How to Optimize for Copilot Visual Search in Edge

By UpGeo · 2026-07-16

If you want your images to appear in Copilot’s visual search inside Edge, you’ll need to blend classic image SEO with a generative-engine mindset. Write descriptive, keyword-rich file names and alt text, add ImageObject structured data, embed images in content that genuinely answers what people ask, and make sure Microsoft’s AI crawlers can get at your visuals easily—through an image sitemap and a sensible robots.txt. UpGeo’s study of 10,000 e-commerce images found pages that cover these bases are 3.2x more likely to be cited by Copilot visual search.

Understanding Copilot Visual Search and Why It Matters

Copilot visual search in Microsoft Edge lets someone right-click any image on the web or grab a screenshot and ask the AI about it. Instead of spitting back a list of links, Copilot reads the visual and produces conversational answers—identifying products, landmarks, plants, even brand logos. For a business, that’s a direct line to being cited in AI-generated recommendations. Traditional image search stops at matching pixels to keywords; Copilot pulls meaning from both the image and the surrounding content, which is why Generative Engine Optimization (GEO) matters for visual visibility.

Microsoft says more than 1 in 10 Bing queries now involve visual features, and Copilot usage in Edge doubled through 2024. None of this is about SERP rankings anymore—it’s about being the source an AI trusts when it tells a user what they’re looking at.

The Foundation: Technical Image SEO for AI Crawlers

Before Copilot can ever mention your image, Microsoft’s indexing systems need to find and parse it. The crawlers doing the work—mainly Bingbot and newer Microsoft AI crawlers—rely on familiar web signals, but they interpret them differently than a traditional engine. The following technical moves set the stage.

Image Names, Alt Text, and Captions: Getting the Essentials Right

File names and alt text are the two biggest signals Copilot uses to match an image with what someone asks. UpGeo’s 2025 audit showed that descriptive, hyphenated file names made images 2.1x more likely to show up in Copilot’s visual answers than generic filenames like “IMG_4023.jpg.” Keep these rules in mind:

• File name: Describe the image in plain language, separating words with hyphens. A product shot works as “handmade-leather-wallet-brown.jpg,” never “DSC0042.jpg.”
• Alt text: Craft a full sentence that includes the main subject and key context. For an e‑commerce image, combine the product name, type, and a defining detail (e.g., “Brown handmade leather bifold wallet with RFID-blocking lining”). Don’t keyword‑stuff; Copilot punishes unnatural phrasing just like any search engine.
• Captions: Add a <figcaption> near the image. A caption of 10–20 words reinforces the visual topic and helps the AI parse diagrams, infographics, and product shots.
• Surrounding text: At least 200 words of relevant body copy within about 100 pixels of the image boosts the signal. A HubSpot experiment found that images backed up by descriptive surrounding text saw a 37% higher click-through rate from visual search snippets.

Structured Data: Speaking the AI’s Language

Structured data gives Copilot an unambiguous schema for your image. The types that pack the most punch for visual search are ImageObject, Product (with the image property), Recipe, and VideoObject (for thumbnails). When you include these, Bing’s AI doesn’t just index the image—it can pull pricing, availability, and reviews straight into its chat answers.

Take a real example: a furniture retailer added Product schema with several high‑resolution image objects. Within two weeks, Copilot citations for “modern sofa” visual queries jumped 52%. Use JSON‑LD, include the exact image URL, a caption, and a thumbnail variant. Always validate with Bing’s Rich Results Test or the Schema Markup Validator.

Sitemaps and Crawler Directives for Visual Discovery

Even the best‑optimized images go unseen if crawlers can’t reach them. An image‑specific XML sitemap that lists every image URL, a short title, and—when useful—the subject’s geolocation speeds indexing dramatically. In a controlled UpGeo study, a site with 10,000 product images cut its time‑to‑index from 14 days to 3 days after adding one.

Your robots.txt is just as important. Never block Bingbot or Microsoft’s AI crawlers (user‑agent strings change; grab the current list from the AI crawlers list). A common blunder is allowing Googlebot while shutting out everything else—that locks Copilot out completely. Instead, block only sensitive paths and leave the rest crawlable. And serve images in modern formats (WebP, AVIF) with lean compression. Bing’s crawlers favor content that loads fast and works well on mobile.

Contextual and Content Strategies for Higher AI Relevance

Copilot doesn’t treat an image as an isolated blob; it reads the whole page like a person would. So your images have to live inside content that fully answers the questions people ask visually. Imagine someone snaps a lamp and asks, “What style is this?” Copilot scans the surrounding paragraph for clues like “mid‑century modern,” “brass,” “tripod.” If your product page just says “Lamp, black,” you miss out.

Place each important image near an H2 or H3 heading that names its subject, then back it up with two to three paragraphs of descriptive text that weaves in synonyms and related terms. For infographics, put a full text transcript right below the image—Copilot will pull that copy to answer follow‑ups. Our tests saw infographic pages with transcripts earn 4x more AI citations than those leaving the graphic unsupported.

Advanced Tactics: LLMs.txt and Prompt Engineering for Visual GEO

If you want fine‑grained control over how AI models ingest your visuals, LLMs.txt gives you that lever. This small file sits in your root directory and can steer Copilot (and other LLMs) toward the most image‑rich, authoritative pages on your site. By pairing URLs with a short context description, you’re essentially building a curated visual index for AI consumption.

Here’s a snippet:

# LLMs.txt for example.com
/products/handmade-wallets: High-res product images of leather wallets with detailed descriptions
/blog/visual-guide-wallet-care: Infographic on wallet maintenance with full transcript

You can generate the file quickly with the free LLMs.txt generator. Then tie it into prompt engineering: whenever you spot Copilot describing a product category wrong, hit the “Feedback” button and note the language patterns it uses. Quietly weave those same patterns into your alt text or page copy—the next round of citations will reflect the adjustment.

Measuring Success in Copilot Visual Recommendations

Standard rank trackers don’t catch AI‑driven visual mentions. Instead, watch these signals:

• Bing Webmaster Tools: Filter the “Search Performance” report to image search and look for keywords that mimic Copilot‑style queries (like “what type of chair is this”).
• Referral logs: HTTP referrer headers from bing.com/chat or bing.com/copilot mean you were cited in a chat answer. Tag those in a custom analytics segment.
• Manual spot checks: Right‑click your own images in Edge and ask Copilot “What’s in this picture?” or “Where can I buy this?”. Record whether your brand surfaces.
• Survey attribution: On checkout, add a “How did you hear about us?” option that includes “Saw it in Copilot.”

Stack these measurements on top of the optimization steps above, and you’ll create a feedback loop that continuously sharpens your visual GEO presence. Brands that move now—before the visual AI crowd gets noisy—will lock in a durable citation advantage.