Multimodal distinctness: when a brand is searched not with words

Throughout the AI100 corpus, we have discussed visibility in the context of text queries: the user types a question, and the model produces an answer. But the world of search has long ceased to be reducible to typing words on a keyboard. A user photographs a product in a store and asks, “How much is this online?” They say aloud, “What model is this?” while pointing the camera at a pair of headphones. They upload a screenshot from Instagram and ask, “Find something similar, but cheaper.” They record a video and add a text question: “What material is this made of?”

These are not exotic scenarios. Google Lens processes more than 20 billion visual queries per month, and 20% of them are shopping-related [1]. AI Mode is integrated with Google Lens: a user can take a photo or upload an image, and the system, using Gemini’s multimodal capabilities, analyzes the entire scene—objects, their context, materials, colors, and shapes—and produces a composite answer [2]. ChatGPT with GPT-4o processes images, voice, and text simultaneously. 27% of mobile users already use voice search [3].

For a brand, this means that text optimization is a necessary but no longer sufficient condition for visibility. If your product cannot be recognized from a photo, if your YouTube video has no transcript, if a voice assistant cannot link the spoken company name to the correct entity, you lose the audience that searches without words.

Visual search works in a fundamentally different way from text search. The user does not describe what they are looking for—they show it. Convolutional neural networks (CNNs) transform an image into a numerical vector and compare it with a database of indexed images [4]. This means that the quality, consistency, and technical accessibility of images on a site directly affect whether your product will be found.

For e-commerce, the consequences are most obvious. A shopper sees a dress on the street, photographs it, and Google Lens shows similar products with prices from different online stores within three seconds. If your product images are low quality, lack descriptive alt text, lack Product schema, and do not follow a consistent photography style, they will not make it into that selection. A competitor with clean, marked-up photos will.

Visual consistency across platforms is also becoming a factor. Google Lens is better at recognizing brands that use the same photography style across their website, marketplaces, and social platforms. A fragmented visual layer makes it harder to tie content to an entity [5].

Voice queries differ from text queries not only in modality, but also in structure. A person speaking aloud uses natural sentences: “What’s the best café near me that’s open right now?” instead of “cafe near me open.” Queries in AI Mode are, on average, three times longer than ordinary search queries [6]. This means that content optimized for short keyword phrases may not match the way people formulate queries by voice.

For a brand, the practical implication is clear: FAQ sections written in a question-and-direct-answer format work better for voice search than long marketing texts. Structured data (FAQ schema, HowTo schema) helps voice assistants extract a specific answer. The brand name should be pronounceable and unambiguous—a model that cannot connect the spoken “Exco-Data” to the entity “ExcoData” will lose the brand in a voice query.

AI systems are making increasing use of video content. YouTube video transcripts become a source for citation: if an expert in your video explains in detail how the product works, and the transcript is available, the model can extract a passage from it for an answer. If there is no transcript, the video remains invisible to the text layer of the answer system.

Google explicitly states that AI Mode uses multimodal analysis: the system works simultaneously with text, images, video, and context [2]. For a brand that publishes educational videos, reviews, or product demos, a clean and accurate transcript is not optional—it is a condition of discoverability.

Multimodal optimization does not require a revolution. It requires extending familiar work into new formats.

Images: high quality, descriptive filenames and alt text, Product schema tied to specific products, and a consistent photography style across platforms.

Voice: FAQ sections in question-answer format, HowTo schema for instructions, and a pronounceable, unambiguous brand name.

Video: transcripts for every video on YouTube and on the site, VideoObject schema, and descriptive titles and metadata.

General layer: the same principle as for text visibility—structured data, machine readability, and external confirmation. Multimodality does not replace these foundations; it adds new input channels on top of them.