← All retail signals
Creator economy

The Taste Graph

The machine maps what you love so precisely that discovery stops feeling like search and starts feeling like being known.

What is it?

Recommendation is becoming a model of your whole taste, not a log of clicks — and that model is what platforms now sell. What once ranked results behind a search box moves to the front, where you ask it directly. Whether it knows you or just guesses is unsettled.

Why now?

Clicks only recorded what you did; foundation models are cheap enough to infer what you want. And because the modelling is now much the same for everyone, the edge stops being who models best and becomes who holds the most behaviour — and controls the surface you ask through.

Who’s building it?

Platforms with dense first-party data — the model layer is off-the-shelf; the moat is the behaviour beneath it.

Data-rich platformsModel teamsQuery surfacesMerchants · advertisersUsers editing the model
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — platforms holding the deepest behaviour logs; technique is generic.
Buyers — advertisers buying access to the graph; increasingly the user, handed edit controls.
Gatekeepers — whoever owns the surface the demand routes through.
What’s at stake
The moment of demand shifts from a ranked list a seller can shape to an answer the platform controls.
An editable profile you can’t export stays a lock-in.
Recommendation moats have reset before — scale alone won’t hold one.
What’s unsettled
Does first-party data compound across model swaps, or reset with them?
Is being known something people want, or guessing dressed up?
Editability — a real check on the model, or theatre that changes nothing?
02 · How the demand matureda progression, not a timeline
Ranking
A log of clicks orders supply behind the box, easily swapped.
Taste modelling
Foundation models read taste whole; suggestion starts to feel anticipatory.
The queryable graph
The model moves to the surface you ask — capturing the moment of demand.
Owning demand
Whoever holds the graph and the surface holds the route to demand.
03 · Where the market goesthe forks, not a prediction
→Compounding or resettable — whether years of behaviour survive a model swap decides if scale is a moat or a head start.
→Portability is the tell — a ported profile commoditises the edge; a locked one keeps the platform’s power.
→Value concentrates at the surface you ask — the moment of demand — not the ranking behind it.
→Editability resolves either way — a genuine user check, or a governable-looking veneer.
Underneath every fork is an unsettled question: do people want to be known this way, or tolerate a good guess? If tolerance, there’s little to own.
04 · The Taste Economy deskadjacent markets, moving now
Post-Search Discovery
The answer arrives whole; blue links fade.
Google said at I/O 2026 that AI Mode has over 1 billion monthly users.
9to5Google
The Model Mirror
People shape themselves to be legible to AI.
50% of consumers use AI search; $750B flows through by 2028.
CMSWire · McKinsey
Hyper-Personal Media
Content spun up for a single viewer.
4M+ advertisers used Meta’s generative AI ad tools by 2025.
Marketing Dive
05 · The Wire
01
Spotify lets listeners edit the profile behind their recommendations
The Next Web ↗
02
Pinterest’s ‘Taste Graph’ maps people to interests and aesthetics
TechCrunch ↗
The panel on this signal
They agreed on
–The technique is going generic — the moat is first-party data and the interface, not who models best.
–The fight is a user-owned tool versus a platform-owned asset.
–Value concentrates where the graph becomes the surface you query, capturing demand.
The behavioral economist disagreed on
–Consumer response to accurate prediction splits into comfort and unease depending on framing; whether users treat an edited recommendation as evidence of trust or as a signal the model is guessing badly remains untested.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.