A model can draft a listing in nine seconds. It cannot tell you which claim will get the page pulled, or whether the second bullet sounds like a robot. We use the machine for the first ninety percent and a person for the part that decides whether anyone buys.
Harvest, draft, sign off.
Three steps, in that order, every time. The last one is not optional.
Harvest
Atlas · our data layer
Atlas reads the category every night: what competitors rank for, where their listings are thin, and the complaints buried in a few thousand reviews. That comes back as keyword clusters and a brief, not a spreadsheet.
- Category and competitor benchmarks
- Gap analysis against ranking listings
- Pain points mined from review text
- Search intent mapped to real queries
Draft
Machine first pass
The brief becomes titles, bullets and A+ copy built around how each marketplace currently ranks. It is fast, it is thorough, and it is never the version that ships.
- Titles and bullets against the brief
- A+ and Enhanced Brand Content
- Backend terms and structured attributes
- Image and video direction
Sign off
Kevin Park · Head of Listing
A person reads every line before it goes live — for brand voice, for claims that would get the listing pulled, and for whether it actually sounds like a human wrote it. Nothing ships on a machine draft alone.
- Brand voice and tone
- Claims checked against the rulebook
- Read on a phone before it ships
- A/B test defined at launch
The machine is fast. The person is accountable.
Atlas is our internal data layer — it reads categories, tracks ranking drift and drafts. It is software, and we name it as software. Kevin signs off every line that reaches a customer, and his name is on the result.
Atlas does
- Keyword clusters
- Competitor gap analysis
- Intent mapping
- First-pass drafts
- Ranking drift alerts
Kevin decides
- Brand voice
- Claims and risk
- Emotional register
- What gets tested
- What actually ships
Kevin Park
Head of Listing & Conversion
“The draft is never the problem. The problem is the one claim nobody checked, and the bullet that reads like it was written for a search engine. That is the job.”
Four things every listing has to survive.
Search, not guesswork
Keyword clusters built from live category data every week, not a tool export from last quarter.
- Category-specific keyword mining
- Gap analysis against ranking competitors
- Search intent mapping
- Backend term optimisation
Refreshed weekly
Built for a phone
Most of your traffic is on a 6-inch screen holding it one-handed. We design for that first and check it there before launch.
- Mobile truncation checked on device
- Graphic-led imagery that reads small
- Infographics and comparison modules
- Video thumbnails that earn the tap
Mobile-first, verified
Legal before live
Every line passes a compliance check before publish. A pulled listing in peak season costs more than a year of optimisation earns.
- Licence and labelling validation
- Bilingual requirement checks
- Claims verification
- Restricted term detection
Checked before publish
Written to be parsed
Marketplace search is increasingly answered by a model rather than a keyword index. We write so the machine understands the product too.
- Complete structured attributes
- Semantic clarity over keyword stuffing
- Question and answer coverage
- Consistent entity naming
Built for AI-mediated search
“A listing has two readers now: a person deciding in four seconds, and a model deciding whether to show it at all. Write badly for either and you lose.”
Kevin Park · Head of Listing & Conversion