Nearly 9 out of 10 online stores ship images with no alt text and files 5 to 10 times heavier than they need to be. It's the most neglected SEO lever in 2026 — and one of the few that directly affects your Google Images traffic, your page speed, and now your visibility inside AI search engines.
This guide covers what actually matters: formats, compression, naming, alt text, image sitemaps — plus the angle most SEO guides still skip: how AI engines (Google Lens, AI Overviews, ChatGPT, Perplexity Shopping) "read" your images to build their answers.
What Image SEO Actually Means
Image SEO covers five distinct technical levers:
- Format and compression — file weight, which directly affects your LCP (Core Web Vitals)
- File name — a signal Google uses when the surrounding textual context is weak
- Alt text (
alt) — the description that stands in for the image for engines and screen readers - Technical delivery —
srcset, declared dimensions, lazy loading - Discoverability — image sitemaps,
ImageObject/Productstructured data
Each lever works independently. A perfectly named file with weak alt text loses half its value; a perfect alt text on a 3MB file still tanks your LCP.
Formats and Weight: the Duo That Decides Your Speed
WebP is the default standard in 2026: 25–35% lighter than JPEG at equal quality, with transparency support. AVIF goes further (up to 50% lighter than JPEG) but is slower to encode and less universally supported.
| Format | Weight gain vs JPEG | Use case |
|---|---|---|
| WebP | -25 to -35% | Default standard, all use cases |
| AVIF | -40 to -50% | High-impact hero images |
| PNG | baseline | Only when complex transparency is required |
| SVG | variable | Logos, icons, vector illustrations |
Target weight by use case:
- E-commerce product image: under 200KB
- Hero / lifestyle image: under 400KB
- Icons and thumbnails: under 30KB
Lossy compression at 80% quality reduces file size by 70–80% with no visible quality loss. Tools like Squoosh, TinyPNG, or an automated pipeline (Next.js Image, Cloudinary, imgix) make this effortless on every upload.
Naming and Describing Your Images: the Alt Text + Filename Checklist
Google uses the file name as a signal when textual context is missing. IMG_00234.jpg says nothing; long-floral-print-dress-blue-size-8.webp immediately describes the subject, color, and size.
Naming best practices:
- Lowercase, hyphens (not underscores)
- Descriptive, no keyword stuffing
- Maximum 60–80 characters
For alt text, the rule is simple: describe what the image shows as if you were describing it to someone over the phone, without forcing the keyword.
| Bad alt text | Good alt text |
|---|---|
"" (empty on an informative image) | — |
"product image" | too vague |
"blue dress summer dress cheap dress" | keyword stuffing |
"Long floral print dress in blue, flowing cut, summer 2026 collection" | descriptive and precise |
Purely decorative images (backgrounds, visual dividers) don't need alt text — leave it empty (alt="") rather than inventing a useless description.
Image Sitemaps, Lazy Loading, and Responsive Delivery
Three technical settings complete the picture:
srcsetandsizes— serve the right resolution for the device, instead of sending a desktop-sized image to mobile- Declared dimensions (
width/height) — prevents CLS (Cumulative Layout Shift) caused by images that push content around while loading - Lazy loading (
loading="lazy") — delays off-screen images, except for the hero image (LCP), which should stayeagerwithfetchpriority="high"
If your catalog spans several hundred images, a dedicated image sitemap (image:image tag in your XML sitemap) speeds up discovery and indexing in Google Images — especially useful for product pages sitting deep in your internal linking structure.
Structured Data: the Shortcut to Shoppable Images
Product markup with an image sub-property, or standalone ImageObject, explicitly tells Google and AI engines that an image is tied to a product, price, and availability. That's what makes an image eligible for enhanced shopping results. If you haven't structured your product data yet, our Schema.org and structured data guide covers the full JSON-LD implementation.
The GEO Angle: How AI Engines "See" (and Cite) Your Images
This is the part most SEO guides still skip: AI engines don't "look" at your images the way a human does. GPTBot, PerplexityBot, or Google Images crawlers rely first on the text surrounding the image — alt text, caption, filename, page context — to understand what it represents, before computer vision even comes into play.
Three signals matter specifically for GEO:
- Visual search is exploding. Google Lens hit 10 million monthly visits in 2025, up from 1 million two years earlier. Perplexity Shopping and ChatGPT rely on similar mechanisms to identify products from images.
- AI Overviews surface shoppable images. A well-tagged image (
Productschema + descriptive alt text + rich textual context) has a far better chance of appearing in these enhanced blocks than an isolated image with no metadata. - Multiple angles help GEO. Photographing a product from several angles multiplies the chances of being picked up by a visual search engine — each angle is another matching opportunity.
In practice: the combination of precise alt text, descriptive filename, and product schema is today's most actionable signal for an image to be understood — and cited — by an AI engine, well before pure image recognition becomes reliable at scale on general content.
A Concrete Example: What a Full Reset Changes
On a typical audit of a fashion e-commerce store (around 400 product pages), here's what changed before/after a full image SEO reset (formats, compression, alt text, schema):
| Metric | Before | After (6 weeks) |
|---|---|---|
| Average image weight | 1.4MB (JPEG) | 140KB (WebP) |
| Pages with alt text filled in | 8% | 100% |
| Mobile LCP | 4.8s | 2.1s |
| Google Images traffic | baseline | +61% |
| Pages appearing in enhanced shopping results | 3 | 22 |
Not a single line of text content changed on these pages — only the image layer. It's the kind of gain most sites leave on the table because image SEO is treated as a minor technical detail.
Key Takeaways
- WebP (or AVIF for hero images) and 80%-quality compression are non-negotiable in 2026
- Descriptive alt text, without keyword stuffing, remains the most reliable text signal for Google and for AI engines
- Name your files descriptively: it's a backup signal when context is missing
Product/ImageObjectschema is the shortcut to enhanced shopping results and AI citation- An image audit can unlock a meaningful traffic gain without touching a single line of text content
FAQ
Does every image on a site need alt text?
No. Purely decorative images (visual dividers, backgrounds) don't need a description — leave the attribute empty. Focus on images that carry information: products, illustrations, charts, editorial photos.
WebP or AVIF: which one in 2026?
WebP remains the safest default (universal support, fast encoding). Reserve AVIF for high-impact visuals (hero images, banners) where the extra weight savings justify slower encoding.
Does the file name really affect rankings?
Yes, but as a secondary signal: Google mostly relies on it when the surrounding textual context is weak or missing. On a product page already rich in text, the impact is marginal; on a thin-content page, it can make the difference.
Do AI engines actually analyze my images visually?
Increasingly, yes — but the associated text (alt, caption, filename, context) remains today the primary and most reliable signal for an image to be correctly understood and retained in a generated answer.
To see where your own site stands across these five axes (image, technical, content, GEO, trust), run your free SEO + GEO audit — or grab the full PDF report with a prioritized action plan. If you want to understand why GEO changes the game first, start with our GEO guide.