Why the Algorithm Is Rejecting You Everywhere
Remote work made a promise to talented people in Lagos, Bridgetown, Nairobi, and Kingston: your skills, not your postal code, would decide your future. A laptop and a good portfolio could put you in front of an employer anywhere.
That promise is quietly breaking. Before a human reads a word of your CV, software now parses you, scores you, ranks you, and very often rejects you. About 90% of U.S.UK, and German employers use AI to screen applicants, and most of them buy it from the same few vendors. So the question for African and Caribbean job seekers isn't really "Can I do this work from here?" anymore. It's "Can the machine at the door even see me?"
A run of research over the past year says: often, no. And it comes with numbers.
What the data shows
Researchers writing for Brookings ran almost 40,000 résumé-to-job comparisons through three widely used language models. They kept the content identical and changed only the names. The results are hard to read as anything other than a wall.
Résumés with white-sounding names were preferred 85.1% of the time. Black-sounding names led in 8.6% of cases. The two came out even just 6.3% of the time. Male names beat female names 51.9% to 11.1%. And the worst result of the lot landed on Black men: against an identical résumé with a white man's name, a Black man's name was picked 0% of the time.
Same skills, same experience, different name. That zero is what exclusion looks like once you stop dressing it up.
A team at Stanford HAI took the question out of the lab and into the actual job market, in what they call the first large-scale study of hiring algorithms in the wild. Using the U.S. EEOC's "four-fifths rule," which flags when one group gets recommended at under 80% of the top group's rate, they found plenty of roles where Black and Asian applicants were screened out at unfair rates.
When the whole market runs the same code
The researchers have a phrase for this: systemic rejection. The study behind it, "Algorithmic Monocultures in Hiring" by Bommasani, Bana, Creel, Jurafsky, and Liang, names the mechanism too. They call it algorithmic monoculture. Because employers screen with tools from the same handful of vendors, the same people, and the same racial groups, get rejected everywhere at once.
They tracked roughly 3 million applicants across 4 million applications. About 25.87% of Black applicants and 14.74% of Asian applicants applied to jobs where the algorithm worked against their group. Here's the part that stuck with me: among people who applied to 10 positions, 4% were rejected by all ten. That's more than chance would give you. You're not getting unlucky across fifty applications. You might be meeting the same algorithm fifty times, and it keeps making the same call. The authors' own advice is bleak in its way: apply as widely as you can, just to get a human to look at you.
The bias doesn't stop when a person takes over
You'd hope a recruiter reading the shortlist would catch this. A study from the University of Washington suggests the opposite. Kyra Wilson, a doctoral researcher, led an experiment where 528 people screened candidates across 1,526 scenarios. When the AI was badly biased, people went along with it about 90% of the time. Wilson's point: unless the bias was obvious, people were happy to take the machine's word for it.
So the algorithm doesn't just filter. It coaches. It hands a human a ranked list, and the human, trusting the "objective" tool, signs off on the discrimination as their own judgment. One hopeful detail: when people did a short bias-awareness exercise first, biased picks dropped 13%. Awareness helps. Left alone, though, we tend to inherit the machine's prejudice.
Why this hits us twice
Most of those studies measure bias against names inside Western job markets. For us, two more layers pile on top.
First, the training data was never built for our credentials. These models learned what a "good" candidate looks like from histories where companies rarely hired anyone from Accra or Castries. A journalist at TechCabal tested this, asking an AI to rank five equally qualified people for remote roles. "Tunde Afolabi" from Lagos came 4th of 5. "Sophia Smith" from San Francisco came first, and the model actually cited her "proximity to a tech hub" as a plus, for a fully remote job. Run against a real posting, Tunde again came in below the Western and Latin American applicants. A University of the West Indies degree, a solid regional certification, English shaped by local usage: the model often can't read these as signs of quality, because nothing taught it to.
Second, there's the location filter that kills the search before it starts. Plenty of jobs advertised as "remote" are only remote inside one country or region. A quiet location field disqualifies you in milliseconds, and no human ever learns you applied. Lawyers in South Africa have warned that a CV screener trained abroad can reject local candidates simply because their CV's language and layout don't match metrics built for somewhere else. It works like indirect discrimination even when nobody meant any harm.
And when the door does open, the deal can be crooked. One African remote worker put it plainly to TechCabal: "When they come to Africa, they try to hire the best talents at the lowest price." A product designer making about $1,500 a month in Lagos may be doing work that pays $10,000 in New York. That gap is real, and I won't pretend otherwise. But it argues for going into the global market with your eyes open, not for staying out of it.
The trouble with "objective" hiring
The whole sales pitch for automated hiring was that it would scrub out human bias and find the best person on merit. The research keeps showing the reverse. It launders old bias into a new form that's harder to spot and nearly impossible to appeal.
A recruiter who overlooks a Caribbean CV can be questioned, argued with, given context. A score can't. And when the whole market runs on the same few engines, there's no shopping around, because the next employer is quietly using the same one. This is why I keep landing on a people-over-technology view of work. The tools should widen the door for capable people. Used carelessly, they narrow it while claiming to be neutral.
What you can actually do Monday morning
This isn't a counsel of despair. It's a map, and knowing where the traps sit is already an edge. A few moves genuinely shift the odds.
The first gate is usually keyword matching, so beat the parser without lying to it. Mirror the job description's exact wording wherever it honestly fits. If they say "stakeholder management," use that phrase. Keep a clean, single-column CV that machines can read, and call your tools by their standard names.
Then give a human something a score can't flatten. A live portfolio, a public GitHub, a two-minute Loom walkthrough, a written case study. A ranking model can rank you low; it's much harder for a person to ignore work they can actually see.
Better still, go around the gate entirely. A warm referral often skips the automated screen completely, and that's the strongest argument I know for putting real time into community and networking. A connection does what no keyword can. It puts your file in front of a person from the start.
It also pays to be picky about where you apply. Async-first, genuinely distributed firms have usually stripped out the location filters and tuned their process for international talent. Build a short list of those and go deep, rather than spraying applications at companies that are remote in name only.
Last thing, and I can't say this strongly enough: the job search is itself a skill. The people who get through usually aren't the most gifted in the pile. They're the ones who figured out the game and prepared for it. That can be learned, and at this point it isn't optional.
The bottom line
AI hiring tools aren't going anywhere, and used well they can surface someone a swamped recruiter would have missed. I want to be fair about that. But the evidence on how they work as built is hard to wave away: they put Black applicants, women, and talent outside the Western default at a disadvantage, and because the market has standardized on a few of them, one rejection can follow you everywhere. The humans downstream mostly inherit the bias rather than fix it.
Remote work was supposed to make the world flatter. The least we can do, as workers and as a community, is refuse to let the machines at the door quietly tilt it back.
