The frameworks underneath
Tessera does not invent a theory of personality. It uses the two frameworks that career guidance has used for decades: Holland's RIASEC model of vocational interests, the same model behind the O*NET Interest Profiler that many schools already use, and the Big Five personality dimensions, the standard model in personality research.
RIASEC interests
Big Five personality
Whatever is novel about Tessera is in how the evidence is gathered, not in what it is measured against. The destination is the same as a counselor's inventory; the road is different.
Reading the two frameworks against each other is not novel either, and we would rather say so plainly. The correlations between RIASEC types and Big Five traits have been meta-analyzed twice over (Larson, Rottinghaus & Borgen, 2002; Barrick, Mount & Gupta, 2003), and pairing an interest inventory with a personality instrument is ordinary practice in career guidance. Barrick and colleagues state the useful conclusion directly: the two are related, but "they are not merely substitutes for each other." That is the whole reason Tessera measures both.
How much the second measure adds depends on which teen you are looking at, and that shapes what Tessera will and will not say. The link between interest and personality is uneven across the six types. It is strongest for Enterprising and Artistic, where Barrick and colleagues report multiple correlations of .47 and .42, so an Enterprising teen's interests already imply a fair amount about how they are wired. It is weakest for Realistic (.11, with no relation to any Big Five trait), Investigative (.26) and Conventional (.27), where the interest score tells you close to nothing about personality and measuring it separately is the only way to get it. The synergies and friction points Tessera raises are therefore better evidenced for some profiles than for others, which is part of why they are offered as reasoning a teen can push back on rather than as findings.
How much evidence does a picture need?
Validated short-form instruments are the useful benchmark here, because they are the published answer to exactly this question: how few items can you get away with?
| Instrument | Items | What it delivers |
|---|---|---|
| O*NET Interest Profiler, Short Form Rounds, Su, Lewis & Rivkin, 2010 |
60 10 per interest area |
Internal consistency about .81 to .86 |
| Mini-IP Rounds, Ming, Cao, Song & Lewis, 2016 |
30 5 per area |
Internal consistency .74 to .81; agrees with the 60-item form on the leading interest (Cohen's kappa .73), profile correlation about .92 to .95 |
| 18REST Ambiel et al., 2018; validated partly on high-schoolers, mean age 16 |
18 3 per area |
.68 to .81; correlates .87 to .94 with the full item pool |
| BFI-2-S Soto & John, 2017 |
30 6 per dimension |
About .77, retest .76 to .84, captures about 91% of the full instrument's variance |
| Mini-IPIP Donnellan et al., 2006 |
20 4 per dimension |
Acceptable, retest comparable to the 50-item parent measure |
| TIPI Gosling, Rentfrow & Swann, 2003 |
10 2 per dimension |
6-week retest averages .72; convergence with a full Big Five measure about .77 |
There is no published rule of thumb for a minimum number of items, and these instruments do not suggest a simple one: the TIPI gets usable retest and convergence evidence out of two items per dimension, while longer forms buy higher internal consistency and more detail. What the set does show is a trade-off. Shorter forms stay serviceable for a broad read and lose precision, facet-level detail, and internal consistency as they shrink.
Tessera's arc collects roughly 100 to 120 answers, split between the interest side and the personality side, so something like 50 to 60 answers for each. That is the same order of magnitude as the instruments above, and several times more than the quick ten-question quizzes teens usually meet online.
What the AI does, and what it does not do
This is where people are right to be careful, so it is worth being precise.
The AI writes the questions
It phrases them around things a particular teen actually cares about, in language that suits their age, so that a teen keeps answering. That is its main job, and it is the reason the conversation works at all where a 60-item checklist would be abandoned by item twelve.
The AI does not decide what an answer means
When a teen picks one of the offered options, which is how most answers are given, the interpretation is already fixed: every option carries a predetermined mapping to interest and personality dimensions, written into the app, and the scoring is done on the device by ordinary arithmetic. The model is not consulted about what the choice says about the teen. The same is true for the rapid-fire rounds.
No single answer carries much weight
The picture is an accumulation. One answer nudges it; it takes a consistent pattern across dozens of answers to move anything the teen is eventually told. An odd mood, a joke answer, or one bad day cannot rewrite the profile.
It is not a standardized test
And we do not present it as one. Tessera does not issue a score, a percentile, or a diagnosis. It describes patterns in what a teen has told it, in plain language, with the reasoning visible so the teen can disagree.
Counselors should treat it as a rich conversation starter that arrives with evidence, not as an instrument to file.
Why the answers are spread over weeks
Because spreading them out makes the result better, not just easier to swallow.
Fleeson (2001) had 46 students report on their own behavior five times a day for thirteen days. Any single report predicted another single report only weakly (correlations of .28 to .54): one moment is a poor guide to a person. But the average of about twenty-five reports matched the average of the other twenty-five almost exactly (.87 to .94). The person is in the average, not in any one answer.
This is the aggregation principle Epstein (1979) described: measurements averaged over many occasions are far more stable than any single occasion. It is also the single strongest argument for Tessera's shape.
A teenager answering forty questions in one sitting on a Tuesday afternoon gives you one mood, one day, one level of patience. The same forty questions spread over three weeks give you the teen, not the Tuesday.
Why three weeks and not three months
Because we do not believe most teens would get there.
