TL;DR

Google Images drives roughly 22% of all web searches, and Google Lens now processes over 20 billion visual queries per month. Yet only 26% of websites use alt text correctly, and the majority of pages still serve images larger than necessary. Fix your formats (AVIF/WebP), write descriptive alt text, add structured data, and never lazy-load your LCP image. Those four moves cover most of the gap.

Published June 21, 2026 · By Guru Editorial

Image SEO has always been the underdog tactic that practitioners neglect until they audit a slow page and realize half their payload is a 4MB JPEG someone uploaded from their phone. In 2026 the stakes are higher, because images now surface through three distinct channels: traditional Google Image Search, Google Lens visual queries, and AI-generated answers that pull visual context from your pages. Miss any one of them and you leave real traffic on the table.

Google Lens alone processed over 20 billion visual search queries per month as of early 2026, continuing a growth trajectory that saw it triple its volume from roughly 3 billion monthly queries in 2021 to more than 10 billion by early 2023. Meanwhile, the shift toward AI Overviews has compressed organic click-through rates: Ahrefs reported in December 2025 that position-1 CTR drops 58% when an AI Overview appears. That means every incremental traffic channel, including image and visual search, carries more weight than it did two years ago.

This guide covers the complete image optimization stack: file formats, compression, naming, alt text, structured data, sitemaps, page-speed signals, and what actually changes when you optimize for AI visual search rather than classic image search.

1. Choose the Right Image Format Before You Do Anything Else

Format selection is the highest-leverage decision you make for image performance. Most teams default to JPEG or PNG out of habit. Both are outdated choices for most use cases in 2026.

AVIF produces files roughly 50% smaller than equivalent-quality JPEGs and around 20-25% smaller than WebP, thanks to its AV1-based codec. Browser support hit 94.7% globally as of mid-2026, which makes it viable for production without aggressive fallback logic.

WebP remains a solid default with ~99% browser support and 25-35% file size reductions versus JPEG. If your image CDN does not yet support AVIF transcoding, WebP is the correct fallback.

PNG belongs on images with transparency where quality loss is unacceptable. JPEG is still appropriate for legacy pipelines and email, but should not be your web-delivery format in 2026.

The recommended implementation pattern: serve AVIF to browsers that support it, WebP to those that do not, with JPEG or PNG as a final fallback inside a <picture> element. Reducing image file sizes by 40-50% through next-gen format adoption typically improves Largest Contentful Paint (LCP) scores by 20-30%, often moving pages from "needs improvement" to "good" thresholds (Hashmeta, 2025).

Format Decision Reference

FormatTypical Size vs JPEGBrowser Support (2026)Best Use Case
AVIF~50% smaller94.7%Hero images, product photos, editorial content
WebP25-35% smaller~99%General fallback, wide compatibility required
PNGSame or larger100%Logos, icons, transparency-heavy graphics
JPEGBaseline100%Legacy systems, email only
SVGVariable100%Icons, logos, diagrams, inline graphics

2. Compress Without Destroying Quality

Choosing AVIF or WebP does not eliminate the need for compression. Format conversion alone does not guarantee a small file if the encoder quality setting is maxed out.

Target a perceptual quality score between 75-85 for most editorial and product images. Tools like Squoosh (browser-based), Sharp (Node.js), or Cloudinary's auto-quality parameter handle this programmatically. For ecommerce catalogs running thousands of SKUs, automation is the only scalable path: any image that comes into your CMS should be auto-transcoded at upload time, not manually touched before publishing.

A useful mental model: the file size limit for an image that is not the LCP element is whatever keeps your total page weight below 1MB for mobile. The LCP image specifically should load inside the 2.5-second threshold that Google's good/needs-improvement/poor scoring uses.

Avoid double-compressing. Running a lossy compressor on an already-lossy JPEG and then re-saving it degrades quality without a proportional size benefit. Always compress from the original raw or lossless source file.

3. Name Files and Structure URLs for Crawlers

Image file names are a confirmed Google ranking signal. Google uses the filename to understand image content, especially when alt text or surrounding context is sparse. A file named IMG_4821.jpg tells crawlers nothing. A file named titanium-road-bike-frame-side-view.avif tells them exactly what to index.

