In a reply to Sadie on X, who asked about Google images! News from Google explained why some of image results aren’t great.
Agreed — some of these results aren’t great. This is actually a difficult and long-standing technical challenge for all search engines (you can compare results on others). That’s because search engines reflect how things on the web are labeled — they rely primarily on the text… pic.twitter.com/kGtv4jQBmm
— News from Google (@NewsFromGoogle) September 5, 2025
Google wrote
Agreed — some of these results aren’t great. This is actually a difficult and long-standing technical challenge for all search engines (you can compare results on others). That’s because search engines reflect how things on the web are labeled — they rely primarily on the text surrounding the image, which doesn’t always match the content of an image as you might expect. As a more trivial example of this, look at the query “yellow dress and green socks”. We’ll continue to look for ways to improve this.
And speaking of what Google said “We’ll continue to look for ways to improve this” Google has already been working on better understanding and processing of images using Gemini family of models. Like in July they put out a new technique “Conversational Image Segmentation” which enables Gemini 2.5 model to now actually understand what is there in an image and its almost like Gemini getting eyes to see!
This bridges the gap between proper understanding of search user query and what Google can show to a user, even though there isn’t any public information from Google saying if they are going to use this technique (or some modification of it) for Google Image Search but research like this from Google Gemini and DeepMind teams is promising & in future may be will be powering Google Image Search making it way better than what it is now.
Leave a Reply