Your Career Test Isn't Wrong. It's Answering a Question the Market Never Asked.

A career assessment test measures preference. Your next career move is decided by demand. Those are different variables, and no amount of accuracy on the first one produces an answer to the second. The test can be psychometrically sound and still leave you with nothing to do on Monday.

Most people get this backwards. They treat a low-confidence result as the reason they're stuck, so they take another quiz, hoping a sharper reading of their personality will finally reveal the job. But preference was rarely the binding constraint. You probably already know roughly what you like. What you don't know is which of those roles are being hired for right now, at what level, near you or remote, at what pay, and what the path in looks like from where you actually stand.

What does a career assessment test actually measure?

Interests. And it measures them better than critics usually admit, so let's be precise before we're skeptical.

John Holland's RIASEC model (Realistic, Investigative, Artistic, Social, Enterprising, Conventional) is the backbone of nearly every serious career assessment, including the Strong Interest Inventory and the US government's own instrument. Fifty years of data sit behind it, and the report commissioned to update O*NET's interest model notes that interests have been linked to job performance and job satisfaction across a series of recent meta-analyses. In a review of the field, James Rounds and Rong Su report that a meta-analysis of seventy years of studies found interest-scale scores correlated .20 with job performance, and interest congruence, the fit between a person and their environment, correlated .36. The same review reports congruence correlating .34 with persistence in school and .36 with persistence in jobs. A separate meta-analysis of 74 studies found single interest scales predicting job performance at .14 and training performance at .26, with a negative relationship to turnover.

Those aren't trivial numbers. A .36 correlation with performance is in the same range as the personality predictors that industrial psychologists take seriously. And there's a genuinely striking finding in the same review: analyzing 400,000 high school students followed up eleven years later, interests accounted for 83.3% of the explained variance in income, against 12.0% for cognitive ability and 4.7% for personality (the three together explained about a quarter of income variance overall). Interests are also more stable than personality across the life span.

So the fair verdict is this. Interest inventories do two things well: they generate options you hadn't considered, and they give vocabulary to a vague feeling. That's real value, and it's why a Holland code career test is a reasonable first hour. It is not a plan.

So why does the result feel useless?

Because every number above describes fit, and fit is a statement about you and a hypothetical environment. It says nothing about whether that environment is currently hiring.

Take the classic output: "you're Investigative-Artistic, consider archivist." Archivist is a real job with a real Holland code. It is also, according to O*NET's occupational data, a job held by about 9,300 people in the entire United States with roughly 1,100 projected annual openings. Software developers, an Investigative-Conventional profile, number about 1.69 million with roughly 115,200 projected annual openings. Two occupations, both defensible outputs of an interest test, separated by a factor of a hundred in how often the door actually opens.

The test doesn't weight for that, because it isn't built to. The BLS employment projections for 2025-2035 are a separate dataset entirely, so a quiz that hands you a list of titles has quietly left the hardest part of the decision to you. That's the same error behind the advice to chase a feeling before you've built anything: it treats the internal signal as the whole problem.

Is the MBTI any good for career decisions?

This is the one to be genuinely skeptical about, and the most persuasive critic is the publisher.

The Myers & Briggs Foundation states plainly that "the MBTI assessment is not designed to be used for hiring purposes" and that "it does not measure ability, competence, or skill." That's the people who sell it, telling you it isn't a selection or placement instrument.

The reliability data explains why. In a review in Consulting Psychology Journal, David Pittenger reports that across a five-week retest interval, 50% of participants received a different classification on one or more of the scales, and that the MBTI manual itself reported 35% of individuals getting a different four-letter type after four weeks. The same paper notes that the bimodal score distribution the type theory predicts is "conspicuously absent," meaning most people sit near the midpoint and get sorted into a category by a coin-flip margin. Pittenger also quotes the 1991 National Research Council committee behind In the Mind's Eye: "At this time, there is not sufficient, well-designed research to justify the use of the MBTI in career counseling programs."

Be careful what this does and doesn't say. Type language is not worthless for self-reflection or team conversation. It is simply too unstable to carry a career decision, and the publisher agrees it was never built to.

Why is every free career quiz trying to sell you something?

Because the quiz is the funnel, not the product. Worth naming, because it shapes what the free tier optimizes for: a result satisfying enough to share and vague enough to need upgrading.

