SEO for design agency websites: what we found on 81 agency sites

The short answer

If you fix one thing on an agency site, link each case study to the service it demonstrates, in the text of the case study. A link in the menu does not count, because it appears on every page and says nothing about this project. We found that link in the content of only 41% of the case-study pages we read. It is a five-minute fix per page.

How we checked

On 12 October 2026 we searched for "web design agency" in 12 cities: London, Manchester, Bristol and Leeds in the UK, New York, Austin, Chicago, Los Angeles and Denver in the US, and Sydney, Melbourne and Brisbane in Australia. We took the agency websites from the results, leaving out directories, review sites and articles, which gave 89 agencies. Eight blocked our checker, leaving 81: 31 in the US, 28 in the UK and 22 in Australia.

For each site we read the homepage, robots.txt, sitemap and a missing page, then found its portfolio or case-study pages and read up to five, chosen at random. That came to 286 case-study pages from 59 agencies. Every check used public pages only, with a pause between requests.

One correction to our own method, because it changes the headline. Our first pass counted any link from a case study to a service page, and found one on 98% of pages. Almost all of those were in the navigation menu, which is on every page. When we counted only links in the content of the page, the figure fell to 41%. That second number is the one that matters, and the one we report.

The basics were fine

None of this is surprising for businesses that build websites for a living, and it is a useful reminder that the technical checklist is not where most agency sites lose. We compared the platforms themselves in Webflow vs Framer vs WordPress for SEO.

Where case studies fall short

Most case studies are titled with the client's name, and nothing else. That is natural, because it is how a designer thinks about the project. It is also why, as we found when we looked at our own case studies, these pages tend to rank for the client's name and not for the service that won the work. A title like "Brand identity and website for [client]" tells a reader and a search engine what the page proves. "[Client]" alone tells them who it was for. When we planned the titles for a branding agency’s new website, its own Search Console data showed buyers searching for "brand identity design agency" far more than "branding agency", and that one detail shaped every title.

The missing body links matter for the same reason. Google says it uses links to discover pages and that good anchor text describes the page it points to. A sentence in the case study such as "we rebuilt their website on Webflow", with "website on Webflow" linking to that service page, connects the proof to the service in a way a menu never can.

Portfolio images

On the case-study pages with at least three images, the median page described 71% of its images with alt text. A third of pages (96 of 283) described fewer than half. On 19% of pages, the descriptions were largely repeats, with fewer than 6 in 10 different from each other, which often means a template is filling the field rather than a person.

For a design agency, the images are the work. Google says it uses alt text, together with the content of the page, to understand what an image shows. "Packaging design for a coffee brand, front and side view" is useful. A blank field, or the project name on every image, is not. We wrote about how to find repeated alt text and other text a template left behind.

AI crawlers and llms.txt

Of the 79 agencies with a valid robots.txt, 16 (20%) named any AI crawler at all. Nine named OpenAI's search crawler, eight to allow it and one to restrict part of the site. None blocked it outright. The rest are open by default, which is fine, as long as it is a decision rather than an accident.

25 of the 81 (31%) published an llms.txt file. That is a different sample from our 48-agency study, so the two numbers should not be read as a trend. There is still no published evidence that the file affects whether AI tools recommend a business.

How to fix your own agency site

1. Link every case study to its service, in the text. One sentence near the top, with the service as the link text.
2. Put the service in the case-study title. "Website redesign for [client]", not "[client]".
3. Describe the work in the alt text. What the image shows, in a phrase. Leave purely decorative images empty, and never repeat one description across a gallery.
4. Give each case study its own meta description that names the problem and the result, without inventing numbers the client has not approved.
5. Link each service page back to two or three relevant case studies, so the connection runs both ways.
6. Decide about AI crawlers on purpose. Read which crawler does what, then write the rule you actually want.

If you would rather have someone do this for your own site or your clients', that is part of how we work with design agencies.

Limits

• The sample is agencies that appear in search for their city, so it leans towards agencies that already take search seriously. The real picture is probably weaker.
• "Links to a service page" depends on recognising service pages by their addresses. A site with unusual addresses may be undercounted.
• "Names a service" in a title is a word match. It will miss some creative phrasings and catch some generic ones.
• We read up to five case studies per agency, chosen at random, not every page.
• Eight agencies blocked our checker, and we could not find case-study pages on 22 more. We report on the sites we could read.