What our first 51 AI gift designs showed

Upload je foto: hij verschijnt meteen op het product, met prijs.

Mijn foto uploaden ✦Of beschrijf je ontwerp, de AI tekent het

T-shirt vanaf $23.00 incl. verzending

Measured 25 September – 1 October 2026 · about 5 min read

The short answer

Over the first week our AI design studio was open, three out of four people did not upload a photo — they described an idea in one sentence and let the AI draw it. The second finding surprised us more: after the T-shirt, the most-opened product was the framed print, ahead of the mug. Sample: 51 generations by 47 visitors, 25 September–1 October 2026.

Upload a photo and see it on the itemInstant preview, no account. From $23.00, shipping included.

Printed to order, shipping included.

Finding 1 — 75% described an idea instead of uploading a photo

Of 51 generations, 13 started from an uploaded photo (25%) and 38 from a written description (75%). We built the shop expecting the opposite ratio: the whole category — photo mugs, photo frames, photo calendars — assumes the customer arrives with an image in hand.

What people typed instead were things only they could have asked for: a pet in a costume, a private joke between two people, a scene from a holiday nobody photographed. The practical consequence for anyone selling personalised gifts: the blocking question is not “do you have a good photo?” — it is “can you say what you want?”.

Finding 2 — the framed print beat the mug

Counting which product people opened the studio on: T-shirt 52%, framed print 19%, mug 12%, then canvas, hoodie, tote, postcards and calendar at 2 to 5% each.

The T-shirt leading is unremarkable. The framed print taking second place, ahead of the mug, is not what we would have predicted from the price alone — the frame costs roughly twice the mug. A plausible reading, which we cannot prove at this sample size: a generated artwork is treated as art, and art goes on a wall before it goes on a cup.

Finding 3 — 90% of generations completed, in 3.6 seconds on average

46 of 51 generations (90%) completed. The average time between asking and seeing the image was 3.6 seconds. Five attempts failed or were abandoned mid-way; that is the number we are trying to move next.

Method, in full

Data from our own counter on posetondesign.fr, covering 25 September to 1 October 2026 — the first full week the AI studio was publicly open. 51 generations, 47 distinct visitors. Internal testing and identified bot traffic are excluded. We record counts and timings only: no prompt text, no uploaded image and no personal data are stored or published.

The honest caveat: n=51 is small, from one young shop, over seven days. These are observations, not laws. We publish them because very little public data exists on what people actually do with AI design tools in a buying context, and because a dated, transparent small sample beats a confident round number.

What we are doing about it

The finding points at our own copy: like everyone else in the category, we lead with “upload your photo”, while three quarters of visitors want to type a sentence instead. We are reworking the entry point so that describing an idea is presented as a first-class path rather than a fallback, and we will report whether it moves the completion rate in the next measurement.

Citing this These figures may be quoted freely with a link to this page and the measurement date (25 September–1 October 2026, n=51, posetondesign.fr). We will republish with a larger sample.

Questions we get

How many designs is this based on? 51 AI generations requested by 47 different visitors between 25 September and 1 October 2026 — the first full week the studio was open to the public. It is a small sample from a young shop, and we say so rather than rounding it into a bigger-sounding number.

How was the data collected? By our own counter, on our own site: each generation records a timestamp, whether a photo was attached, whether the generation completed, and how long it took. Internal tests and bot traffic are excluded. No prompt text and no personal data are stored or published — we report counts, never content.

Why does the photo-versus-description split matter? Because the whole personalised-gift industry is built on the assumption that you arrive with a photo. Three quarters of our visitors did not. If that holds at scale, the addressable moment is not “when you have a good picture” but “when you have an idea”.

Is a 90% completion rate good or bad? It is the figure we watch. Five generations out of 51 failed or were abandoned mid-way. We would rather publish it than hide it: it sets the bar for the next measurement.

Can other people cite these numbers? Yes, freely, with a link back to this page and the date. If you are a journalist or researcher and want the breakdown in more detail, the contact form reaches us directly.

Will you update this? Yes — this is the first measurement, dated. We plan to republish with a larger sample and, if we add the tracking, a breakdown by art style. The method and the caveats will stay published alongside the numbers.

Start with a photoNothing is sent before you order: the preview is computed in your browser.