The 500M Profile Study: How Work Really Works

Over a three-month study we analyzed over 500 million professional profiles to understand hiring, jobs, education, relationships, and professional development more broadly. This report covers the data, the methods, and what we found.

Data methodology

The dataset combines three sources:

1. Public professional signals. Openly available information about what people do, where they have worked, and how they present themselves.
2. Activity data. Behavioral signals over time, showing how people move and engage rather than only how they appear on paper.
3. Proprietary data. Our own first-party data, used to ground and validate the picture the public sources suggest.

Each source is partial on its own, so we cross-check across all three and weight the places where they agree. Before analysis, records were de-duplicated, normalized into a common shape, and cleaned of obvious noise.

Technical analysis methodology

Each profile is represented as a high-dimensional vector: a numerical summary of what a person does and how they show up. Two complementary approaches turn that into structural insight:

• Vectors, embeddings, and language models to cluster people and interpret the content of profiles at scale.
• Traditional machine learning to measure similarity between profiles and to run graph analysis across their relationships.

Together these let us run structural analysis across the full dataset. The study ran five lines of analysis, each detailed below.

Groups, relationships, and clusters

People do not sit in one flat network. They form dense communities held together by a few highly connected hubs, linked to other communities by a thinner set of bridges. What matters is not how many connections a person has but how clustered those connections are: a tightly linked cluster behaves like a shared antenna, where opportunity that reaches one member quickly reaches the whole group.

Most of those ties form through a handful of familiar channels. The majority trace back to shared history, education, and the workplace, with in-person business settings and online interactions adding the rest.

Position inside that structure beats seniority or credentials as a predictor of opportunity flow:

• Bridges see the most. People spanning two clusters get the widest range of opportunities.
• Density beats distance. Most roles were found within 2 to 3 degrees of connection, so one well-placed relationship outweighs dozens of distant ones.
• Ties need not be local or formal. Strong online relationships can transcend educational background entirely.
• Visibility compounds. A small share of people, roughly 1 in 10, with a strong public profile captured an outsized portion of inbound interest.

Career progression and experience

Careers rarely move in a straight line. The fastest-progressing people reached senior roles through lateral moves and cross-domain jumps, not a single ladder. Roughly 6 in 10 significant advances followed a non-linear step: a move sideways into an adjacent function, or a jump into a new domain that reused existing skills.

What carries people across those jumps is rarely the job title. When someone moved into a new domain, the through-line was almost always a set of transferable capabilities that travelled with them, not the specific role they were leaving. Titles describe where a person has been; portable skills describe where they can go next.

Education and work interleave rather than run in sequence. On-the-job and formal learning reinforce each other, and the people who kept adding capabilities mid-career compounded their options over time. Those who specialized early and stopped learning tended to plateau, even when their initial credentials were strong, because the market moved and their signal did not move with it.

Hurdles, communication, and work patterns

Progression stalls for structural reasons, not personal ones. The most common blockers are a thin or closed network, unclear signaling of what someone can do, and limited visibility into where demand is moving. Skill and credential gaps appear too, but rank below the network and visibility problems.

Two failures stood out as especially costly:

• Misdirected effort. Many people stack courses, licenses, and certificates that do little for their marketability, because they misread what actually moves hiring decisions.
• Imperfect information. Without a clear view of the going rate or the alternatives, people negotiate from weakness and accept worse roles and pay than their profile would command.

Communication and rhythm matter as much as the hurdles themselves. People who communicate clearly and collaborate across group boundaries moved more easily than equally skilled people who stayed siloed. The effect held regardless of work pattern: what separated fast movers from stalled ones was not remote versus on-site, but whether they reached beyond their immediate group. Work pattern changed the mechanics of how people connect, not whether connecting paid off.

Occupation and geographical trends

Where work concentrates is splitting along economic lines.

• Dispersing: online, remote, data, and gig work is expanding fastest in emerging economies, where people are taking up digital opportunities that were previously out of reach and building careers around them.
• Concentrating: in-person and physical work is moving the opposite way, showing a very acute recent shift into a small number of pockets in advanced economies rather than spreading out.

The result is two diverging maps: one dispersing across borders online, one concentrating into a few high-cost centers.

Demand is also moving between occupations, not just between places. The roles that grew fastest were hybrid ones, combining a technical core with people-facing or judgment-heavy work that is hard to automate or offshore. Narrowly defined single-skill roles grew slowest, squeezed from both sides: cheap to automate, and easy to source anywhere.

The two shifts compound. As routine work disperses to wherever it is cheapest and automatable work thins out, the durable roles are those that combine skills, sit close to real-world constraints, or require trust that cannot be sourced remotely.

AI impact

A clear recent shift in the data is how people present themselves. AI-assisted resumes, skill listings, and project write-ups grew from uncommon to more than half of new profiles over several quarters, and that share is still climbing.

This corrodes what a profile is worth as evidence:

• Harder to tell real from fake. When anyone can generate a fluent resume, it is increasingly difficult to know which claims reflect genuine capability.
• More exaggeration. Profiles are more polished and overstated than before, with achievements inflated to match what a model can plausibly write.
• Wording stops distinguishing. As a clean, well-written profile becomes the default, it no longer separates strong candidates from weak ones.

The weight therefore shifts toward proof that is hard to fabricate: verified work, demonstrated projects, and real relationships. As generated content becomes the default, trustworthy evidence becomes the scarce resource.

AI is also reshaping the kind of work on offer, and not only by displacing it. Lower-value online gig work has thinned out, with much of that activity moving toward AI and data-related tasks. At the same time, demand for some human services rose: as more goods and output are labeled as AI-made, "made by humans" has become a premium people will pay for.

But the impact is far narrower than the headlines suggest. In the data, AI has made a real dent in only two areas so far:

• Software, where much routine work is now assisted or automated.
• Freelance and online work, where lower-value tasks are quickest to be replaced.

Everywhere else the effect is faint. Physical work and regulation-intensive roles have barely been touched, insulated by hands-on requirements, licensing, and compliance that generated output cannot satisfy.

What this means

The findings point in a consistent direction for the people who act on them:

• For those looking for work: invest in position and proof, not just credentials. A single well-placed relationship, a demonstrated project, and a legible profile outperform another certificate. Being findable within 2 to 3 degrees of the right people beats applying into the void.
• For those hiring: as generated content floods profiles, lean on signals that are hard to fabricate. Verified work and real references separate genuine capability from polish, and looking one or two connections beyond the obvious pool surfaces stronger, less-contested candidates.
• For educators and institutions: teach transferable capability and help people signal it credibly. The market rewards skills that travel across domains and penalizes narrow specialization that cannot adapt.

The common thread is that structure and trust are becoming the scarce resources. Access to the right network and the ability to prove what is real now matter more than the presentation anyone can now manufacture.

The through-line

Across these lines of analysis, one pattern recurs: the structure a person sits in, their cluster, their position, their reach, shapes outcomes more consistently than any single attribute on their profile. As AI makes stated credentials easy to generate, where a person sits in the network only matters more.

This report was produced by Kariaa Research. All data, resources, and analysis are proprietary. For questions, contact research@kariaa.com.