File naming conventions that work:

  • Use lowercase letters only
  • Separate words with hyphens, not underscores or spaces
  • Lead with the primary subject, then qualifying detail (product-name, color, angle)
  • Keep names under 60 characters
  • Match the filename vocabulary to the page's target keyword cluster where natural

For URL structure, keep images hosted on the same domain as your content when possible. CDN hosting is fine, but cross-domain images require you to verify both domains in Google Search Console and include the CDN domain in your image sitemap. Google's image sitemap format explicitly permits cross-domain <image:loc> entries, unlike the restrictions on regular sitemaps.

4. Write Alt Text That Serves Both Humans and Crawlers

Alt text is the single most important metadata signal for image indexing. Google uses alt text alongside its computer-vision algorithms and the surrounding page text to understand what an image shows. Only 26% of websites use alt text effectively, which makes this one of the easiest competitive advantages available (Amra and Elma, 2026).

Alt text is an image ranking factor, not a general on-page ranking factor. It can help your images appear higher in Google Images results and increase the likelihood that your page image gets pulled into AI Overviews or visual search surfaces. It does not directly boost your page's position in standard blue-link results.

What good alt text looks like:

  • Describes the image as you would to someone on a phone call
  • Stays under 125 characters (screen-reader and crawler convention)
  • Integrates a relevant keyword naturally, not as a forced suffix
  • Omits "image of" or "photo of" as a prefix (Google ignores it)
  • For decorative images (separators, abstract backgrounds), use alt="" to signal explicitly that the image carries no semantic meaning

What keyword stuffing in alt text looks like: alt="SEO, search engine optimization, best SEO, image SEO, SEO tips, image optimization SEO". Google's guidelines flag this as a spam signal.

For ecommerce product images, include the product name, key differentiating attribute (color, size, material), and the view angle. For editorial images, describe what is happening in the scene with enough specificity that a blind reader understands the visual context.

You can audit your existing alt text gaps at scale using Guru's on-page analysis tools, which flag missing, duplicate, and over-long alt attributes across your full URL inventory.

5. Handle LCP Images Correctly

Images are the LCP element on the majority of pages. Google's Core Web Vitals 2025 standards mandate LCP under 2.5 seconds for a "good" rating, and the ranking impact is real: sites that pass all Core Web Vitals thresholds tend to rank measurably higher than peers that do not, with some analyses placing the average advantage at multiple ranking positions (BKNDdevelopment, 2025).

The most common and costly image-specific LCP mistake is applying loading="lazy" to the hero or above-the-fold image. Lazy-loaded LCP images hit a 75th-percentile load time of 720ms, nearly double the 364ms achieved by preloaded images. The 2025 Web Almanac reported that 16% of pages still lazy-load their LCP image.

The correct approach for LCP images:

  1. Set loading="eager" (or omit the loading attribute entirely, since eager is the default)
  2. Add fetchpriority="high" to the <img> tag to signal render-critical priority to the browser
  3. Add a <link rel="preload" as="image"> tag in the <head> pointing to the LCP image
  4. For responsive images, add imagesrcset and imagesizes attributes to the preload link

For images below the fold, loading="lazy" is correct and recommended. The rule is simple: lazy-load everything the user cannot see on initial load, and eager-load everything they can.

Cumulative Layout Shift (CLS) is a secondary image-related Core Web Vitals issue. Always declare explicit width and height attributes on <img> elements so the browser reserves the correct space in the layout before the image loads. Undeclared dimensions are the most common cause of image-driven CLS.

Pair image performance fixes with the broader technical checklist in Guru's 40-point technical SEO audit for a complete picture of speed and crawlability signals.

6. Add Structured Data and Image Sitemaps

Schema markup and image sitemaps serve different purposes, but both increase the probability that your images get discovered, indexed, and surfaced in AI-generated answers.

Structured Data for Images

Google's documentation recommends embedding image information inside the structured data types that already apply to your content: Article, Product, Recipe, VideoObject, and LocalBusiness all support image properties in JSON-LD. Structured data with schema markup has a 2.5x higher chance of appearing in AI-generated answers, according to analysis of Google's Gemini-powered AI Mode (Stackmatix, 2026).

