How Folkso finds the person you'll work well with

Folkso brings you to the right person

Skills and a rate give you a list. Which person on it you'll enjoy working with for the next three months, the list doesn't say.

Folkso is built for that question. It finds people who do the work and checks them. Your AI compares them with the way you work and tells you who is closest. Both halves rest on research about people at work, and we measure both.

Suiting you best means fitting both the task and the way you work. The numbers come from 300 test searches whose right answers were known in advance and which took no part in tuning. They ran on the same code that answers you in ChatGPT and Claude.

Four steps from your request to the right person

1. Folkso finds people who do the work. Only people open to this kind of work make it through, and anyone whose own conditions rule out your task drops out. A model reads the 50 closest profiles in full, without contacts, and reads the first few again one by one before you see them.
2. Your AI gets the facts. For each person it sees a sentence from their own profile that shows they can do it, and something concrete you share, if there is one. It sees what doesn't match your conditions, such as a rate above your budget. And it gets a short portrait of how the person works and what they care about in work, estimated from the text of their profile. Only confident estimates make it into the portrait.
3. Your AI compares them with you. It has seen how you work, in your chats and in the work you've done together, and knows you better than any profile you'd fill in. It weighs pace and rhythm, whether someone works on their own or needs check-ins, how they argue and take feedback, and what matters to them in work. Sometimes the right person works like you. Sometimes it's the one who covers your weak spot, like an organised partner for someone who works in bursts. The AI writes its take on each card and can mark one person Closest.
4. You decide. The cards stay in Folkso's order. You read the takes and choose whom to write to.

A take starts from a fact in the profile and names the "but" when there is one. Compatibility percentages and "perfect fit" are left out. In user studies by Bansal and colleagues, an AI's explanations made people accept its suggestion more often, whether the suggestion was right or wrong. Name, photo, gender, age and origin play no part in the comparison.

Folkso doesn't read your chats or your assistant's memory. Only the conditions of the task go into the search, and if your AI wants to turn something it remembers about you into a condition, it asks you first.

The takes are seen only by you and deleted after 30 days. The person a take is about doesn't see it.

What decides whether two people work well together

Folkso's matching rests on meta-analyses and long studies of people at work. These are the ones it leans on.

Kristof-Brown, Zimmerman and Johnson combined 172 studies with 836 effect sizes on how people fit their job, their organization, their team and their manager. Fit went with being satisfied and wanting to stay. Fit with the manager correlated .46 with satisfaction with that manager, and fit with the team .47 with how close-knit the team was. The links with performance were weaker, .07 to .20. Team fit went with overall job satisfaction more weakly than the authors expected, and they offer one reason. Many of those studies measured it by comparing personality scores.

Liden, Wayne and Stilwell followed 166 new employees and their managers through their first six months together. What the two expected of each other in the first five days predicted how good their working relationship was at two and at six weeks. Similarity they saw in each other, and liking, predicted it at most points. Demographic similarity didn't.

Edwards and Cable compared the values of employees from 4 organizations with the values of their organization. Shared values went with job satisfaction and wanting to stay, mostly through trust in the organization and its people, and next through easier communication. Fit with one colleague wasn't measured there.

Time adds to this. The longer people work together, the less demographic differences matter and the more differences in attitudes do, in work groups (Harrison, Price and Bell) and in 144 student project teams followed over four waves (Harrison and colleagues).

Mentoring shows the same pattern. In a meta-analysis by Eby and colleagues, 173 samples with 40,737 people, how well protégés rated their mentoring went most closely with how similar they were to the mentor in attitudes, values, beliefs and personality. Among everything that came before the relationship, these were the strongest links, .38 to .59.

Couples are the most studied pairs of all, and even there similar personalities explain little. In three national samples of married couples, how alike the partners' personalities were added less than 0.5% to their satisfaction on top of each partner's own personality (Dyrenforth and colleagues). Across 43 studies of 11,196 couples, how each partner saw the relationship itself predicted up to 45% of its quality at the start of each study (Joel and colleagues).

