Intuit · Sr. Content Designer · 2023

An AI Bill of Rights

People were being asked to let a model act on their money. Before we wrote a single prompt, we wrote down what it would never be allowed to do.

Fintech · AI integration

We wrote a position before anyone asked us for one

Intuit's published guidance for writing content for AI experiences

Intuit’s public guide for writing AI content. The internal principles came first.

This was 2023. The models were new, the press was loud, and customers were being asked to let software act on their behalf inside their own books. Nobody was worried about tone. They were worried about whether a company with their bank feed should be allowed to make moves without asking.

So content design wrote an AI Bill of Rights: a statement to leadership about what AI in our products would and would not do to someone’s financial data. Not a style guide. A set of limits, written before the thing that needed limiting had shipped.

We didn't argue about quality. We scored it.

Generated outputScoreVerdict
  • Your books are looking great! Let's keep the momentum going.2bad
  • We sorted 42 transactions. Check the 6 we weren't sure about.4not bad
  • Effortlessly manage your finances with smart automation.1bad
  • This expense looks personal. Move it out of business?4not bad

Representative rows, rebuilt. The sheet itself is behind the same lost login.

My old portfolio said this took “a lot of testing, collaboration, and spirited debate.” That was flattering and mostly untrue. There wasn’t much debate, because we stopped having the conversation that produces debate.

Every generated draft went onto a Likert scale and into a very large spreadsheet. Quality, voice, tone, one row at a time. Once a stakeholder is looking at a column of numbers, “I don’t like it” turns into “I scored it a two, and here is why.” That is a completely different meeting.

The honest finding, in 2023, was binary. Output was either bad or not bad. The models weren’t good enough for finer distinctions, and pretending otherwise would have wasted everyone’s time. Naming that plainly is what let us ship anything at all.

The principles ended up governing a product I never touched

Product teams picked them up when they built governance for Intuit Enterprise Suite, the company’s move upmarket from small businesses toward larger, more complex ones. I wasn’t in those rooms. The document was.

That’s the outcome I’d point at. Not that we wrote something good, but that it travelled to a part of the business none of us worked on.

I can't show you the principles, and that's the part that stuck

They’re behind a login I lost in a layoff. I helped write them and I cannot reconstruct them, not really. The rules I could approximate. The reasoning is gone: why we drew each line where we drew it, which argument each one settled, what we were afraid of in the room that week.

That loss is normal and almost nobody plans for it. Institutional memory in tech is a handful of people who happen to still work there, and the actual reason for a decision is usually one of two things: evidence, or the highest-paid person had an opinion on a Tuesday. Only one of those survives being written down, and neither survives a badge getting turned off.

Everything I’ve built since keeps its decisions, and the why underneath them, in plain files I own. Every rule carries its reasoning, because I have already been the person who couldn’t remember.

Pull me into a conversation →

Project details

Company
Intuit, QuickBooks
Role
Sr. Content Designer
Timeline
2023
Tools and methods
Content principlesPrompt evaluationLikert scoringDesign system contributionCross-functional governance