Pinterest Predicts and the forecast that designs the future

The Forecast That Might Have Been the Weather

In 2023, Mattel and Sweetgreen both cited the same document when explaining certain product and campaign decisions. That document was Pinterest Predicts — an annual report that launched in 2020 with an unusual structural promise: it would name trends before they peaked, not after. Most trend forecasting works retrospectively, confirming what early adopters already sensed. Pinterest Predicts claimed to invert that entirely, using platform search data that, by Pinterest's own account, precedes mainstream cultural adoption by months. The report's stated hit rate exceeded 80 percent. Mattel and Sweetgreen believed it enough to act on it. And when the trends arrived — as trends do — the report got credit for seeing them coming.

Which raises a question that is harder to answer than it first appears: at what point does a forecast stop being an observation and start being an instruction?

The empirical foundation of Pinterest Predicts is genuinely interesting. Pinterest's search behavior differs structurally from most social platforms. People search Pinterest privately, aspirationally, in planning mode — they are looking for something they want to do or make or become, not something they want to share. That makes the data a reasonable proxy for latent intent rather than performed preference. WGSN and Trendalytics have built substantial businesses on trend forecasting, but their source material tends to be public-facing: runway data, social posts, retail signals. Pinterest's claim was that it could see the desire before the behavior, which is a different kind of evidence. On that basis, calling it research seems fair.

But the report was never just data. James Hurst, working as Global Creative Director at Pinterest, built a visual language for the predictive work — which means the forecast was also a designed artifact, with its own aesthetic logic, its own palette, its own way of making an unverified future feel credible and actionable. That word, actionable, is where things get interesting. A designed forecast is not a neutral data release. It is an argument dressed in the visual grammar of authority. It says: this is what is coming, and here is what it looks like, and here is how you might move toward it.

When Mattel and Sweetgreen read that argument and moved toward it, the trend became more likely to arrive. Not because the forecast was wrong — it may well have been right — but because the brands that shape consumer culture were now pointing in the same direction the forecast had pointed. The prediction and the outcome became difficult to separate. Pinterest Predicts had, in some meaningful sense, designed the future it then got credit for forecasting.

This is not a criticism. It is a genuinely strange and underexplored dynamic, and it maps onto something James has written about in a different register. In Use Design To Design Change, there is a recurring argument that the most durable brands do not describe culture — they shape it, and then the culture catches up and calls the brand visionary. Amazon did not predict that people wanted same-day delivery; it built the logistics infrastructure that made same-day delivery a desire people could articulate. The brand did not follow the behavior. The behavior followed the brand. Pinterest Predicts seems to operate in the same territory, except the mechanism is compressed: the shaping happens through the act of forecasting itself, distributed to the brands who will then make it real.

The question that sits underneath this — and that a talk on creative direction at Pinterest would probably circle without quite landing on — is whether there is a meaningful ethical distinction between predicting a future and designing one. The honest answer is probably that the distinction has always been unstable. Every trend forecaster who publishes widely enough is also, to some degree, a trend generator. The difference with Pinterest Predicts is the precision of the claim. An 80 percent hit rate is not modesty. It is a number that invites trust, and trust invites action, and action closes the loop.

There is also something worth sitting with about what gets left out. A forecast with above-80-percent accuracy is, by definition, a forecast that does not surface the genuinely strange. The weird thing, the lateral thing, the thing that does not yet have a search cluster — that is not going to appear in a report built on existing platform data, however early-stage that data is. Pinterest Predicts is, structurally, a forecast of the almost-mainstream. Which may be exactly what consumer brands need. But it is not the same as seeing the thing that has no precedent yet.

James' second book, Weird the Normal / Normal the Weird, frames a tension between two moves: estranging the familiar until its assumptions become visible, and giving strange ideas enough structure to survive. The Pinterest Predicts report is interesting precisely because it does neither — it finds the thing that is already becoming normal and accelerates it. That is a third move, and it may be the most commercially powerful of the three. But it raises a question the report itself cannot answer: if the forecast only surfaces what is already emerging, who is responsible for the things that never emerge at all?

Maybe that is the question the next version of this kind of work needs to hold — not just what is coming, but what the act of predicting it makes impossible to see.