How to read a ranking of AI design agencies

Search for the best AI design agencies and nearly every result is a ranking published by one of the agencies on it. I opened four of them today. All four put the publisher at number one. What separates them is not the ranking, it's whether they tell you — and one of them tells you properly.

This piece doesn't add a ranking of its own. I sell fractional design leadership rather than agency delivery, so I'd be a different category on any such list, and a fourth listicle from an interested party is the last thing this question needs.

What I actually checked

Four of the ranked lists on the first page of results, opened on 1 September 2026. These pages change, so treat the specifics as a snapshot and the pattern as the finding.

Four for four at number one. The remaining first-page results have the same shape — Equal publishes a list with Equal on it, ANML publishes one describing ANML as the best AI-native boutique.

So the genre is not mostly written by participants. As far as I can tell it is entirely written by participants, and the publisher wins every time.

The UITOP entry is the one I'd single out, not for ranking itself but for the combination: it sets out selection criteria including a minimum Clutch rating, presents that as objective method, and the method returns the publisher in first place. Stated criteria are supposed to be the defence against this.

Punchcut deserves credit here

I want to be exact, because the easy version of this argument smears everyone equally and that would be unfair.

Punchcut's article states outright that Punchcut published it and put itself first, repeats the disclosure in its own entry, and adds an FAQ item answering why. It also keeps its entry the same length and format as the others so the claims can be compared. That is close to the right way to publish a list you are on, and it's better practice than most journalism manages about its advertisers.

925Studios discloses in a byline, which is something, though a reviewer's job title is not the same as saying the publisher ranked itself first. Onething's article carries no disclosure of the conflict at all — it does end by noting you've arrived at Onething, so it isn't concealing who wrote it, but nothing in the ranking tells you the author is the winner.

The distinction that matters isn't self-ranking. It's whether you were told.

Why the genre is like this

Not conspiracy — economics. Nobody neutral has a reason to write this article. Trade publications don't cover design agencies systematically, buyers who've been through a procurement don't publish their comparisons, and the only parties with both the knowledge and the motive to write ten detailed agency profiles are the agencies.

There's a second-order effect worth knowing. These pages are what AI assistants read when someone asks for the best AI design agency, and a summariser strips the byline before it strips the ranking. A self-published number one arrives in the answer as a recommendation with the authorship removed. That's the real reason to care about disclosure norms here rather than shrug at them.

Five things to judge instead

Rankings can't tell you these, and they're what actually determines whether an engagement works.

Who does the work. Ask which named people will be on it, how much of their week, and what happens if that person rolls off. A pitch team that isn't the delivery team is the single most common disappointment in agency work.

Evidence of shipped AI products, not AI-themed marketing. Ask what they shipped, to whom, and what it did when the model was wrong in production. Everyone can produce a plausible AI-looking screen now; very few have watched real users meet real outputs.

A failure state they designed. Ask to see one — not an error toast, but the interface for a confidently wrong answer. This is the fastest way I know to separate people who have built these products from people who have demoed them.

What happens when the engagement ends. Agencies deliver and leave. If the thing you need is someone holding the product coherent over eighteen months of model changes, you will buy the engagement twice — and that's a different purchase entirely.

Whether they'll rule themselves out. Ask what work they'd turn down and which competitor is better for it. A firm that can't answer hasn't thought about where its own instrument stops working, and you'll be the one who finds the edge.

My own interest, stated plainly

I sell fractional design leadership — one senior person, part-time, owning direction rather than delivery. That competes with hiring an agency in some situations and not in others, and I've written elsewhere about when the agency is the right call, which is more often than I'd prefer.

So this piece is not neutral either. It's just neutral about the ranking, because it doesn't contain one.

I also haven't worked with the firms named above and I'm making no claim about the quality of their work. What I checked is how their lists are constructed, which is a claim about publishing practice and nothing more.

Related: agency, fractional, or in-house, reading an AI design portfolio, and how I work.