Everything AI makes looks the same. How to teach AI your brand

Why everything is indigo

In August 2025, Adam Wathan, co-creator of the popular Tailwind CSS framework, posted a tongue-in-cheek apology on X. Five years earlier, every button in the ready-made Tailwind UI components had been given an indigo default. The post was viewed more than one and a half million times, because anyone who had generated interfaces with AI got the joke immediately.

“leading to every AI generated UI on earth also being indigo.”

Models learn from what’s most common online and reproduce it when they get no other guidance. Without a description of your brand, you get the internet’s average: the same colours, similar slide layouts, familiar phrases.

What sloppy AI output costs

This kind of material already has a name. In 2025, researchers from BetterUp Labs and Stanford described “workslop” in Harvard Business Review: AI output that looks polished but lacks substance, so it offloads the work onto colleagues. In their research, 41% of workers had encountered it, and each case cost nearly two hours of rework. In a newer, not yet peer-reviewed study by Jeff Hancock of Stanford, a co-author of that piece, covered by The Guardian in April 2026, 40% of 1,150 surveyed US desk workers had run into workslop within a month. Dealing with it took them an average of 3.4 hours a month.

In sales and HR the cost is double. A deck that doesn’t look like the brand weakens the company in front of a client. A job ad in a random style tells candidates more about the company than it would like. And if the company has a designer, they end up fixing it anyway.

Workday and 23,000 slides in six weeks

Workday is a good example. When the company refreshed its visual identity, its solution consulting team, which supports sales, had six weeks to update 23,000 customer-facing slides and demos: new fonts, colours and messaging. There was no time for manual updates or for handing it to an agency.

The team used an AI tool that moved old slides into the new template according to the brand guidelines. According to a case study published by the tool’s vendor, Prezent, every slide was converted on time, AI removed about 80% of the manual work, and the savings exceeded 675,000 dollars. These are vendor figures, so we treat them with care. The mechanism itself is simple, though: brand rules written down once and applied automatically to thousands of files, with people refining the details.

Stripe and Anthropic build the rules into their tools

The AI in Design Report 2026 by Designer Fund and Foundation Capital shows that some companies go further. At Stripe, two tools grew from the bottom up. Protodash lets designers and engineers build interactive prototypes on the company’s design system. Dante is an AI agent that drafts, edits and reviews content to apply Stripe’s standards, built by the content team to keep quality up as the volume of content grows. The report’s authors note that Stripe increasingly builds best practices into the tools themselves instead of only documenting them.

At Anthropic, designer Nate Parrott built a tool in late 2025 that generates interactive prototypes grounded in the company’s design system. Designers used it, and so did non-designers, including salespeople and educators, who made visual artifacts with it that they couldn’t have made before. It shows what a rule built into a tool gives you: someone with no design background uses the design system without even thinking about it.

A brand book PDF isn’t enough

Most companies keep their brand rules in a brand book, a PDF few people open. Alex Dousie of Frontify describes in a post on brand governance what this looks like in practice. Teams paste bits of the guidelines into prompts, upload old files and summarise tone-of-voice rules by hand. Dousie’s point is that prompting alone doesn’t replace a proper brand system.

An AI-readable brand is the same set of rules, written down so tools can apply them without guessing. Colours and typography as concrete values. Templates for the most common materials, from a sales proposal to a job ad. Tone of voice shown through example sentences that fit the brand and ones that don’t. A photo and illustration style. All of it connected to where people actually work: brand kits, instructions and projects in the AI tools the team uses.

Then sales, HR and marketing can make materials that look like your brand on their own. That includes companies without an in-house designer that would like to automate content but don’t know where to start. In our service this is the AI that knows your brand area. For how the same principle works in campaign production, read our post on marketing asset adaptations. If your company doesn’t have a consistent identity yet, the work starts one step earlier, because AI can only reproduce what someone designed first. We describe what that project looks like on our visual identity page.