DiscoRadar
A Google Discover radar for newsrooms. It reads the feed every five minutes and shows which stories, publishers, and topics Google is recommending across Ukraine.
- My role
- Product Owner, Data & AI Architect, Full-stack Engineer
- System
- React, TypeScript, Express, Firestore, BigQuery, Gemini, OpenAI embeddings
What editors could not see
Google Discover sends a large share of traffic to Ukrainian news sites, and it tells a publisher almost nothing. Search Console reports your own clicks a day or two late. It says nothing about the story a competitor has held at the top of the feed since morning.
I built DiscoRadar for the Telegraf.com.ua and OBOZ.UA newsrooms to answer four questions an editor asks all day. What is in demand right now? Which competitors own which topics? Where is our own coverage missing? What should we publish next, and at what hour?
How a reading is taken
The radar measures the feed the way a reader meets it, as an ordered list of cards on a phone.
- 01
Read the feed
Signed-out mobile sessions read Discover as it appears in Kyiv, Lviv, and Dnipro, in Ukrainian and in Russian. A scan runs every five minutes.
- 02
Keep the sighting
The article and its appearance are stored separately. One article can show up in several cities, at different positions, for hours.
- 03
Score it
Each story gets a visibility score. A higher place in the feed and a longer stay both raise it.
- 04
Weight the cities
City readings combine into one national number, so a story seen only in Lviv does not outrank one seen across the country.
- 05
Group the stories
Headlines about the same event are merged into one topic by meaning, so an editor reads a list of stories and not a wall of near-identical cards.
Reading the feed
The first screen is the feed itself, ranked by visibility. Each row has its score, a trend line, and the time the story has spent in Discover.
The priority label is the part editors act on. It says whether a subject is worth picking up first or after stronger ones.
The score measures placement in the feed. It is not traffic, and the interface never presents it as views or clicks.
Topics
Stories about the same subject collapse into one topic. Over a week the radar tracked 169 topics across 8,464 stories.
A topic card shows its visibility curve, the stories in the feed now, and which publishers hold the largest share of it.
Competitors
The same sightings, grouped by publisher. The table separates volume from staying power.
In this week one outlet led with 557 stories that lasted about two hours each. Another reached third place with 65 stories that each held the feed for more than seven hours.
Who owns which category
The matrix crosses categories with publishers. A section editor sees at once who holds the category and where the gap is.
When to publish
Google reports a publication time for every card. The radar joins it with the article's first-day visibility and compares hours and weekdays, for the newsroom and for its competitors.
The view also measures how long an article takes to reach Discover after it goes out.