How we count.
M1 · People reports
A report counts the moment it’s filed, unless it’s obviously spam or a duplicate. The story is separate: it only appears after a person reads it.
- Names of people and employers (we black them out, or ask you to edit).
- Health details, about you or anyone else.
- Threats, hate, or anything aimed at a person.
- Stories that aren’t about losing work to AI or automation.
Flagged stories go back to the queue. Turned-down reports still count; only the text stays off the site.
M2 · The news crawler
- 1CrawlEvery morning we read a fixed list of news sources and press-release feeds for layoff stories.
- 2ExtractAn AI model drafts the company, jobs cut, country, date and a suggested confidence. Yes, we use AI here, openly.
- 3CheckA human reads every draft against the source, fixes it, merges duplicates, or throws it out.
- 4FileApproved items go into the ledger with their source link and the date they were checked.
AAI explicitly cited. The company itself named AI or automation as the reason for these cuts.
BAI mentioned as context. Credible media linked the cuts to AI; the company didn’t say so. Speculation, opinion pieces and anonymous rumours don’t count at all.
M3 · Why we never add them together
Different units. Pink counts people who reported. Yellow counts jobs that companies said they cut.
Unknown overlap. Someone in a news story may also have reported here. Adding would count them twice.
Different blind spots. Freelancers rarely make the news; big layoffs rarely produce reports. Each copy misses what the other sees.
M4 · Limits of the data
- Self-selection
- People who feel strongly are more likely to report. These are stories, not a survey.
- English only
- The site and crawler work in English, so some countries are under-counted.
- Announced isn’t done
- Companies sometimes announce cuts that change later. We correct when we see it.
- “AI” is fuzzy
- People and companies use the word for very different tools. We record what they said.
- No causation
- A report shows what someone experienced, not proof that AI was the only reason.
- Small numbers
- Below 5 reports a job has no page of its own, because the numbers swing too much.
Questions
Can I trust a self-reported story?
Not as proof, and we never present it that way. It’s someone’s account. That’s why we show the verification level on every slip and keep self-reported, confirmed and news numbers separate.
Why do you use AI in a project about AI job loss?
For one narrow job: drafting fields from news articles. Every item is then checked by a person before it appears. Stories from people are never edited or judged by AI.
Could someone flood the site with fake reports?
We use an invisible bot check, a small daily limit per connection, and duplicate detection. Confirmed reports are shown separately so you can choose to look at those only.
Why don’t you show company names from people’s reports?
Because it could identify the person, and because one person’s account isn’t a fair basis to name a company. Company pages show news only.
How often is the data updated?
People reports: counted when filed, statistics refreshed every hour. News: one crawl every morning, published after checking. Every page shows its last update time.
Can I use the data?
Yes, under CC BY 4.0. Download it on the Data page and credit “I lost my job to AI”.