Google researchers analysed 2.1 million Gemini Nano Banana conversations and found people using the image generator for everything from marketing and product visualisation to educational and technical material. The study gives a rare, large-scale look at what people ask an image model to do, beyond the memes that usually get all the attention.
Google DeepMind Watch analysis
What happened
Search Engine Journal reports that the researchers grouped conversations into five broad categories: social and viral images; brand and marketing; product visualisation and e-commerce; architectural and spatial design; and educational and technical information. They examined how people described image editing, artistic styles, composition, interactions and other features.
The reported findings include people appearing in nearly 80 per cent of images, and requests to start again being about twice as common as requests for minor changes. Memes and humour were the least prevalent features in the social and viral category. In marketing conversations, prompts often specified exact text and typography, and the researchers found many examples involving small businesses, individuals or fictional brands rather than established companies.
The researchers used prompt and conversation patterns, rather than examining uploaded or generated images, to study use. That is the method described in Search Engine Journal’s account, not a blanket guarantee about privacy.
Why it matters
This is a more useful picture of image-generation demand than a parade of viral examples. People are using these tools for practical work, including visualising products, building promotional material and creating educational illustrations. That broadens the conversation from “can it make a funny picture?” to what kinds of creative and commercial work are becoming easier to attempt.
The study does not settle whether AI is enabling entirely new work or making existing work cheaper, and conversation patterns cannot tell us everything about the finished images or their impact. Still, millions of interactions give researchers a substantial window into how people are experimenting.
Our read
The telling result is not that people make memes. It is that memes were the least common feature in the social-and-viral category, while detailed commercial and practical uses showed up across the wider sample. The public image of image generation may be lagging behind what people are actually doing with it.
What to watch
- Whether the full paper explains the sampling and classification methods in enough detail to assess the findings.
- How use patterns change as image generators add editing and multi-step features.
- Whether follow-up research can distinguish new creative work from work that has simply become cheaper to produce.
Discussion spark: If image generators are increasingly used for business and educational work, should we judge them mainly by what they enable, or by the creative and professional work they may replace?
Sources and evidence
- Source update (11 November 2026, 14:18 UTC)
not affiliated with, endorsed by, or operated by Google or Google DeepMind