So Folkso looks at the work, at what people expect, at what they value and at how they work. There are no personality types in it. Comparing all that with you is your AI's part, because it knows you.

How we test it

Folkso's lab runs 1,000 searches with right answers known in advance, on the same code that answers you. 700 of them tune the weights. The other 300 stay aside, and the numbers on this page come from them. Here is how the test works.

The strictest measure asks whether the single best person for you, out of 100 people in the test pool, stands exactly first. Folkso puts them first in 48.9% of searches. Random order after the filters manages 4.6%, so Folkso does it more than ten times as often. That's why your AI sees five people, and the best one is among them in 84.5% of searches.

On 21 October we checked the record against new right answers, where similar personalities count for little and values and expectations count for more. It held. Best person first stayed at 48.9%, and whether the first person can do the job doesn't depend on the new answers at all.

What we tried and dropped

So the order of results rests on what held up, in the studies and in our test.

Folkso reads what people expect from work

Some of what decides a yes is already written in the profile. "Writing over calls." "Projects of three months or longer." Folkso reads such lines, and on 100 test profiles it didn't read a single stated condition the wrong way round. When one clashes with your task, your AI sees it next to the person and tells you. Whether it matters is up to you.

What these numbers can't tell you

• The test runs on fictional people and searches. Real profiles and requests are shorter and messier, with typos and two languages in one text.
• Live outcomes are still few. Accepted requests and people set aside with a reason will show how often Folkso gets it right in real life.
• The studies above are correlations. They show what goes along with a good working relationship and can't prove what causes it.
• How the first week goes, only the two of you will find out. The choice stays with you.

Folkso can bring Leo to Mira's card and tell him why she's closest. Whether her first sketch lands at three in the morning, he'll see for himself.

Folkso is free. Connect it once, and the next time a task needs a person, your AI finds them right in the conversation.

Sources

• Kristof-Brown, Zimmerman, Johnson, Consequences of individuals' fit at work. Personnel Psychology 58(2), 281–342, 2005. Meta-analysis of 172 studies with 836 effect sizes.
• Liden, Wayne, Stilwell, A longitudinal study on the early development of leader-member exchanges. Journal of Applied Psychology 78(4), 662–674, 1993. 166 new employees and their supervisors over their first six months.
• Edwards, Cable, The value of value congruence. Journal of Applied Psychology 94(3), 654–677, 2009. Employees from 4 organizations.
• Harrison, Price, Bell, Beyond relational demography. Academy of Management Journal 41(1), 96–107, 1998. Work groups.
• Harrison, Price, Gavin, Florey, Time, teams, and task performance. Academy of Management Journal 45(5), 1029–1045, 2002. 144 student project teams, four waves of data.
• Eby et al., An interdisciplinary meta-analysis of the potential antecedents, correlates, and consequences of protégé perceptions of mentoring. Psychological Bulletin 139(2), 441–476, 2013. 173 samples, 40,737 people combined.
• Dyrenforth, Kashy, Donnellan, Lucas, Predicting relationship and life satisfaction from personality in nationally representative samples from three countries. Journal of Personality and Social Psychology 99(4), 690–702, 2010. Married couples from national samples in Australia (N = 5,278), the UK (N = 6,554) and Germany (N = 11,418).
• Joel et al., Machine learning uncovers the most robust self-report predictors of relationship quality across 43 longitudinal couples studies. PNAS 117(32), 19061–19071, 2020. 43 datasets from 29 labs, 11,196 couples.
• Bansal et al., Does the whole exceed its parts? The effect of AI explanations on complementary team performance. CHI 2021. User studies on three datasets.
• Folkso match lab, 300 held-out searches. Record of 27 September 2026, checked against new right answers on 21 October 2026. How the test works.