Guide · Image SEO
Image SEO: How to Optimise Images for Google and Users
Five points decide good image SEO
Image SEO means: descriptive file names, meaningful alt texts, WebP or AVIF instead of JPEG, correct dimensions with srcset, and lazy loading for everything except the LCP image. AVIF often saves 40 to 50 % file size compared to JPEG. Since the EU AI Act, you additionally have to label AI-generated images. Per page, implementation rarely takes longer than 30 minutes.
File names: the first ranking factor everyone ignores
Google reads the file name before any algorithm analyses the image itself. IMG_20260806_1432.jpg says nothing. road-bike-carbon-frame-black.jpg says everything.
- Lowercase, hyphens, no umlauts. „Küchenmaschine“ becomes kuechenmaschine.
- 3 to 5 words that describe what’s visible – not what you want to rank for.
- Name before uploading. Renaming afterwards in the CMS often creates new URLs.
For shops with thousands of product images, a naming scheme like productname-variant-view.avif from the ERP is worth it. One hour of configuration instead of hundreds of manual steps.
Alt texts: accessibility first, then SEO
The alt text exists for people who can’t see the image – screen reader users, but also anyone with a dropped connection. SEO is the side effect. Whoever reverses the order writes alt texts like “images seo optimisation agency berlin” – useless for the blind, risky for rankings. In its own image search documentation, Google names the same order: first an honest description, then visibility.
- Describe what’s visible, in one sentence: “A mechanic trues a wheel in the assembly stand.”
- 80 to 125 characters are a good corridor.
- Decorative images get an empty alt attribute (alt=””), so screen readers skip them.
- No “image of…” – the software already knows it’s an image.
This is how file name, alt text and fixed dimensions interlock in the markup – purely as an example, without a real image loading here:
<img src="rennrad-carbon-rahmen-schwarz.avif"
alt="Black carbon road bike frame photographed from the side"
width="800" height="600" loading="lazy">WebP and AVIF: the format question is settled
JPEG and PNG are legacy for web use. According to Google’s own WebP study, WebP saves around 25 to 35 % file size compared to JPEG. According to web.dev, AVIF goes further: 40 to 50 % savings at the same perceived quality, available in all major browsers since 2024.
Recommendation for 2026: AVIF as standard, WebP as fallback via the picture element. PNG only for lossless graphics, SVG for logos and icons. Most CMS handle the conversion server-side. Important: test the quality level – AVIF at quality 50 to 60 usually looks better than JPEG at 80 and weighs half.
Dimensions and srcset: never deliver more pixels than needed
A 2,400-pixel-wide image in a 400-pixel container is the most common performance mistake in my audits. How much is at stake is shown by the Web Almanac 2024 by HTTP Archive: at 1,054 KB, images are the largest single item in the median page weight of a desktop homepage – ahead of JavaScript, fonts, CSS and HTML. In cases like the one described above, the browser loads a multiple of the data needed for the visible container.
The solution is srcset with sizes: the same image in several widths (e.g. 400, 800, 1200, 1600 pixels), the browser picks the right one. Modern CMS generate the variants automatically – just check that the sizes attribute matches the layout. This includes fixed width and height attributes in the HTML: they prevent layout shifts and keep your CLS value stable.
Lazy loading – but not for the LCP image
loading=”lazy” only loads images when they scroll into view. Mandatory for everything below the first viewport. The exception: the largest image in the first viewport – your LCP element. That needs the opposite: loading=”eager” plus fetchpriority=”high”. A lazy-loaded hero image regularly delays the LCP by 500 to 1,000 milliseconds.
Practical rule: First image eager, all others lazy. Many themes blanket-set everything to lazy – the second most common mistake in my audits. This too can be shown as pure example markup, without a real image loading here:
<img src="hero-rennrad-shop.avif"
alt="Road bike workshop with a view of the bike wall"
width="1600" height="900" loading="eager" fetchpriority="high">AI images: labelling obligation under the EU AI Act
Since August 2026, Article 50 of the EU AI Act applies fully: AI-generated or AI-manipulated images must be recognisable as such. Concretely: AI images need machine-readable labelling (many generators embed C2PA metadata – don’t strip it when compressing). For realistic-looking people, places or events, the labelling must additionally be visible, for example in the caption.
Stock images too: Check licence and personal rights. “Royalty-free” doesn’t mean rights-free – editorial licences often don’t allow commercial use, and depicted people need a model release. Document source, licence type and scope of use per image – a spreadsheet is enough.
Image sitemaps: only needed in special cases
Almost never needed. If images are embedded regularly in the HTML, Google finds them via the normal crawl. Only sensible if images are loaded via JavaScript – or if image search traffic is a real business goal, for example with recipe or product databases.
An image project in numbers
Numbers weigh more than claims. An example from an ongoing client project shows what the individual levers – format, srcset, lazy loading rule – deliver together in practice.
“For a bike parts shop, we converted around 12,000 product images to AVIF, fixed srcset and set the hero image to eager. The LCP dropped from 3.4 to 1.9 seconds in the CrUX report, and image impressions in the Search Console rose by about 30 percent within three months.”
— Viktor Pásztor, SEO freelancer
None of the levers is spectacular on its own. Taken together – format, right size, correct loading priority – they make the difference that shows up in the Core Web Vitals.
Common mistakes in image SEO
All images lazy-loaded
Even the LCP image gets lazy-loaded. That costs 0.5 to 1 second on the largest element in the first viewport – exactly where it counts most.
Keyword stuffing in the alt text
Instead of an honest description, there’s a keyword chain like “images seo optimisation agency berlin” – useless for screen reader users and risky for rankings.
Original size without srcset, missing dimensions
4,000-pixel photos land uncropped in 400-pixel containers, and without width and height attributes the layout jumps while loading – bad for the CLS value.
AI images and stock photos without documentation
AI-generated images without labelling have been a legal risk since 2026, and stock images often lack the documented licence including personal rights.
Frequently asked questions about image SEO
How long should an alt text be?
80 to 125 characters are a good corridor. One descriptive sentence is enough; screen readers read out the complete text.
Does image SEO bring measurable traffic?
Yes, depending on the industry. For visually driven topics, 5 to 15 % of impressions often come via image search.
WebP or AVIF – which do I take in 2026?
AVIF as standard, WebP as fallback. AVIF saves 40 to 50 % file size compared to JPEG.
Do I have to label every AI image?
Yes, machine-readable always (Art. 50 EU AI Act). Additionally visible if the image depicts real people, places or events deceptively realistically.
Viktor Pásztor
Viktor Pásztor is an SEO freelancer in Berlin, has been in digital marketing for over 15 years and manages around 15 client projects in parallel, mostly e-commerce and B2B. He has been working 100 % remotely for more than five years – with WordPress, Shopify, Shopware and TYPO3. More about Viktor Pásztor
Keep reading
Technical SEO
What a crawl finds and what happens next.
On-page optimisation
How content, structure and images interact.
SEO pricing
What working with me costs.
Images slowing down your site?
In a free initial call, we’ll look at your Core Web Vitals and the biggest image problem areas. Reply within 24 hours on business days.
→ Request an SEO audit now