Can AI Design Shelf Talkers? I Tested a Stopper, a Wobbler and a Shelf Flag

I wanted to find out one thing. Ask an AI for three shelf pieces with trade names, a stopper, a wobbler and a banderola, and see whether it knows they're three different objects doing three different jobs.

I picked Oreo as the brand and ran it through three AIs: Gemini in Pro mode, ChatGPT with Images 2.5, its latest image model, and Bellto, an AI agent for POP campaigns that takes a campaign from the brief to shop drawings. It was a test. Oreo and the company behind it had nothing to do with it and aren't clients.

I've spent close to four years in point-of-purchase, and Bellto is what we're building now. The video is in Spanish. This is the written record of what came back.

What's the difference between a stopper, a wobbler and a shelf flag?

All three live on the shelf edge, in the family US stores loosely call shelf talkers. They make a shopper look in different ways, and that difference is the whole test.

A stopper is a rigid printed panel fixed to the shelf edge and standing perpendicular to it, out into the aisle. The US trade calls it an aisle violator. It has to be visible from down the aisle so it stops you as you walk, which is where the name comes from.

A wobbler is a small card on an arm, facing the shopper standing in front of the shelf. The arm lets it sway as people pass. It also has to be rigid enough to hold the card up, because a wobbler that can't carry its own card doesn't wobble, it falls.

A banderola is a narrow printed strip hanging flag-style from the shelf, which makes the closest US term a shelf flag. It usually hangs perpendicular to the shelf, the way a stopper stands, so it reads from far down the aisle, and it comes in more than one variant.

Turn a stopper parallel to the shelf and you have a strip. Give a wobbler an arm that can't hold and you have nothing.

How did I run the test?

I asked Gemini and ChatGPT for the pieces and let them work while I set up the campaign in Bellto.

With Bellto the request was different. I asked for all three in one message, a stopper, a wobbler and a banderola for Oreo, because Bellto is built to tell the pieces apart and make each one separately. It came back with a brief, which I accepted as it was, and I told it to go ahead. It proposed the three pieces, and once I created them it started generating two versions of each.

Since Bellto gives two versions per piece, I asked Gemini and ChatGPT for a second version too.

One thing I didn't give Bellto was a product photo, artwork or any reference. That shows in its images, and I'll come back to it.

Which AI got the stopper right?

Gemini. Correct. Its stopper stood perpendicular to the shelf, visible from down the aisle, which is the job, and the image reads like a real photograph. How it was fixed to the shelf is debatable, but on the whole it passed.

ChatGPT. Its stopper didn't do a stopper's job. It sat flat against the shelf, which makes it something closer to a shelf strip. That one failed.

Bellto. The first version was a stopper, exactly. Bellto sticks closely to its inputs and is focused on renders, so with no product input the piece came out bare. The second version was less correct, because it wasn't perpendicular to the shelf. With a right first version, the piece passed.

And the wobbler?

Gemini. This was the curious one. The arm was a spiral, and I don't believe a spiral like that would hold the card rigid. It would just drop. The two versions changed the graphic, not the hook or the arm. On top of that, it set the card perpendicular to the shelf, the way a stopper stands. Gemini didn't get the wobbler.

ChatGPT. Slightly better. It had trouble drawing the arm and came out as a kind of stopper, but you could argue the arm was rigid enough to make the wobbling movement, that sway the piece is named for. Its two versions were very similar, with a different die-cut shape.

Bellto. Two wobblers, and on both the arm is rigid enough to hold, unlike the spiral. The graphics were minimal, because I'd given it no product image. The second version also had a correct arm and came out much bigger than the shelf itself, again with no product to size it against.

A wobbler is small, and it still decides whether the shopper sees a moving card or a fallen one. If your next launch includes shelf pieces like these, the Bellto waitlist is where that brief goes.

And the banderola?

Gemini. Good. It had two different fixings, which is arguable, and it hung stopper-style, perpendicular to the shelf. There are different kinds of banderola, and I'd pass this one.

ChatGPT. The hardest one to read. At first glance the image was odd, with the product and a kind of information strip coming out of it. It read more like a stopper than a banderola, though it could fall inside the category, and both versions were curious. It didn't make much sense as a banderola, so ChatGPT missed this one.

Bellto. The first banderola again came without graphics, very much a render, because I'd given it no reference. The fixing was fairly correct and it hung stopper-style, perpendicular. The second version wasn't perpendicular, carried a little more graphic, and was also quite good.

So which AI knew the difference?

Piece	Gemini (Pro mode)	ChatGPT	Bellto
Stopper	Perpendicular to the shelf, reads from the aisle; fixing debatable	Flat against the shelf, closer to a shelf strip	First version a stopper; second not perpendicular
Wobbler	Spiral arm that wouldn't hold, card set perpendicular like a stopper	Arm drawn with trouble, came out stopper-like	Two wobblers with arms rigid enough to hold; second oversized for the shelf
Banderola (shelf flag)	Hung perpendicular, two different fixings	Product with an information strip coming out, closer to a stopper	First perpendicular with a fairly correct fixing; second not perpendicular, more graphic
All three right	No	No	Yes

Bellto got all three without the inputs it needs to do its best work. Gemini missed the wobbler. ChatGPT missed the stopper and the banderola.

What do I take from this test?

A photograph of the wrong piece is still the wrong piece. Gemini's images read like real photos, and its wobbler would have dropped off the shelf. A shelf piece is judged on its orientation and its arm, and on whether it does the job its name describes.

The names of these pieces carry geometry, and a general image model reads them as style. I went through why in why generic image AI fails at POP displays. This test is that problem on three small pieces.

Bellto took one request and split it into three pieces, each one the piece it was named. That's the part a campaign needs, because a launch rarely has one shelf material. It has a stopper here, a wobbler there and a flag down the aisle, and each one has to come out as itself.

Bellto sticks to what you give it. With no product photo and no graphics, it gave me plain renders of the right pieces. With the graphics and a product photo in the brief, I expect the result to be better still. The longer run, a full launch with a floor display, a wobbler and a shelf strip, is in my Bellto campaign test.

Two versions per piece mattered too. On the stopper the first was right and the second wasn't, so choosing is part of the work.

If you're a manufacturer and a client needs a concept image of one shelf piece rather than a campaign, shelf material, stoppers and wobblers included, is one of the display types in AI POP Displays. If you're a brand with a campaign that mixes stoppers, wobblers and flags across stores, put it on the Bellto list.

What we're building next: Bellto

Bellto is an AI agent for brands and agencies that takes a POP campaign from brief to shop drawings, and of the three AIs in this test it's the only one that ends in a manufacturable package. It proposes the campaign mix, floor stand, glorifier, shelf strip, stopper, with the budget split per store, and designs every piece with you at real scale in parametric 3D, with proportions derived from the product and the facings. A dimension change recomputes the whole piece and re-runs the manufacturing checks in seconds. Every approved piece comes out as a 3D model, a part-by-part cutlist, dimensioned drawings, a STEP of the assembly and a cut DXF per part, a package a workshop can quote without redrawing. The fuller reference is in what is Bellto.

We're building it now. The waitlist is open to any brand, and we're contacting the first ones soon to run the first real campaigns end to end. It opens by invitation, in small groups. Pricing goes first to the people on the list, and there's no card. POP manufacturers can sign up too.

If your next launch has shelf pieces that need to come out as the right pieces, join the Bellto waitlist and tell us about it.