Key schema properties to include for every image you care about:

  • image (URL to the highest-resolution version)
  • name (descriptive label)
  • description (what the image depicts)
  • width and height in pixels
  • contentUrl for the actual file, url for the canonical landing page

Note: HowTo rich results were removed from Google SERPs in 2023, and FAQ rich results were removed on May 7, 2026. Neither schema type generates a SERP feature anymore, but both remain useful because they help AI engines parse your content structure during answer synthesis. The same applies to ImageObject markup: no dedicated SERP feature, meaningful signal for AI extraction.

For a thorough breakdown of which schema types still produce rich results in 2026, see Guru's schema markup guide.

Image Sitemaps

An image sitemap tells Google about images it might not discover through crawling alone, including images loaded via JavaScript or hosted on a CDN subdomain. You can add image tags to your existing XML sitemap or maintain a separate image-sitemap.xml.

Each <url> entry supports up to 1,000 <image:image> child elements. At minimum, include <image:loc> (the image URL) and <image:title> for each image. Submit the sitemap in Google Search Console and verify any CDN domains listed in <image:loc>.

Google supports indexing BMP, GIF, JPEG, PNG, WebP, SVG, and AVIF formats through image sitemaps. If you serve AVIF files, they will be indexed normally, so there is no need to list JPEG fallbacks in the sitemap.

Traditional image search and AI visual search require overlapping but not identical optimizations. Google Lens uses computer vision to match images against its index, which means the visual content of the image itself, not just the surrounding metadata, influences how it gets matched and retrieved.

For Google Lens to match your product or editorial images accurately:

  • Resolution: Minimum 1,200px on the longest dimension for editorial images. For primary ecommerce product images, 2,400px or higher allows Lens to resolve product detail features.
  • Subject clarity: The subject should occupy the dominant portion of the frame. Cluttered backgrounds reduce matching confidence.
  • Multiple angles: For products, include front, side, back, and detail shots. Each angle is indexed separately and expands the surface area for visual matches.
  • Consistent lighting: Even, diffused lighting makes feature extraction more reliable than dramatic shadows or mixed color temperatures.

Google Lens integrates directly with Google Shopping Ads in 2025-2026, meaning ecommerce product images that are well-optimized and paired with Product schema can appear in Shopping surfaces triggered by camera-based queries. Visual search adoption in ecommerce is accelerating, with market analysts tracking significant year-over-year growth in visual-search-driven conversions across major platforms.

For AI Overviews and AI Mode specifically: Ahrefs data shows AI Mode and AI Overviews share only about 13.7% of the URLs they cite. The implication for image SEO is that you cannot assume a well-ranked page automatically surfaces its images in AI-generated answers. Pages need their images embedded inside well-structured content, supported by schema, with descriptive surrounding text, to give AI crawlers the multimodal context they need to pull the image into a generated response.

For a broader treatment of GEO optimization signals, see the Guru GEO module and our guide on optimizing for Google and AI answer engines simultaneously.

Image SEO Signal Stack: From File to SERP 1. File Layer Format (AVIF/WebP) · Compression (75-85 quality) · Filename (hyphenated, descriptive) 2. HTML Layer Alt text · width/height · fetchpriority="high" · loading="lazy" (below fold only) 3. Schema Layer ImageObject · Article/Product image property · width, height, contentUrl 4. Discovery Layer Image sitemap · GSC submission · CDN domain verification Each layer multiplies the signal strength of the layer above it

The image SEO signal stack: each layer reinforces the one above it. Skipping the schema layer leaves AI visual extraction with weaker context.

8. Image SEO Checklist

Use this before any page goes live and as a recurring audit item for existing content.

Pre-Publish Checklist

  • [ ] Image exported in AVIF (with WebP fallback) or WebP minimum
  • [ ] File compressed to 75-85 perceptual quality from lossless source
  • [ ] Filename is lowercase, hyphen-separated, and describes the subject
  • [ ] Alt text written (under 125 characters, descriptive, no keyword stuffing)
  • [ ] Decorative images have alt=""
  • [ ] width and height attributes declared on every <img> element
  • [ ] LCP image has loading="eager" and fetchpriority="high"
  • [ ] LCP image has <link rel="preload"> in <head>
  • [ ] Below-fold images have loading="lazy"
  • [ ] Product/editorial images included in ImageObject or parent schema type
  • [ ] Image URL included in image sitemap
  • [ ] Image sitemap submitted in Google Search Console
  • [ ] CDN domains verified in Search Console (if applicable)
  • [ ] Product images meet 1,200px minimum (2,400px for primary ecommerce shots)
  • [ ] LCP score verified in PageSpeed Insights after publishing

9. Common Image SEO Mistakes and How to Fix Them

Understanding where teams go wrong is as useful as knowing what to do right. These are the most frequent failures encountered in production audits.

