What alt text actually does
Alt text serves three audiences, and the same text needs to work for all three. Screen reader users hear it read aloud and decide whether to skip or engage. Google's crawler uses it as a primary signal for what the image depicts and as a strong relevance signal for the page it sits on. Browsers display it as a fallback when the image fails to load. The same string serves all three — write it for the screen-reader user first and the other two benefits arrive for free.
There's a fourth, newer audience: large language models powering AI Overviews and chat answers. These models lean heavily on alt text when deciding which images to cite alongside generated answers. Pages with crisp, descriptive alt text are getting cited; pages with generic or missing alt text are not.
The four rules of writing alt text that ranks
Across thousands of pages we've audited, the same patterns hit and the same patterns miss. Apply these four rules to every image:
- Describe what's in the image, not why it's there. Bad: 'Click here for our pricing.' Good: 'EazyAITools pricing page showing four plan tiers with monthly and annual toggle.'
- Include the primary subject early. Bad: 'A close-up shot taken with a 50mm lens of a steaming cappuccino on a wooden table.' Good: 'Steaming cappuccino on a wooden table, shot close-up.'
- Match the page's intent without keyword-stuffing. If the page targets 'best wireless headphones 2026' and the image is a Sony WH-1000XM6, alt text 'Sony WH-1000XM6 wireless headphones, the top pick for 2026' is honest and relevant. 'Wireless headphones 2026 best top buy review headphones cheap' is stuffing — Google's spam classifiers down-rank it.
- Leave decorative images blank (alt=""). Spacer images, decorative dividers and pure styling images should have empty alt text so screen readers skip them. Wrong alt text on a decoration is worse than none.
File naming: the second-biggest image SEO lever
Google uses the image's file name as a ranking signal, especially when alt text is missing. 'IMG_0473.jpg' tells Google nothing; 'sony-wh-1000xm6-wireless-headphones.jpg' tells it the subject. Adopt a simple convention: lowercase, hyphens (never underscores or spaces), descriptive, no more than 5–6 words. Don't try to keyword-stuff the file name either — '-best-cheap-2026-review.jpg' looks spammy to humans and to algorithms.
For programmatically-generated images (product photos, user uploads, thumbnails), bake the descriptive name into the upload pipeline. A CMS that saves uploads as 'product-1234.jpg' loses ranking signal compared to one that saves them as 'sony-wh-1000xm6-front.jpg'.
Step-by-step: optimise an image for SEO before publishing
- 1Pick the right image. The image must visually answer the question the page targets. A page on 'how to remove backgrounds in Photoshop' needs a real screenshot of the Photoshop UI, not a stock photo of a person at a laptop.
- 2Rename the file. Use lowercase-with-hyphens, 3–6 descriptive words, no stop-words. Example: 'remove-background-photoshop-select-subject.jpg'.
- 3Resize and compress. Cap width at 1200px for body images, 1920px for hero images. Compress to WebP at quality 80 (or JPEG quality 75) — typically 80–150 KB.
- 4Write alt text in plain language. Describe what's in the image; include the primary keyword naturally if it fits.
- 5Add a caption beneath the image where it adds clarity. Captions get read more than body text and are a strong relevance signal.
- 6Place the image near the relevant body text. Google strongly associates an image with the paragraph immediately before or after it; misplaced images miss the relevance boost.
- 7Add width and height attributes in HTML to prevent layout shift, which Google now penalises directly via Core Web Vitals.
- 8If the image is the subject of the page (product photo, recipe photo), add ImageObject schema with caption, contentUrl, license and creator fields.
Image structured data (schema.org/ImageObject)
Most pages don't need image-specific schema, but two categories should always include it: product pages (image is part of Product schema) and how-to / recipe content (image is part of HowTo or Recipe schema). The schema lets Google show your image in rich results like the 'recipes' carousel or the product image pack, which take up significantly more SERP real estate than a regular blue-link result.
