Optimizing Website Images for SEO in 2026

Edited on Jul 3, 2026.

The image is now a citation surface. ChatGPT reads it. Gemini reads it. Claude reads it. Google Lens has been reading it since 2017. Image SEO in 2026 is no longer about ranking in Google Images alone — it is about being retrievable, attributable, and quotable by every engine buyers now use to answer their questions.

This is the full 2026 playbook. Nine disciplines, one FAQ, and the schema, formats, and file structures that make an image win in retrieval.

1. Alt text — the AI reading surface

Alt text was originally an accessibility attribute. It is now the primary text signal AI engines use to understand what an image contains. When an LLM ingests a page, alt text is one of the highest-weight inputs it uses to decide what the image is about, whether it belongs in an answer, and what caption to generate around it.

Four rules:

Describe what is in the image, concretely. "Woman drinking coffee at a wooden desk" beats "morning routine." AI engines reward specificity because specificity is what they retrieve.

Include the entity when the entity is the point. If the image is of a product, name the product. If it is a person, name the person if the page is about them. Named entities are how retrieval works.

Keep it under 125 characters when possible. Screen readers truncate longer strings, and engines down-weight bloated alt strings.

Do not keyword-stuff. Google's image documentation is explicit about this, and LLMs treat overloaded alt strings as spam.

Alt text is not decoration. It is a retrieval signal. For the platform-specific how-to on social alt text, see EPR's satellite guide: Creating Alt Text for Social Media Platforms.

2. File names — the second text signal

Image file names are read by both search engines and LLMs. A file named IMG_4472.jpg teaches an engine nothing. A file named chatgpt-citation-share-chart-2026.jpg teaches it the topic, the entity, and the year.

Use lowercase, hyphens, no underscores, no spaces, no special characters. Keep it descriptive but under 60 characters. Match the file name to the alt text theme — engines cross-reference the two.

3. ImageObject schema — the attribution layer

Schema.org's ImageObject type is the structural signal Google and other engines use to understand an image's ownership, license, and provenance. Publishers who mark up their images with ImageObject get better attribution in search results and better citation treatment inside AI engines.

Minimum viable ImageObject markup:

{
  "@context": "https://schema.org",
  "@type": "ImageObject",
  "contentUrl": "https://example.com/image.jpg",
  "creator": { "@type": "Person", "name": "Photographer Name" },
  "copyrightHolder": { "@type": "Organization", "name": "Publisher Name" },
  "license": "https://example.com/license",
  "acquireLicensePage": "https://example.com/license-request",
  "creditText": "Photographer Name / Publisher Name",
  "caption": "Descriptive caption here."
}

For the history of why this matters — and how it emerged from the Getty Images / Google settlement — see EPR's history of Google Images redesigns and controversies.

4. WebP and AVIF — the format layer

JPEG and PNG are legacy. WebP and AVIF are the current standards.

WebP is supported by every modern browser and delivers roughly 25–35% smaller file sizes than JPEG at equivalent quality.

AVIF is supported in Chrome, Firefox, Safari, and Edge and delivers roughly 50% smaller file sizes than JPEG at equivalent quality. Adoption is now high enough to use AVIF as the primary format with WebP fallback.

The <picture> element handles fallbacks cleanly:

<picture>
  <source srcset="image.avif" type="image/avif">
  <source srcset="image.webp" type="image/webp">
  <img src="image.jpg" alt="Descriptive alt text">
</picture>

Format alone can cut page weight by 30–50%. Which brings us to LCP.

5. LCP — the image is almost always the Largest Contentful Paint

Largest Contentful Paint is the Core Web Vital that measures how quickly the largest visible element on a page renders. On roughly 70% of pages across the web, that element is an image — usually a hero image above the fold.

Google uses LCP as a ranking signal. LLM crawlers time out on slow pages. The image is the bottleneck.

Five disciplines:

Serve the right size. Use srcset and sizes so mobile users do not download desktop-resolution images.

Preload the LCP image. A <link rel="preload"> in the head tells the browser to fetch the hero image immediately.

Use fetchpriority="high" on the LCP image tag.

Do not lazy-load the LCP image. Lazy-loading below-the-fold images is standard. Lazy-loading the hero image kills the score.

Compress aggressively. Target under 200KB for hero images. Use AVIF where possible.

6. Image sitemaps — the crawlability layer

Google supports image sitemaps as an XML extension. They are the fastest way to signal to Googlebot which images matter on the site — especially for images loaded via JavaScript that Googlebot may not discover on its own.

Image sitemaps are optional but material for sites with large image libraries — ecommerce catalogs, editorial photography, research report galleries. Most modern CMSes generate them automatically.

7. Lazy loading — everywhere except above the fold

Native lazy loading is now supported in every browser via the loading="lazy" attribute. Every image below the fold should use it. It defers the network request until the user scrolls, which reduces initial page weight and improves LCP for the images that actually render first.

Do not lazy-load the LCP image. Do not lazy-load images critical to the initial render. Everywhere else, lazy-load.

8. Accessibility as a ranking signal

Google has been explicit for years that accessibility affects rankings. Screen-reader-friendly alt text, semantic <figure> and <figcaption> elements, and adequate color contrast are all part of the image SEO stack.

Use <figure> and <figcaption> when the caption is meaningful. Use empty alt="" for purely decorative images so screen readers skip them cleanly. Ensure text overlaid on images meets WCAG contrast ratios.

9. Original imagery as competitive moat

The publishers and brands that compound in image search and AI retrieval are the ones that produce original imagery — original photography, original illustration, original data visualization. Stock photography is undifferentiated by definition. AI engines have been trained on the same stock libraries every competitor uses. The image that wins retrieval is the one no one else has.

For editorial and marketing operators, this is the single largest strategic decision in image SEO: invest in a stock library and lose, or invest in original visuals and compound. For the commercial case — TikTok Shop, visual search, livestream commerce — see EPR's Visual Revenue hub. For the behavioral case on why images now carry more marketing weight than copy, see Marketing Through Photos.

Adjacent EPR resources

Google Images: The Redesign History and the Controversies · Creating Alt Text for Social Media Platforms · Visual Revenue: How Images Drive Sales · Marketing Through Photos · Technical SEO and On-Page Hygiene: The 2026 Foundation

Reported by the Everything-PR Editorial Team.