How to Optimize for Google's New Multimodal Search Box

CH.01What actually changed at I/O 2026

Google called the redesign the biggest upgrade to its Search box in over 25 years. Per the official Search at I/O 2026 announcement, the box dynamically expands to give you space to describe exactly what you need, surfaces AI-powered suggestions that go beyond autocomplete, and lets you search across modalities, using text, images, files, videos, or Chrome tabs as inputs. It started rolling out in all countries and languages where AI Mode is available, on both desktop and mobile.

Two facts make this more than a UI refresh. First, a sixth of all queries already involve something other than typed text. Second, the system answering a multimodal query reasons over the inputs together rather than treating the picture and the words as separate lookups. We cover the broader shift in our breakdown of what AI Mode being the default means for search visibility.

The practitioner takeaway: your job is to make every modality on your pages legible to that reasoning. For how this fits alongside every other I/O 2026 change, see our complete breakdown of Google I/O 2026 for search marketers.

CH.02Why non-text optimization is now core SEO

When a sixth of queries are multimodal and the answer engine is a model that reads pixels as readily as text, the image you ship is a ranking input on equal footing with your H1. The surfaces that win are the ones that hand the model clean, structured, unambiguous signals.

A blurry stock photo with a filename of IMG_4821.jpg and empty alt text is invisible to a system trying to visually match a user's camera input. An original, high-resolution product shot, named descriptively, marked up with Product structured data, and surrounded by accurate copy, is a strong candidate. Optimizing the non-text assets is now a core discipline.

The query is no longer a string. It is a string plus a picture plus a tab plus context, and the engine reasons over all of it. Capconvert GEO practice

Here is how to do it, modality by modality, starting with the highest-leverage work for almost every site.

CH.03Image SEO: the foundation of multimodal visibility

Image optimization is the highest-leverage multimodal work for most sites, because images feed Google Images, Google Lens, visual matches, shopping surfaces, and AI Mode's visual reasoning at once. Google's image SEO best practices define the floor. Hit every item.

• Ship original, high-quality images. Use relevant, high-resolution photos and avoid generic logos or extreme aspect ratios. Original imagery beats stock for visual matching because it is distinctive.
• Write descriptive filenames. Google explicitly prefers my-new-black-kitten.jpg over IMG00023.JPG. Name files for what the image shows, in plain words.
• Write accurate, informative alt text. Describe the content, not keywords. Google's own example is "Dalmatian puppy playing fetch" over generic terms.
• Use real HTML image elements. Google finds images in the src of an img element and does not index CSS background images.
• Place images in topical context. Position images near relevant text on topically appropriate pages so Google can confirm what a photo depicts.
• Serve responsive images with picture or srcset and a fallback src. Supported formats include JPEG, PNG, WebP, SVG, and AVIF; match the file extension to the actual file type.
• Optimize for speed. Compress and size images correctly and verify with PageSpeed Insights, because heavy images suppress the page they live on.

CH.04Image structured data and image sitemaps

Structured data and sitemaps are the two technical levers that move images from indexed to eligible for rich treatment and reliably discovered. Add structured data so images become eligible for rich results and prominent badges in Google Images. The type depends on the page: Product for commerce pages, Recipe for recipes, and ImageObject where you need to describe a standalone image with its license, creator, and caption. Licensable-image markup can earn a badge that links to your licensing terms, which is useful for any site whose imagery gets reused.

Submit an image sitemap to surface images Google might not otherwise find, including images loaded via JavaScript. The rules are specific.

• Nest one or more image:image tags inside each url entry; each URL can contain up to 1,000 image:image tags.
• Put the image URL in image:loc inside each image:image.
• Declare the namespace http://www.google.com/schemas/sitemap-image/1.1.
• Images can be hosted on a different domain, for example a CDN, as long as both domains are verified in Search Console, and image URLs must not be blocked by robots.txt.

