Why do our case studies rank for our clients' names but not for the services we sell?

The short answer

A case study can name the client in the title, the heading, the images and the URL, and never once state plainly what the agency did, for whom, under what constraint. When that is true, a search engine has been given a page about the client. It will treat it accordingly.

That is the common diagnosis and it is often right. It is not the only one, and jumping to a rewrite before checking the alternatives is how agencies spend a quarter retitling projects for no movement.

Four queries, one complaint

"We rank for the wrong things" almost always turns out to be four different observations added together. Split them before deciding anything.

Assign these in a fixed order and keep them disjoint, so a query lands in exactly one bucket. Then look at the counts. An agency convinced it has a case-study problem often turns out to have almost no non-brand service impressions at all, anywhere on the site, which is a different and larger problem.

What the platforms document

Two facts here change what a rewrite can be expected to do.

So if a project page is titled with the client's name, that name is probably reinforced by the H1, the images, the anchor text of every link pointing at it from your own portfolio index, and the way other sites link to it. Editing one field of that set is a weak intervention. Making the agency's contribution legible across the page is a strong one.

This matters more for a small agency than almost anyone. If your site is low traffic, a large share of your real queries never appear in the table at all, and the visible rows are a biased sample skewed toward your most common query, which is usually your own brand name. Percentages built on visible rows are not population shares, and the honest report says so.

We failed our own audit

Before writing this we ran the check on our own two case studies. Neither passed.

We extracted every internal link inside the article body of both pages, excluding global navigation, the footer, and any call to action outside the article element. Checked 23 September 2026.

Both pages demonstrate work we sell. Neither contains a single contextual link, inside the body, to the page that sells it. We had connected our evidence to more evidence and stopped there.

This does not mean the site has no service links, that a reader cannot find the contact page, or that anything is being suppressed. Navigation exists on every page. What it means is narrower and more useful: there is a difference between publishing evidence and connecting that evidence to the service it supports, and we were on the wrong side of it while planning to write an article telling other people about it.

Crawlable links with descriptive anchor text are how a search engine learns that one page is the evidence for another. Navigation says every page exists. Body links say which pages belong together.

Three jobs a portfolio page can do

These are our editorial categories, not a scoring system Google runs. Most case studies do the first well and the other two by accident.

A deliberately visual case study that does only the first job is a legitimate business decision, not a mistake. Design studios sell on craft and a wall of beautiful work does that. The mistake is having made that choice unconsciously and then being surprised that the page does not generate enquiries.

What a rewrite actually changes

An illustration, not a client. The company below is invented.

Northline is fictional and so is the work. Do not publish an illustration like this as a commissioned project or attach invented performance figures to it.

Worth noting the counterexample too, because the simple version of this advice is wrong. Pentagram's project pages lead with the client name as the heading, then carry discipline and sector labels and a narrative separating problem, strategy and solution. A client-name heading is not the diagnosis. What surrounds it is.

The six-field worksheet

Code each portfolio page present, absent or unclear. No scores. A score invites an average, and an average across six unrelated fields means nothing.

1. Service described. Does the page state what discipline this was, in words a buyer would use?
2. Audience or industry named. Can a reader tell whether this is relevant to them?
3. Problem stated. What was wrong before?
4. Scope stated. What did the agency do, versus the client's team or another supplier?
5. Evidence and limitations. Any measured result, and any honest statement of what is not known?
6. Contextual service link. A body link, with descriptive anchor text, to the page that sells this.

Fix your selection rule before you audit, such as the four oldest still-relevant projects, rather than four chosen because they support the conclusion. If you can, have a second person code a subset independently and resolve the disagreements. That single step is what separates an audit from a justification.

When the client-name ranking is fine

Four reasons not to act, which the usual version of this advice skips.

• It has real value. Ranking for a client's name supports referrals and due diligence. A prospect who heard about you second-hand will search the client, find your page, and that is working.
• Different pages have different jobs. Forcing a case study and a service page to compete for one phrase blurs the architecture and often demotes both.
• The page may be assisting rather than landing. A page can shape a decision without ever being the first click, and last-click reporting will call it useless.
• The problem may be elsewhere. Before blaming portfolio copy, check the service page is accessible and indexed, and check what the results page actually favours for that query. If Google is showing directories and listicles, no case study is going to displace them.

Limits

• The own-site audit covers two pages on one site, ours. It is an illustration of a distinction, not evidence about agency websites generally.
• The Pentagram observation covers two pages from one studio, inspected as public examples. No performance data was available and two pages cannot establish what is common.
• We have not run the controlled version of this: matched pages, half improved, half held back, over a predeclared window. Until someone does, the causal claim is unproven and we are describing a hypothesis with a good mechanism behind it.
• Query classification depends on visible Search Console rows, which exclude anonymised queries. Every percentage from that source is a share of what you can see.
• Platform documentation checked 23 September 2026.

If the deeper question is whether the site's evidence is reaching AI answers at all, that needs a fixed prompt set rather than a rankings report, and our own record of building a search layer for a branding agency is published as a case study. If you want the underlying model of how Google decides any of this, we set that out in full.