More and more buyers start their research in ChatGPT or Perplexity, and most companies have no idea what those tools actually see when they read their site.
Vedran, the founder of Vizura Studio, a one-man branding studio in Zagreb, couldn’t answer that question for his own website either, so he built a free scanner to find out. Brand MRI checks 34 signals across 7 layers and reports on how readable a brand is to AI tools.
We asked him about the build, the limits of the scoring, and what a few weeks of real-world use has already changed.
Was there a particular client conversation that sparked the idea?
Not one conversation, more a pattern. My clients are mostly B2B tech and AI companies, and more and more of them mentioned that their buyers start research in ChatGPT or Perplexity. Nobody could tell them what those tools actually see when they read the site. I didn’t have an honest answer for my own site either, so I built something to check it.
How did you build the scanner, and did you need any outside help?
No outside people. I’m a designer, not an engineer, so I used Claude as a coding partner. I decided what to check, how to score it, and how the report should read, and Claude helped me write and debug the code. It runs as a small site on Cloudflare, with the scoring engine kept on the server and a 3D scroll story on the front, because that part is my day job.
Roughly how long did it take to get the first version live?
The first working version went live within a few days in late September, mostly late nights between client deadlines. The polishing since then has taken longer than the build.
What can Brand MRI reliably tell someone, and where does it still need human judgment?
It reliably checks the mechanical side: 34 signals across 7 layers. These include whether AI crawlers are allowed in, whether the page can be read without running JavaScript, whether the company is named and described the same way everywhere, whether there’s proof with real numbers, and whether things like structured data and llms.txt exist.
What it can’t judge is whether the brand is any good. A famous brand can score low because its proof lives in videos and images, and a clear positioning sentence still needs a person to say whether it’s the right one. The free scan also doesn’t ask ChatGPT anything directly, since it reads the signals those tools rely on. Nobody can promise an AI will recommend you, so the tool doesn’t promise that either.
Has any early feedback surprised you or prompted a change?
Yes, mostly from running it on a lot of sites myself. Scores bunched up in the middle, some very well-known brands scored low, and some sites scored badly only because they’re built in a way a crawler can’t read without running code. My own site scored very high, which told me the first version of the scoring was too shaped around what I had built myself.
So I’m reworking it now. I’m separating “this brand is unclear” from “I couldn’t read this page properly”, and making every point in the report cite something specific on the actual site.
Have people using the free tool gone on to enquire about your studio services?
Not directly yet, honestly, since it’s only been live a few weeks. Where it has helped more is my own outreach. Instead of a generic intro, I can send someone a short note on what a machine concludes about their site, and those get much better replies than a normal cold email.
Do you see it becoming a business in its own right, or remaining part of the studio?
Part of the studio for now. The free scan is the front door, and the paid side is Brand OS, where I turn a company’s brand into a system their team and their AI tools can actually use. I’m sketching a small paid version for teams that want to track their score over time, but it isn’t built yet, and I’d rather the free tool earns trust first.
