Cleaning up 340 help articles across three markets

Problem statement

Back Market's Seller Support Center had 340 help articles across three markets — EU English, EU French, and US English — and none of them had ever been audited at scale. The articles had been written at different times by different people, and it showed: outdated policies, old terminology, misleading information, inconsistent formatting, and emojis that had no place in a professional support context. For sellers trying to resolve issues independently, the SSC was supposed to be the first place they turned. In practice, it wasn't reliable enough to be.

My role

I proposed this project and led it end-to-end. I scoped the work, built the tracking system, coordinated 15 subject matter experts and their managers across teams, ran check-ins, and did the majority of the content editing myself.

Process

Prioritising with AI

340 articles is a lot to triage. To decide where to start, I used AI to rank articles by urgency based on three signals: view count, date last updated, and whether the article was surfaced in our seller support decision funnel. The highest-priority articles were the ones sellers were most likely to encounter and least likely to find useful. That ranking shaped the order we worked through everything.

Getting the facts right first

Before I could fix the writing, the content needed to be accurate. EU and US markets operate under different policies, so articles in each market had to be reviewed separately. I assigned ownership to SMEs based on their area of expertise and built a tracker to manage progress across all 340 articles. Getting 15 stakeholders to actually complete their reviews took consistent follow-up — that was most of the project management work.

Applying the style guide

Once the facts were confirmed, I rewrote each article against Back Market's content guidelines. That meant fixing tone, updating terminology to match our glossary, restructuring for scannability, and cutting anything that didn't belong. The through-line was consistency: a seller reading an article in the US market and one in the EU should encounter the same standard of clarity, even if the policies themselves differ.

Building a tool to help stakeholders contribute

To make it easier for SMEs to draft content that already aligned with our standards, I built a Dust AI agent trained on Back Market's voice and tone guidelines and preferred terminology. It reduced the editing burden on my end and gave stakeholders a way to internalize the style guide through using it, rather than reading it.

Impact

All 340 articles were updated over six months. The SSC went from a fragmented, unreliable resource to a consistent, policy-aligned knowledge base — and one with a more significant role than it had before: the updated articles now serve as the primary knowledge source for Back Market's first seller-facing AI agent. The quality of that agent's responses depends directly on the quality of the content underneath it.

I'm tracking article views as an early signal of whether the updates are landing — higher views on previously underperforming articles would suggest sellers are finding what they need and coming back. In an ideal world I'd also have read-through rates, which would let me build a hierarchy of which articles sellers are finishing and which ones they're abandoning halfway through. That's the signal that tells you whether the content is actually working, not just being found.