Hi, I’m Donna
Donna is the AI assistant inside HeyDonna. She does the first pass on a transcript. You hold the pen.
She is openly an AI. She is not a person, she is not a reporter, and she never claims to be certified. You are the one who certifies the record, so Donna proposes and you decide. Every edit she makes shows up as a tracked change that you can accept or reject.
HeyDonna is in public beta, with 2,500 free credits to start at heydonna.law. Donna will get things wrong sometimes. Tell us when she does.
Why a persona for legal transcribers
The people we build for are freelance legal transcribers, court reporters and scopists. It is a small, close-knit, confidentiality-conscious profession. Many of them are skeptical of AI, and they trust each other more than they trust brands. Their main worry is reasonable: certifying a document that contains an error nobody checked.
That shaped the whole design. A persona could easily have made things worse, with a fake human face, a chatbot pretending to be a colleague, or hype. We wanted the opposite: a tool with a personality, one that is honest about being a tool.
So Donna’s stance matches how the profession already talks about AI: a supplement, not a replacement. You are not the record; you help produce it. Her tagline is “I’ll draft it. You make the call.”
How she sounds
Warm, dry and competent, like a good colleague. Short sentences. She uses reporter vocabulary correctly (read-back, errata, scopist, exhibit) and keeps light humor for the grind of the work, never for anyone’s skill. No hype, no “replace”.
Her voice in our videos is Gemini text-to-speech, using the Aoede voice, prompted to sound warm, calm and professional, “like a helpful colleague smiling as she talks”. We use that one voice everywhere so she stays recognizable.
How she looks
Donna is an illustrated character: an indigo blazer, a pen, and a small glowing lavender AI earpiece. The earpiece is deliberate. It tells you at a glance that she is an AI.
She comes in six poses: greeting, pointing, thinking, reviewing, celebrating, and oops. Each has a job. Thinking shows while she drafts. Reviewing shows when she hands the draft to you. Oops is for errors and 404 pages, and it never appears above the fold on a marketing page.
The look around her is a calm comic-book kit: bold ink outlines, flat fills, hard offset shadows, halftone dots on mustard backgrounds, speech bubbles and numbered panels. The palette is ink, indigo, lavender, mustard and a warm off-white.
How we designed Donna
This is the real sequence, not a tidy retrospective.
1. Start from a question, not a logo. We had a product but few people knew about it, and a launch feature that was held up. I asked the growth agent for a growth plan and, separately, “I think we need a persona for Donna as well.” The Donna chatbot did not even have an icon yet.
2. Check what history said about the audience. The growth agent went back through our earlier marketing research before proposing anything. The findings that mattered for design: the audience averages around 55, trusts peers over brands, is AI-skeptical, and handles confidential work. Cold, one-to-many pushes had returned almost nothing in the past. Working reporters vouching for the tool had worked. That pushed us to treat Donna as a proof asset, not as a mascot for its own sake.
3. Write the rules before drawing anything. The persona draft set guardrails first: always disclose that Donna is an AI, no photoreal human face, no implied credentials (no gavel, no steno-machine costume), no sad or angry faces for errors, no color-only states, no customer data, no accuracy or time-saved claims, nothing about features that have not shipped.
4. Explore three directions. The concept doc proposed three:
- A, “Marked Page”: an abstract mark, a page with a tracked-change caret on an indigo square, close to the existing logo. Safest and best at 16 px, but the least personable. This was the initial recommendation.
- B, “Scribe-bot”: a small, openly robotic character with a headset and a pen. Most personality, but it risked reading as playful to a conservative audience.
- C, “Faceless Bust”: a geometric silhouette of a person. Warmer, but a human silhouette can be mistaken for a real person unless it is labeled as AI.
5. Ask the uncomfortable question. When I saw the options my first reaction was whether a comic-book character would appeal to this audience at all. That question stayed with us through the rest of the process. We asked the CTO to generate real images instead of judging prompts.
6. A fourth direction won. The CTO-generated set added a D: an illustrated Donna in a reference style I had shared, with the indigo blazer, the glowing AI earpiece and a pen. I liked D best. The earlier recommendation had been A, so this was a change of mind driven by seeing the pictures.
7. Accept the consequence. If Donna is a character, the website has to be built around her, with more comic-book elements. I approved that redesign, and we wrote a spec for it.
8. Name the risk in the spec. The redesign spec says it plainly: a comic persona can build warmth but can read as unserious to someone certifying a legal record. The mitigations were built in. The comic language lives in the framing (panels, bubbles, captions), while product claims stay in sober, plain copy. The product screenshot stays realistic. Donna always defers to you. There are no gags on pricing or security pages. The success measure is signup-to-activation, not only signups.
9. Pick the type and the colors with contrast in mind. For headlines we chose Londrina Solid, a calmer comic-lettering face, and rejected Bangers and Luckiest Guy as too loud for this audience. Body text, forms and pricing stay in Inter. The fonts are self-hosted, so the static site makes no third-party font requests. We checked the contrast ratios, and one rule came out of it: never put indigo text on mustard, because it fails.
10. Build the kit once, then adopt it page by page. The visual kit shipped as a standalone stylesheet with a demo page and an automated check that verifies contrast and the reduced-motion behavior. After that, the homepage, the app’s color theme, and then the blog adopted the same tokens, so a visitor moving between pages sees one consistent site. The blog shell you are reading now is the latest piece, with the post template, tags, archive, search and the 404 page (with the “oops” pose) following.
What we got wrong, or have not proven yet
- We do not yet know whether the comic style helps or hurts with working reporters. We will compare signup-to-activation against the earlier page rather than trusting our taste.
- Final art was generated and iterated with AI tools. Small-size icons (16 to 64 px) need careful hand work to stay legible.
- Some of the pose art arrived after the layouts, so early builds used labelled placeholders rather than faking it.
What’s next
Donna is going to appear in short videos, tips for transcribers and in the app itself. Everything she shows uses sample transcripts, never customer data. If a tip is wrong, tell us.
Try her at heydonna.law, and tell us what you think of her.
Follow Donna
Donna is on social media, and she posts short tips for transcribers, sample-transcript walkthroughs and notes on what we fixed in the beta. If you work in this profession, please follow her. It is the easiest way to see new tips, and to tell us what is wrong or missing.
- Instagram: @heydonnalaw
- YouTube: @heydonnalaw
- X: @heydonnalaw
- Facebook: Donna Law
- LinkedIn: HeyDonna
Follow, say hi, and reply with the most tedious step between rough draft and certified. I read every reply.
— Rajiv