Images & Video · Podcast episode · source Feb 11, 2026
Main Image Redesign Polled First, Then Built with ChatGPT and Gemini
Find out why a main image loses on mobile and fix it before paying for a redesign
“from four votes in the initial poll to eight”
Per source From a vendor-affiliated source (see Worth knowing), not independently verified.
Copy-paste prompt
I sell [product] on Amazon. Attached: my main image and the top 3 competitors' main images as they appear in mobile search. Tell me why mine looks weaker at thumbnail size (product fill, aspect ratio, contrast, clutter). Then write one image-generation prompt for Gemini that keeps my product exactly as photographed and fixes those issues, plus 2 alternative concepts I can poll against it.
The source didn't publish its exact prompt. We wrote this starter so you can try the same workflow. Replace anything in [brackets].
What they did
On Helium 10's AM/PM Podcast, Hannah Lyss Tampioc of Mad Cat Creatives walked through her SORT framework on Bradley Sutton's Coffin Shelf listing, which looked small next to competitors in mobile search. Her team ran a PickFu poll on a mobile search-results mockup to learn what shoppers liked, then tested a taller image with no overlapping parts. For the new gift box she dictated her idea to a custom GPT in ChatGPT that writes image prompts, pasted the prompt and a product screenshot into Gemini, and assembled the main image in Canva.
How it went
In the follow-up poll the redesigned image went from four votes in the initial poll to eight. She also added the words 'coffin shelf gift box' to the box, since Amazon can read text in images, and ran a second poll to rank which secondary images matter most to set their order.
We only report what the source says. Results are self-reported unless noted.
Worth knowing
A small poll, not a sales test. The show is Helium 10's own podcast, so expect its tools to come up as the data source. The order is the lesson: decide whether you have a click problem or a conversion problem and test a mockup first, then use AI to make the image.