Baumel, Muench, Edan & Kane (2019) looked at real-world usage data from 93 mental-health apps with over 10,000 installs each. Median retention fifteen days after install: 3.9%. Thirty days: 3.3%. This is not specific to Tessera; it is what voluntary app engagement looks like outside a paid research trial.
The gap between trial conditions and real life is stark. Cohen & Schleider (2022) ran single-session digital interventions with 2,436 adolescents: completion was 84.75% among paid trial participants and 36.86% among real-world users doing it on their own.
Neither study tested a three-week program, and neither one names an optimal length. We know of no research that does. What they establish is narrower and still decisive: voluntary engagement with a self-directed app falls away quickly, and completion rates measured in a paid trial badly overstate what happens in real life. A design that assumes months of consistent unpaid use is assuming something those numbers give no support for.
Three weeks of short sessions is our judgment call in light of that, not a finding we can cite. The five-minute session length is the same kind of call: it has to fit in the gaps of a teenager's day, or it does not happen. Both are worth revisiting against our own usage data as it accumulates, and we would rather say that plainly than dress a product decision up as science.
Why not two sessions and a summary
Because that produces a horoscope.
At the first milestone, around forty answers, there is enough to point at a leading interest and describe a working style. There is not enough to hand a teen a definitive profile and a career list. That is why the first milestone is presented as a first read and a direction, and why the picture is described as sharpening rather than arriving.
It is also why we ask for the commitment up front, on its own screen, in the teen's own words: a picture built from three chats is not worth much, and a teen who finds that out at the end feels tricked.
What the result is, and is not
Interests and personality at this age are genuinely still in motion, and no measurement schedule changes that:
- Low, Yoon, Roberts & Rounds (2005), a meta-analysis of 114 longitudinal studies drawn from 59 publications, found interest stability of about .55 at ages 12 to 14, .57 at 14 to 16, .58 at 16 to 18, rising to .67 at 18 to 22 and then holding steady for roughly two decades.
- Roberts & DelVecchio (2000), 152 studies, found personality rank-order consistency of about .47 at ages 12 to 18, .51 at 18 to 22, and higher after that.
- Interests are the more stable of the two at these ages, which is part of why the interest side leads the picture Tessera builds.
Read that as good news framed honestly: a teen's direction is real enough to act on and stable enough to plan around, but it is not a verdict on who they will be at twenty-five. Which is exactly why Tessera keeps talking to them after the arc is finished, and why the picture updates instead of being printed once.
What we are actually asking of a teen
Twelve short sessions
About five minutes each. Roughly four a week gets them through in about three weeks.
Slower is fine
Two a week means the arc takes six weeks. Nothing expires and nothing is lost; the only cost is time to the first real read.
Stopping halfway is the one bad outcome
Half the sessions give half the picture, and we would rather a teen know that before they start than discover it at the end.
Questions we get
"Can my teen just do it all in one afternoon?"
The app will not allow it: there is a daily cap. And per the aggregation research above, the result would be worse even if it did.
"They missed a week. Is it ruined?"
No. The picture keeps building from where it left off. A gap costs time, not progress.
"Will it tell them what to be?"
No, and it is not built to. It points toward kinds of work that fit how a teen is built, with reasoning they can read and push back on. The choosing stays theirs.
"What do I see as a parent or counselor?"
Only what the teen chooses to share, and only the insights, never the conversations themselves. Sharing is per-person and can be switched off at any time. There is more on this in For Parents, For Counselors, and the Privacy Policy.
Sources
- Ambiel, R. A. M., Hauck-Filho, N., Barros, L. O., Martins, G. H., Abrahams, L., & De Fruyt, F. (2018). Psicologia: Reflexao e Critica, 31:6.
- Barrick, M. R., Mount, M. K., & Gupta, R. (2003). Personnel Psychology, 56, 45-74.
- Baumel, A., Muench, F., Edan, S., & Kane, J. M. (2019). Journal of Medical Internet Research, 21(9):e14567.
- Cohen, K. A., & Schleider, J. L. (2022). Internet Interventions, 27.
- Donnellan, M. B., Oswald, F. L., Baird, B. M., & Lucas, R. E. (2006). Psychological Assessment, 18, 192-203.
- Epstein, S. (1979). Journal of Personality and Social Psychology, 37, 1097-1126.
- Fleeson, W. (2001). Journal of Personality and Social Psychology, 80, 1011-1027.
- Gosling, S. D., Rentfrow, P. J., & Swann, W. B. (2003). Journal of Research in Personality, 37, 504-528.
- Larson, L. M., Rottinghaus, P. J., & Borgen, F. H. (2002). Journal of Vocational Behavior, 61, 217-239.
- Low, K. S. D., Yoon, M., Roberts, B. W., & Rounds, J. (2005). Psychological Bulletin, 131, 713-737.
- Roberts, B. W., & DelVecchio, W. F. (2000). Psychological Bulletin, 126, 3-25.
- Rounds, J., Su, R., Lewis, P., & Rivkin, D. (2010). O*NET Interest Profiler Short Form Psychometric Characteristics. National Center for O*NET Development. onetcenter.org
- Rounds, J., Ming, C. W. J., Cao, M., Song, C., & Lewis, P. (2016). Development of an O*NET Mini Interest Profiler (Mini-IP). National Center for O*NET Development. onetcenter.org
- Soto, C. J., & John, O. P. (2017). Journal of Research in Personality, 68, 69-81.