When seasonal demand actually starts moving

6.6 weeks

before the event, demand leaves baseline

1.1 weeks

before the event, demand peaks

5,571

measured event ramps

"Upload seasonal work three months ahead" is the most repeated piece of advice in stock illustration, and it is almost always given without a number behind it. We have the series, so here is the measurement.

For every event linked to a tracked subject, we take the weekly search series around that event's most recent PAST occurrence and measure two things: when interest leaves its baseline, and when it tops out. Both are counted in weeks before the event date. Using a completed season rather than a projection is the point — these are ramps that already happened. 5,571 profiles across 937 events.

The shape of a season

Across everything we measure, demand leaves baseline a median of 6.6 weeks before the event and peaks a median of 1.1 weeks before it. That second number is the surprising one: the peak is not weeks out, it is days out.

Christmas is the best-measured event in the corpus at 1,101 profiles, and its shape is clean: the ramp starts around 6.6 weeks out — the first half of November — and peaks about 1.6 weeks before the day, at 12.5 times baseline. Easter starts earlier at roughly 8 weeks and peaks harder into the final week. New Year barely ramps at all: onset and peak are the same 0.4 weeks, because it arrives inside the Christmas ramp rather than having one of its own.

What follows from it

Three things, in decreasing order of how confident the data makes us:

• ·Uploading in the month of the event is definitively too late. By the time a December upload clears review and gets indexed, the measured peak has already passed. This is the one conclusion the data states outright.
• ·The ramp, not the date, is the deadline. A file has to be live, reviewed and discoverable BEFORE onset — so for Christmas subjects, before early November, not before Christmas.
• ·The amplitude column is where the ranking is. A 12.5x ramp and a 4x ramp deserve different amounts of your time, and that difference is far larger than the difference between adjacent weeks of upload timing.

Method and limits

• ·Demand is weekly Google Trends search interest for the tracked phrase, not marketplace sales or download data. We have no access to anyone's sales figures, including our own users'.
• ·Each profile is anchored on the event's most recent past occurrence, and only ramps reaching at least 1.5x baseline are kept — below that the onset week is noise.
• ·Medians, not means, throughout. A handful of phrases with enormous ramps would otherwise carry the average.
• ·The per-event rows with single-digit profile counts (Chinese New Year, St Patrick's Day) are included for shape, not for precision. Christmas at 1,101 profiles and New Year at 506 are the rows to trust.
• ·Events with fewer than four measured profiles are omitted entirely.

The events we track and the subjects they move are public — see what to draw next for the method this feeds, or browse the corpus to find a subject and the events linked to it.