CH.05Google Lens and visual-match optimization

To win Google Lens and visual-match queries, give the system distinctive, original images of the exact thing a user is likely to point a camera at. Lens and Circle to Search work by matching a real-world or on-screen image against indexed images, so your odds rise when your images are unique, well-lit, shot from the angles people actually photograph, and unambiguously tied to a product or entity.

1. Lead with clean, isolated product shots plus in-context lifestyle imagesThe isolated shot helps the visual match; the lifestyle shot helps the model understand use and intent.
2. Show the distinctive features a person would photographThe logo, the silhouette, the label, the packaging. If your product has a recognizable shape, make sure an image captures it cleanly.
3. Keep product imagery consistent across your site, your Merchant feed, and third-party listingsThe same entity reinforces itself everywhere Google looks.
4. Tie images to structured product dataGTIN, brand, price, and availability so a visual match can resolve straight into a shopping result.

Visual search rewards specificity. A retailer with original, multi-angle, structured-up product images will out-match a competitor relying on the manufacturer's stock photo that a hundred other sites also use. This matters more as Google moves toward AI-driven shopping; see our piece on getting your products into Google's AI shopping agent.

CH.06Video discoverability for multimodal and conversational queries

Video is now a first-class input and output: users can attach a video to the new Search box, and conversational queries increasingly surface video answers. The same legibility principles apply, so give the engine clean metadata and structure.

• Mark up videos with VideoObject structured data including a thumbnail, name, description, upload date, and duration.
• Provide a representative, high-quality thumbnail.
• Supply transcripts and captions so the spoken content is machine-readable.
• Submit a video sitemap so Google reliably discovers embedded and JavaScript-loaded videos.
• Answer the specific question early and clearly, because the model is extracting an answer, not ranking a ten-minute watch.

Video and YouTube deserve their own playbook, especially now that YouTube has its own conversational search. We go deep on this in how conversational video search changes YouTube SEO; treat that as the companion to this section.

CH.07Entity and topical clarity so Google can reason across modalities

The connective tissue of multimodal optimization is entity clarity. Google can only reason across an image, a tab, and a sentence if it can resolve all three to the same well-defined entity. If your brand, products, people, and topics are ambiguous, the model has nothing stable to anchor a cross-modal query to.

• Use Organization, Product, and Person structured data consistently, and connect them with sameAs links to authoritative profiles like Wikipedia, Wikidata, LinkedIn, and official social accounts.
• Keep names, attributes, and descriptions consistent across your site, your structured data, your image alt text, and your off-site presence.
• Build genuine topical depth so that when a query spans text and an image, your page is the obvious match on the subject, not a thin page that happens to contain the keyword.
• Anchor every asset to its topic, so the image, the video, the copy, and the schema all describe the same thing. Conflicting signals dilute the match.

This is the same E-E-A-T and entity work that drives every other AI surface; multimodal search just raises the stakes by adding pixels and spoken language to the inputs the model weighs.

CH.08A practical multimodal optimization checklist

Work through this list for any page you want visible in multimodal and visual search.

1. Replace stock and low-resolution imagesUse original, high-quality assets shot from the angles users actually photograph.
2. Rename and re-caption every important imageDescriptive filenames and accurate, content-specific alt text.
3. Use real img markup, served responsively and compressedReal elements, srcset, and speed verified in PageSpeed Insights.
4. Add page-appropriate image structured dataAnd designate the primary image of the page.
5. Generate and submit an image sitemapAdd a video sitemap if you host video, and resubmit when assets change.
6. Mark up videos with VideoObjectAdd transcripts and captions, and front-load the answer.
7. Tie product images to complete structured product dataFor shopping and visual-match resolution.
8. Reinforce entity clarityOrganization, Product, and Person schema plus sameAs links.
9. Place every asset in topically relevant contextWith supporting copy that describes it.
10. Re-audit quarterlyBecause the multimodal share of queries is rising and the surfaces keep expanding.