Beyond the structured-data formats, also include an open-graph image tag (og:image), a Twitter card image (twitter:image), and ensure the image URL is absolute (https://yourdomain.com/...) — relative URLs sometimes fail when LinkedIn or X scrape your page.
Compression, formats and Core Web Vitals
Image weight is the #1 cause of slow Largest Contentful Paint (LCP) scores, which directly impact ranking under Google's Core Web Vitals. The simple fix: serve modern formats (WebP, AVIF) with a JPEG fallback via <picture> elements; cap dimensions to what the layout actually uses (don't ship 4K hero images to a 600px-wide column); and lazy-load every image below the fold (loading="lazy").
AVIF wins on file size for photographs (often 50% smaller than JPEG at the same quality) but encoding is slow; WebP is the better default for most sites because every browser since 2020 supports it natively. Whichever format you pick, don't ship the same image twice — that's the most common mistake.
Alt text for AI Overviews and ChatGPT browsing
AI search surfaces — Google AI Overviews, Perplexity, ChatGPT browsing, Bing Copilot — increasingly cite images alongside generated answers. The selection signal these models use most heavily is alt text combined with surrounding paragraph context. Pages with descriptive, accurate alt text are showing up in citation panels; pages with empty or generic alt text are invisible to that surface.
Practical implication: write alt text as if you're describing the image to someone over the phone. "Screenshot of the Photoshop Select Subject tool with a portrait of a woman against a grey background, showing the marching-ants selection around her hair" is verbose but explicit — and exactly what an LLM needs to confidently cite the image in an answer about the Select Subject feature.
Auditing existing image SEO at scale
If your site has hundreds or thousands of pages, manually rewriting every alt text is impractical. The 80/20 audit: export your images with empty alt text or with alt text under 5 words from your CMS, sort by page traffic (most-trafficked pages first), and rewrite the top 10% — these are the images Google is most likely to index and cite. The remaining 90% can be re-written progressively or auto-generated via an AI vision model with human review.
Most major CMS platforms (WordPress, Webflow, Shopify) ship plugins or built-in features that flag images with missing alt text in their dashboards. Turn that warning on and you'll catch new uploads at the source, where fixing them costs seconds instead of an audit later.
Bonus: lazy-loading without breaking SEO
The native loading="lazy" attribute is the right way to defer below-the-fold images. Avoid older JavaScript-only lazy-loading libraries that swap a placeholder image for the real one in <img src> — Googlebot now executes JavaScript reliably but it still discovers images faster via the initial HTML payload than via a post-render swap. Native lazy-loading uses real <img src> attributes from the start, so the indexing pipeline sees every image immediately while the browser defers actual loading until needed.
FAQ
- How long should alt text be?
- Aim for 8–15 words. Screen readers read alt text aloud word-for-word, so very long alt text becomes tedious. Very short alt text ('headphones') misses the relevance signal Google looks for. Most well-optimised pages settle around 10–12 words per image.
- Should I include the keyword in every image's alt text?
- Only if it describes what's actually in the image. Forcing the keyword into alt text on a decorative image is keyword-stuffing and gets down-ranked. Forcing it on an image that genuinely shows the subject is fine and helpful.
- Does file name matter if I have great alt text?
- Yes, but less. File name is an independent ranking signal — Google uses both. Optimising both is straightforward and compounds. Optimising only alt text leaves 10–20% of image SEO upside on the table.
- WebP or AVIF in 2026?
- WebP for general-purpose use (universal browser support, fast encoding, ~30% smaller than JPEG). AVIF for hero images where every kilobyte matters (~50% smaller than JPEG but slow to encode). Serve both via <picture> with JPEG fallback for safety.
- Should I add captions to every image?
- No. Captions help on images where context isn't obvious from the surrounding text — diagrams, charts, photos with attribution. Decorative images and self-explanatory product shots don't need captions.