Google Lens now handles roughly 20 billion visual searches a month, with about 4 billion of those tied to shopping. Pinterest Lens processes 1.5 billion queries a month, up 40% year over year. People aren't just typing 'blue running shoes' anymore, they're pointing a phone at a pair they saw on the street.
This lesson covers two things that work together: the image SEO basics that still drive Google Images traffic, and what visual search engines actually need from your images to find and understand them.
Quick Summary
- Alt text, descriptive filenames, and compression are still the highest-leverage image SEO wins, roughly 80% of the traffic lift comes from getting these three right.
- WebP is the 2026 baseline format; AVIF compresses further but needs a fallback for older browsers.
- Visual search engines (Google Lens, Pinterest Lens) match on objects and shapes, not filenames, so product angle, background, and image quality matter more than keywords.
- 60% of Gen Z say they prefer visual search over typing a query, this is a discovery channel, not a novelty.
- Structured data (ImageObject, Product) helps Google attach licensing, pricing, and authorship to your images in results.
Image SEO Fundamentals That Still Matter
Before visual search, get the boring stuff right. It's still most of the traffic.
Alt text describes function, not keywords. Write 5-15 words that explain what the image shows and why it's there. 'Woman in red jacket hiking a rocky trail at sunset' beats 'hiking jacket red womens outdoor gear' every time. Google penalizes keyword stuffing in alt attributes as a bad user experience signal, and screen reader users are reading this text too.
For purely decorative images (a background texture, a divider), use alt="". An empty alt tells screen readers to skip it, which is more helpful than a forced, meaningless description.
Filenames are a small but real ranking signal. IMG_0238.jpg tells Google nothing. blue-suede-loafers-side-view.jpg tells it exactly what's in the frame before it even opens the file. Rename every image before upload, use hyphens between words, and skip filler words like 'photo' or 'image'.
Compression and format choice affect Core Web Vitals directly. WebP is now the safe default, about 25-35% smaller than JPEG at equivalent quality, with full transparency support to replace PNG. AVIF pushes further, often 50% smaller than JPEG, but encoding is CPU-heavy and a few older Safari versions still choke on it. Serve AVIF inside a <picture> element with WebP and JPEG fallbacks so nothing breaks.
Heavy, unoptimized images are one of the most common Largest Contentful Paint (LCP) killers, and LCP is a ranking factor. An image that looks identical but loads in 200ms instead of 2 seconds is a win nobody sees, except the algorithm.
Format Comparison: JPEG vs. WebP vs. AVIF
Pinterest's engineering team, rebuilding its web app's loading pipeline (including image weight) around performance Slow perceived load times were costing Pinterest signups and search visibility, page weight, much of it images, was the biggest lever available Rebuilt the page-loading experience to cut perceived wait time, trimming image and asset weight as part of a broader performance overhaul
Result: Cut perceived wait time by 40%, which came with a 15% increase in SEO traffic and a 15% lift in signup conversion rate (measured after the rebuild shipped).
SourceYou don't need to memorize codec specs, you need to know which format to reach for by default. Here's the tradeoff in one table:
| Format | Size vs. JPEG | Transparency | Browser support (2026) | Best use |
|---|---|---|---|---|
| JPEG | Baseline | No | Universal | Fallback only, not a first choice anymore |
| WebP | 25-35% smaller | Yes | Universal in modern browsers | Default for everything |
| AVIF | Up to 50% smaller | Yes | Most modern browsers; a few older Safari versions fail | Hero/LCP images, always with a fallback |
The practical rule: serve AVIF first inside a <picture> element, WebP second, JPEG last as the safety net. Browsers pick the first format they support, so nothing ever breaks for an older visitor, it just costs them a slightly bigger download.
How Visual Search Actually Works
Forever 21's iOS app, via a visual search feature called 'Discover Your Style' built with Donde Search Shoppers browsing dresses and tops wanted to search by look (a neckline, a fit, a color) rather than type keywords that don't describe what they're picturing Let customers tap icons describing an item's attributes, silhouette, neckline, fit, to visually filter the catalog instead of typing a text query
Result: Average purchase value in the two test categories rose 20% within one month of launch (one month post-launch).
SourceVisual search doesn't read your page. It reads the pixels.