Serving the same image to all devices. A 2,400px product image served to a 375px-wide mobile viewport transfers 4-6x more data than necessary. Use srcset and sizes attributes to serve appropriately dimensioned images per viewport.

Lazy-loading the LCP image. As covered above, this is the single most impactful performance mistake specific to images. Audit your hero and above-fold images first.

Ignoring image file size on paginated or category pages. Category pages often load 24-48 product thumbnails. If each thumbnail is 200KB rather than 20KB, you have added nearly 5MB to a single page load.

Using generic stock photos without contextual alt text. A stock photo of "business people in a meeting room" adds no semantic signal. Write alt text that connects the image to the page's specific topic.

Not auditing images after CMS migrations. Platform migrations often strip alt text, change filenames, or break image paths. Always run a dedicated image audit post-migration as part of the standard technical checklist covered in Guru's on-page SEO factors guide.

Hosting all images through a slow or uncached CDN. Image delivery time is part of LCP. If your CDN does not serve images from an edge node close to the user, the physical distance alone adds latency. Verify CDN edge coverage for your primary user geographies.

File Size Reduction vs. JPEG Baseline (same perceptual quality) % Size Reduction 50% AVIF 30% WebP ~0% PNG Baseline JPEG 0% 20% 40% 60% AVIF and WebP deliver meaningful file-size savings; the LCP impact follows directly from smaller payloads.

AVIF cuts file size by approximately 50% versus JPEG at equivalent quality; WebP lands around 30%. Both improve LCP, Core Web Vitals scores, and Google ranking signals.

Frequently Asked Questions

Does image alt text directly improve my page's ranking in standard Google Search?

Alt text is an image-specific ranking factor, not a general page ranking factor. It helps your images rank higher in Google Images and appear in AI-generated visual answers. It does not directly move your page's position in standard blue-link results, though it contributes to overall on-page quality signals.

Should I use AVIF or WebP in 2026?

Use AVIF as your primary format, with WebP as a fallback inside a <picture> element. AVIF produces roughly 50% smaller files than JPEG and 20-25% smaller than WebP at equivalent quality. Browser support for AVIF reached 94.7% in 2026, making the AVIF-first approach viable for most production sites.

How does Google Lens decide which images to surface?

Google Lens uses computer vision to match visual features in an image against its index, combined with metadata from alt text, filenames, surrounding text, and structured data. High-resolution images with clear subjects, no visual clutter, and product schema are most likely to rank in Lens-triggered results.

What is the biggest image-related Core Web Vitals mistake?

Applying loading="lazy" to the LCP image is the most impactful single mistake. Lazy-loaded LCP images take nearly twice as long to render as preloaded ones. Always use fetchpriority="high" and a <link rel="preload"> on your above-fold hero image.

Do image sitemaps still matter in 2026?

Yes, particularly for images loaded via JavaScript or hosted on CDN subdomains that Googlebot might not discover through standard crawling. Submit your image sitemap in Google Search Console and verify every domain listed in <image:loc> entries.

Is HowTo or FAQ schema worth adding for image-heavy pages?

Neither HowTo nor FAQ schema generates SERP rich results in 2026 (HowTo was removed in 2023; FAQ rich results were removed May 7, 2026). However, both schema types still help AI engines parse page structure during answer synthesis and are worth implementing for GEO purposes.

How many images should I include per page for SEO purposes?

There is no universal optimal number. Include images that serve the reader's understanding of the content. Each relevant image is an incremental opportunity to rank in image search and be pulled into AI-generated answers. The constraint is page weight: every image adds to load time, so optimize aggressively and only include images that add informational value.

What image dimensions should ecommerce product pages target?

For primary product images, target 2,400px or greater on the longest dimension to support Google Lens feature extraction and zoom functionality. Thumbnails displayed in category grids should be generated at display size (typically 300-600px), not scaled down in the browser from a larger source file.

Sources