptsday-one retention improved about 3 points; the longer horizons held flat
76%new users finishing first-run setup went from 55 percent to 76 percent once "Go" existed on the portal
The test surface included the Go button. It measured the portal surface with its essential-choice CTA, not the metadata model in isolation.
Afterward, I shared the portal strategy with the onboarding team for its prototypes teaching people how to use portals. The product designer and I advised on proposed additions including an AI label, and published internal and go-to-market terminology guidance.
Every content design question I had worked on before started from what got tapped. This one started from where a body was standing, and that swap rewrote the hierarchy: how much appeared on screen at once, and exactly when each piece was allowed to show.