16Personalities, one of the highest-traffic type tests on the internet, is explicit about it. The free test is free, and the site's own Premium Career Suite page prices the upgrade at €29 and tells you the free result shows you around 5% of the available insights. That's not a scandal, it's a business model. But it means the incentive behind the free career quiz that ranks first for your search is engagement, not accuracy about the labor market, and certainly not telling you that the role it just suggested has 1,100 openings a year.

There is a clean alternative if you want the interest data without the funnel. The O*NET Interest Profiler is published by the US Department of Labor, runs 30 or 60 questions, is free, is explicitly built on Holland's RIASEC structure, and links your result to over 900 occupations with real wage and outlook data attached. Take that one, take twenty minutes, and stop paying for type codes.

What would a useful assessment actually produce?

Three things, none of which a quiz can generate on its own: a shortlist tested against live demand, an honest read of the gap between your evidence and what the postings ask for, and the cheapest test that would kill the idea inside a few weeks.

What the quiz gives you	What you actually need to decide
A type code or three-letter Holland code	A shortlist of 5 to 8 specific roles, checked against postings live this week
A list of job titles	Volume: how many of those roles were posted in your market in the last 30 days
"Consider: archivist, curator, librarian"	Level and pay: which seniority bands are open, and what they pay where you are
A description of your preferences	The gap: what the postings ask for that your evidence doesn't yet show
A 40-page PDF report	One cheap experiment that could kill the idea in three weeks
Validation that you're an interesting person	A decision you can act on this month

The gap row is where most pivots quietly fail. A test tells you a role fits your interests; it cannot tell you that the postings for it all ask for two years of something you've never done, which is the difference between an appealing idea and a reachable one. That gap is also why half your transferable skills don't transfer, and why the pivots that work tend to be the boring adjacent ones rather than full reinventions.

The experiment row matters too, because good career counseling has never actually been the test. In the meta-analytic literature, the five critical ingredients of effective career choice counseling are workbooks or written exercises, individual feedback, world-of-work information, modeling, and building support for the decision. Note the third one. World-of-work information is an ingredient the research says does the work, and it's precisely the ingredient a free quiz supplies least of.

What's the cheapest test that could prove you wrong?

Once you have a shortlist, stop researching and start falsifying. The goal is not to confirm the role suits you. It's to find out fast whether it doesn't.

Weak version: "The assessment said I'd suit UX research, which makes sense because I like talking to people and analyzing things. I'm going to do a six-month certificate, then start applying."

Strong version: "The assessment put UX research on my list. There were 40 postings in my city last month; 31 wanted 3+ years and a portfolio of at least two end-to-end studies. I have zero. So before I spend anything, I'm running one unpaid research study for a friend's product this month, and I've asked two UX researchers for 20 minutes each to tell me whether that portfolio piece would clear their bar. If both say no and the postings don't move, I drop it by November."

The weak version buys a credential to settle an uncertainty a conversation would have settled for free. The strong version treats the shortlist as a hypothesis with an expiry date. Three falsifying tests cost almost nothing: a conversation with someone doing the job, a small real project that produces evidence, or a contract piece that pays you before you commit.

Do this before the sunk costs start, not after. The reason people cling to an expensive retraining decision is the same reason they stay in a career because of the ten years already spent: once you've paid, the test stops being an experiment and becomes an identity to defend.

What to do now

1. Take the free government one, once. The O*NET Interest Profiler takes 20 minutes, costs nothing, and hands you occupations with wage and outlook data attached. Then stop taking tests. A second quiz will not add information.
2. Turn the output into a shortlist of 5 to 8 specific roles. Not fields, not "something creative." Actual job titles you could type into a search box.
3. Run each title against live postings. Count how many appeared in your market in the last 30 days, note the seniority band and the pay range, and search live listings rather than trusting a static occupation description. A title with four postings a month in your city is not an option, whatever your Holland code says.
4. Write the gap down honestly. For each shortlisted role, list what the postings ask for and what evidence you can actually show today. The distance between those two columns is your real project, not the test result.
5. Pick the cheapest falsifying test and set a date. One conversation, one small project, or one contract piece, with a deadline by which you either have evidence or drop the role. Weeks, not semesters.
6. Re-run the demand check before you spend money. Markets move. Before committing to a course or a credential, recount the postings. The number, not the report, should unlock the spend.

Got a type code and no plan? Message Praxy on WhatsApp. Praxy will turn the result into a shortlist and check it against 3M+ live postings from 53 sources, so you find out which of those roles is actually being hired for this week, what the postings demand, and what the cheapest test of the idea would be.