Designing an AI writing tool that publishes finished articles people are willing to put their name on. End to end product design, research, system design, and editorial craft — solo.
Most AI writers solve the wrong half of the problem. They generate volume. They do not generate articles you would publish under your own name. Publizo had to fix that, without turning the product into a thirty step wizard.
What I committed to: A free tier that hands the user a finished, ready to publish article on their first try. A quality gate that quietly sends weaker drafts to draft status instead of pushing them live. A visual system that does not look like a generic SaaS template. Pricing that you can understand in one screen.
As a solo team, every decision had to remove future work, not add it. That ruled out custom illustration, mascots, and a marketing site that needed weekly maintenance.
Two weeks of reading SEO forums, Reddit threads, and YouTube comment sections instead of running formal interviews. I wanted to hear people complain in their own words.
I cut the dashboard down to four real jobs. Everything else either lives behind a setting or does not exist. The free tier and the paid tier share the same shell so people are never relearning the product when they upgrade.
Templates suggest you need to learn the tool. Publizo's promise is the opposite. Tell it a topic. Get a finished article. Every competitor groups articles under "projects" too — I tested it and found the layer added clicks without adding meaning. Sites already carry the project context.
The palette is intentionally restrained: warm paper, deep ink, and a single indigo accent. Combined with Fraunces and Inter, it reads as software made by someone who cares about the written word, not someone who reads dashboards.
Typography: Fraunces for display — used for headlines and editorial moments where personality matters, set at 600 weight with tight tracking. Inter for body and UI, limited to 400 and 500 weights to keep hierarchy consistent and the experience calm.
Layout: Twelve column grid, capped at 1120 pixels, 24 pixel gutters. Sections use generous vertical rhythm. Borders are reserved for moments that need a frame. Cards have a single elevation. Shadows are warm, not blue.
Motion: Custom cubic bezier of 0.22, 1, 0.36, 1. Page transitions under 240ms. Nothing bounces. Nothing slides further than it needs to.
The two flows that decide whether a user stays: onboarding and the first published article. I redesigned both around a single principle — show the value before you ask for the work.
Time to first published article dropped from over four minutes to under ninety seconds for users who connected on the first try. Drop-off at the credentials step fell by roughly a third.
This was the real design problem. Not the UI. Not the dashboard. Whether the article that lands on the user's site reads like a person wrote it. Most of this work was prompt design, eval design, and writing house style rules.
If the model returns a piece below the publish threshold, the system retries silently with a stricter prompt. Only after a second miss does the article go to draft — protecting the user's credits and reputation at the same time.
Voice presets. Each site picks a voice. News reads like a wire desk. Astrology reads like a thoughtful column. Commercial intent reads like a careful buyer's guide. The same engine, completely different rhythms.
Pricing is usually the page where design effort dies. I treated it as a product surface. Same type, same grid, same restraint as the rest of the app. No comparison table that needs a magnifying glass.
The paid tiers list what they include, not what the cheaper tiers do not. Negative selling makes the page feel hostile.
Selecting a paid plan now holds the intent through the sign-up flow, then sends the user straight to checkout instead of dropping them on a dashboard and expecting them to find billing on their own.
Founder's note. While working on design projects, building products, and doing content research, I found myself neglecting my own blogs. Writing, optimizing, formatting, and publishing every article took more time than I could spare. That is what inspired Publizo — a tool that automates the busywork so I can focus on creating content people actually want to read.
What I would do differently. Ship the quality gate on day one. Most early complaints were about articles that should never have been published in the first place. The fix was the gate, not better copywriting on the marketing page.
Five free articles every month. No credit card. Full editorial quality.