This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
Most reputation work starts with a query. Someone types a company, a product or an executive’s name, and the job is to understand what comes back. Google’s update to AI Mode this week adds a path where nobody types the name at all.
What Google added
On Monday, Google said it is opening AI Mode, its experimental search mode in Labs, to “millions more Labs users in the U.S.” and adding multimodal search. Users can “snap a photo or upload an image, ask a question about it and get a rich, comprehensive response with links to dive deeper.” The feature pairs Lens with a custom version of Gemini and is available in the Google app on Android and iOS. Google also said AI Mode queries are, on average, twice as long as traditional searches.
How a photo becomes a set of searches
Google’s description lays out a three-step pipeline, and each step matters.
The first is scene understanding. Google says Gemini can “understand the entire scene in an image, including the context of how objects relate to one another and their unique materials, colors, shapes and arrangements.”
The second is identification. Lens “precisely identifies each object in the image.” In Google’s example, that means every book on a shelf.
The third is fan-out. AI Mode “issues multiple queries about the image as a whole and the objects within the image.” For the bookshelf, those queries look up each title and find similar, highly rated books, and the answer is a list with links to learn more and to buy.
Last month I wrote about query fan-out for typed questions, where the searches that retrieve a page are ones the system writes for itself. The image version adds a step in front. The system first decides what is in the picture, and that decision becomes the query.
What this means for brands
You can be retrieved because you were in the frame. A kitchen counter, a store shelf, a hotel lobby or a street of storefronts contains brands nobody asked about. If the system identifies them, it can search for them. A user who asks “what is a better option than this?” may get an answer that names, compares or ranks your product without ever typing it.
Identification is now a reputation step. If Lens reads your packaging as a competitor’s, or matches your product to an older or discontinued version, the fan-out queries are wrong before any ranking happens. The answer can be well sourced and still be about the wrong thing. That makes visual consistency part of being recognized correctly: distinctive packaging, a legible product name on the item itself, and current product images on your own site and with your retailers.
Context becomes part of the query. Google stresses that the model understands how objects relate to each other. A product photographed next to a rival, or in a setting associated with a particular use, may be looked up in that context. You cannot control the photos people take. You can make sure the text about each product (name, model, category, intended use) is clear wherever the system is likely to look it up.
The answer comes with a purchase path. Google’s example ends with links to buy. A comparison generated from a photo leads straight to a transaction, so a misidentification costs more there than in a summary someone reads and closes.
The practical shift
Keyword hygiene asked what people find when they search for you. Visual hygiene asks what a system decides you are when it sees you, and what it searches for next. There is no list of queries to monitor, which makes it harder to track. A reasonable start is to photograph your own products the way customers would, in ordinary settings and next to common alternatives, run the images through Lens and, if you have Labs access, AI Mode, and record what the system identifies and what it recommends.
AI Mode is still a U.S. Labs experiment. But the pipeline Google described (see, identify, fan out, answer) is a design choice, and design choices in search tend to spread.