OLX Jobs: Job Post Templates
AI-written templates that help employers write job ads candidates can use
Role
Lead Product Designer
Team
PM, data analyst and engineering manager
Outcome
807% increase in template adoption · 1.4% shorter time-to-post
Summary
Job descriptions on OLX had a median of 59 words, and the description field was where most employers dropped off. Over three iterations I moved from splitting the field to optional, AI-written templates that work both for listers writing from scratch and for those pasting a ready ad.
Helping employers write job ads candidates can use, without slowing them down
Context
OLX is Poland’s leading classifieds platform, with over 13 million monthly users, and Jobs is one of its key categories. When we looked at job ads, the median description was 59 words. That isn’t enough to answer the questions candidates told us they ask first: what will I do day to day, and what do I need to get this job? Thin descriptions also weakened search and recommendations, so good candidates skipped ads and poorly matched ones applied.
For employers, the description was the hardest part of posting. It was a blank field with one generic tooltip shared across all categories, and it was the step where we saw the most drop-offs in the posting form.
From v0 to v3: a single empty field, the split field listers pushed back on, then two iterations of optional templates. In the last one, templates are written with AI and shown as job-title chips above the field.
The problem
Two groups needed different things:
Candidates compare offers across several job sites. They told us they like the clear, repeatable structure competitors use: responsibilities, requirements, and what’s on offer.
Employers often didn’t know how to start, and the description field was where most of them abandoned the form.
Success meant fewer drop-offs in the posting form and ads that answer candidates’ key questions.
First attempt: splitting the description field
My first idea followed the benchmark: split the field into three sections, each with its own hints. After design critiques and team refinement, I tested it with 5 employers. The feedback was promising, so we ran an A/B test with a small group of users.
Before launch, we briefed customer care and sales, the teams closest to our users. That paid off. Through them we learned that a large group of listers, including important business clients, arrive with a ready-made description and simply paste it in. For them, copying text into three fields instead of one was real friction, and they pushed back hard.
This gave us a new constraint: the solution had to work both for people writing from scratch and for people pasting a description they already had.
Templates, version 1
We kept a single field and added optional description templates on top. They gave people starting from scratch a structure and the content candidates need, without getting in the way of anyone pasting their own text. The team wrote the first templates by hand for the categories with the weakest descriptions. We tested them the same way as before: usability sessions first, then an A/B test.
What we learned: once people found a template and saw its preview, they were eager to use it. The problem was that too few people found them.
Templates, version 2
In the second version I made templates easier to notice and browsing them more fluid. Writing templates by hand didn’t scale to more categories, so we tried generating them with AI. Before relying on them, we tested AI-written templates with users, and they rated them better than the ones our team had written. That let us cover more categories, faster.
Impact
807% increase in template adoption from the first to the second version, measured in an A/B test.
1.4% shorter time-to-post: adding structure didn’t slow listers down.
Key decisions
Optional, not required. Forcing a structure would have helped candidates but hurt listers who paste ready descriptions, including key business clients. Making templates optional kept both groups on board.
Customer care and sales in the loop from day one. It took a few extra briefings, but that’s how we learned about the paste problem quickly, and they could support users during the experiments. We also checked whether pasted descriptions differed much in structure from our templates, and ruled that risk out.
Testing AI output like any other design. We used AI to scale content, not to replace judgement. AI-written templates went through the same user testing as the hand-written ones.
Templates, version 1: a “Use our templates” link opened a sheet where listers picked a job title and previewed the template before adding it.
Templates, version 2: job titles shown as chips right above the description field. One tap fills in a template that listers can edit before posting.
What I’d do differently
I usually explore three directions: a “sky is the limit” one that ignores current constraints, the best one for users that we can realistically build, and a quick, cheap one to validate the idea. On this project I leaned too far toward cheap. I held myself to solutions that were as cheap as possible to build, and that narrowed the alternatives I explored for the first templates. Next time I’d keep cost out of the exploration phase and bring it in only when choosing what to build.
The project also reminded me how different our job listers are, from small businesses writing their first ad to large clients with ready-made descriptions. Designing for all of them means constantly balancing their needs against each other and against business priorities. The people who hear from them every day are the fastest way to learn where a design breaks.
What’s next
The next step we planned was letting listers generate or improve their own description with AI, not just start from a template.
Previous project
Next project