When someone points Google Lens or Pinterest Lens at an object, the system runs computer vision models that identify shapes, colors, textures, and patterns, then matches them against a visual index of images it's already crawled. Your alt text and filename help Google decide whether to index the image at all, but they don't influence whether the visual match succeeds. That's determined by the image itself.
This means a few things change in priority:
Image quality and framing become ranking factors in their own right. A cluttered background, awkward crop, or heavy filter confuses the matching model. Clean, well-lit, single-subject product shots on a neutral background match far more reliably than lifestyle shots with multiple objects competing for attention.
Multiple angles multiply your match surface. A shopper's photo could be a front view, a side view, or a close-up of a logo. If your catalog only has one hero shot, you only match one kind of query. Retailers seeing 25-40% higher conversion from visual search traffic are typically the ones with 4-6 angles per product, not one.
Context on the page still counts for shopping results. Google and Pinterest pull surrounding text, price, and availability to turn a visual match into a shoppable result. An image with no product name, price, or description nearby can be visually matched but never surfaced as buyable.
One in three online shoppers now uses an AI-powered discovery tool, chatbot, visual search, or conversational product finder, somewhere in their purchase journey. Adoption skews young: about 22% of 16-34 year-olds use visual search to find or buy something, versus 17% of 35-54 year-olds and just 5% of the 55+ group. If your audience is younger and your catalog is product-heavy, this is no longer optional groundwork.
Practical Image Optimization Checklist
Run every new image (and a batch of your top pages' existing images) through this list:
- Rename the file with descriptive, hyphenated words before upload, no camera IDs or generic terms.
- Write alt text that describes the image's content and purpose in under 15 words; use
alt=""for decorative images only. - Convert to WebP, with AVIF plus a JPEG fallback for image-heavy product pages where every kilobyte counts.
- Compress without visible quality loss, aim for the smallest file size where a side-by-side comparison still looks sharp.
- Add
widthandheightattributes (or CSS aspect-ratio) so the browser reserves space and avoids layout shift. - Submit an image sitemap for catalog or CDN-hosted images that Google's normal crawl might miss.
- Add ImageObject or Product structured data with
contentUrl,name,description, andlicensewhere relevant, this is what lets Google attach pricing, authorship, and licensing badges in image results.
Visual search rewards clean, well-lit, single-subject photography the same way voice search rewards clear pronunciation. You can't caption your way out of a blurry, cluttered product shot, fix the photo first, then optimize the metadata around it.
Worked Example: One Product Photo, Start to Finish
Here's the checklist above applied to a single real file, a pair of loafers a photographer just handed off.
Before: IMG_4821.HEIC, 4.2MB, shot on a cluttered desk with a phone charger visible in the background, no alt text, not on any sitemap.
- Crop and reshoot the background. Move the shoe to a neutral surface, single subject, side angle. This step alone determines whether the visual match model can lock onto the product at all.
- Rename the file.
IMG_4821.HEICbecomesblue-suede-loafers-side-view.jpg. - Export three formats. AVIF (~58KB), WebP (~82KB), JPEG fallback (~145KB), all served through one
<picture>element. - Write the alt text.
"Blue suede loafers, side view, tan rubber sole", 8 words, describes what's in frame and nothing else. - Add
width/height. Locks the layout so the image doesn't cause a late shift as it loads. - Add ImageObject structured data.
contentUrl,name: "Blue Suede Loafers",licensepointing to the brand's usage terms. - Add it to the image sitemap. So Google's crawl picks it up even if the product page itself is buried deep in the catalog.
After: a 145KB JPEG shrinks to a 58KB AVIF (a 96% file-size cut from the original HEIC), the file has a real name and alt text a screen reader can use, and the image is now eligible for both a Google Images click and a Lens visual match, where the original cluttered, unnamed photo was eligible for neither.
Key Takeaways
- Alt text, filenames, and compression remain the foundation, do these before anything else.
- WebP is the baseline format in 2026; pair AVIF with fallbacks rather than shipping it alone.
- Visual search matches pixels, not metadata, invest in clean multi-angle photography for anything you want discovered by camera.
- Structured data turns a visual match into a shoppable result by attaching price, name, and licensing data.
- This channel already reaches roughly a third of online shoppers and skews younger every year, treat it as core SEO work, not an experiment.