Hired as the only UX writer. Ten months later: a style guide, a cross-functional workflow, and an AI plugin that catches copy issues before they ship.
1 · Diagnose
I was hired as eToro's sole UX writer. The brief: "Here's a mess. Let us know when you've fixed it."
First move: a formal audit — to gauge the severity and win the mandate to fix it.
2 · Audit
I needed to gauge the severity before I could change it. Combed every core flow, screenshotted every screen, AI-analyzed what I found. Compiled the findings into a deck for the VP of Product.
The deck proved the problem — and bought me the mandate to fix it. Next: winning over the PMs who'd been creating it. Most of them had never worked with a UX writer.
Open the audit deck →3 · Establish
Most PMs had never worked with a UX writer before. Some didn't know what UX writing was. Before any process could stick, I had to introduce the role itself.
So I worked 1:1 with every PM and ran a UX-writing-101 deck at the Product Unit's monthly meetings. Once trust was there, I rolled out a Monday-board submission flow — PRDs, Figma links, context all in one place. First time I had a real queue. But the queue didn't multiply me. Next I'd need automation — because one writer can't review every string going live at the volume PMs were shipping.
Open the UX-writing-101 deck →4 · Automate
Even with my queue, only one me. Live copy was shipping at volume across the product — well beyond what any single writer could review by hand.
So I partnered with the head of R&D to add an AI English-vetting layer in front of production. Every string passes a hygiene + style check before it ships. PMs upload a screenshot so the AI knows a button from an H1. Live spelling mistakes were dead — but whether the slower 1:1 work was actually moving the PMs, I still couldn't tell from inside.
5 · Measure
By January I'd been at eToro six months. Time to ask the PMs on the receiving end if any of it had actually worked.
A pulse survey of 16 came back: 100% rated the work 4 or 5 out of 5. Weekly UX-writing collaboration had tripled — 13% to 40%. Now I had proof to pitch upward.
6 · Standards
R&D leadership bought my pitch: shift from fixing bad copy late to preventing it. Prevention needed a foundation first — rules concrete enough that an AI could enforce them. So I wrote eToro's first comprehensive UX writing style guide.
Voice, tone, financial-risk language, UI standards, terminology, localization, compliance, governance. 157 rules. Built specifically for trading — how to communicate risk without creating fear, how to stay calm during payment failure or market volatility. The rules were the spec. Next: the agent that would enforce them automatically — capitalization, terminology, voice — at design time.
Open the style guide →7 · Build
Click once. It reads every screen the way I would — flags what's wrong, explains why, learns from PMs' pushback.
Live in the eToro Figma org since May 13. Built solo with Claude Code. But the plugin still checks copy from outside. Next: the rules move inside the components themselves.
See the plugin in detail →8 · Apply
The voice guide gave us the framework. Now I'm working with the head of our design system to build content specs into the components themselves — versioned in the same Figma library as the visual tokens, part of the component contract.
Four components shipped so far: Toast, Confirmation Modal, Unsaved Changes Dialog, Destructive Modal. Each has its own spec — formulas, examples, constraints — calibrated to a friction hierarchy from "announce and move on" to "confirm irreversible." More components in flight.
See the components →The throughline
Audit gave me credibility. Trust gave me access. AI vetting stopped the bleeding. The survey proved the work mattered. The style guide became the spec. The agent put it in every PM's Figma. The specs are moving into the components themselves. What started as one writer firefighting is now an AI content system — live, in use, every day.
