Advanced AI-driven MBTI personality test

[Carl-Richard]

Here is a ChatGPT o3 prompt I developed (together with ChatGPT o3) for typing your MBTI type based on your posts on this forum:

Quote

Using a purely academic understanding of MBTI (i.e., grounded in peer-reviewed literature rather than internet lore), estimate [insert forum username here]’s MBTI type from Actualized.org.

Methodological requirements

1. Sampling – State your sampling frame across years and sub-forums, then gather ≥ 100 distinct posts, avoiding topical clustering; list counts per year.

2. Linguistic corpus analysis – Provide a table showing mean sentence length, type–token ratio, jargon density, hedging frequency, metaphor density, and rhetorical-device counts per 1 000 words.

3. Evidence excerpts – Quote at least ten passages (≤ 40 words each) that best exemplify the patterns you discuss, with thread titles and dates.

Analytic instructions

4. Base your typing on recognised functional models; map observed behaviours to dominant/auxiliary functions.

5. Compute Bayesian posterior probabilities for the four most plausible types, explaining priors and likelihood cues.

6. Cross-reference traits with Big Five and HEXACO findings to triangulate your conclusion.

Bias & rigour

7. Explicitly list which cognitive biases (confirmation, anchoring, etc.) you checked for and how you mitigated them.

8. Cite at least three peer-reviewed MBTI validity/reliability papers and one corpus-linguistics study on personality-language links (APA style).

9. MBTI Misconceptions to Avoid

(Each item names a misunderstanding/trope [✗] and immediately replaces it with the scholarly corrective [✓]):

• ✗ Treating the four-letter code as a complete personality description → ✓ Use it only as a coarse preference map; report evidence for preference *strength*, not destiny.  

• ✗ Assuming type is immutable across the lifespan → ✓ Note test–retest drift (≈40 % change) and frame conclusions as probabilistic snapshots.  

• ✗ Equating “Extraversion” with loud sociability and “Introversion” with shyness → ✓ Define the axis in Jungian terms (energy orientation), then check behavioural evidence, not social anxiety cues.  

• ✗ Reducing “Thinking” to unemotional logic and “Feeling” to irrationality → ✓ Clarify that T/F index the *decision-making criteria* (impersonal logic vs. values), not presence/absence of emotion.  

• ✗ Reading “Judging” as judgmental/stubborn and “Perceiving” as open-minded/chaotic → ✓ Explain J/P as preferred *lifestyle structure* (decisive closure vs. adaptive exploration).  

• ✗ Mixing dichotomies with the eight cognitive-function stack without theoretical grounding → ✓ Keep dichotomy evidence separate from any function-level hypotheses, and justify any stack claims.  

• ✗ Treating type descriptions as stereotypical “career match” or “compatibility” horoscopes → ✓ Cite empirical findings only where effect sizes exist; otherwise mark as speculative.  

• ✗ Pathologising types (e.g., “INTPs lack empathy”, “ESFPs are shallow”) → ✓ Use neutral language; avoid clinical labels unless supported by peer-reviewed data.  

• ✗ Over-mapping MBTI letters onto the Big Five (e.g., “N ≙ Openness”) → ✓ Acknowledge areas of overlap but keep construct boundaries explicit; reference cross-inventory correlations.  

• ✗ Ignoring cultural/linguistic variation in type expression → ✓ Specify that behavioural indicators may shift with sociolinguistic context, so weight linguistic cues accordingly.  

• ✗ Cherry-picking quotes that confirm a favoured type → ✓ Actively sample for disconfirming evidence; document selection criteria and coder reliability.  

Output format

10. Organise your answer under numbered headings: Methodology, Linguistic Analysis, Typological Reasoning, Results Table, Bias Checks, Limitations & Ethics, Future Work.

11. Close with practical next steps (e.g., automated LIWC pass, direct MBTI Form M) to improve estimate reliability.

Keep the tone scholarly, precise, and transparent. Avoid anecdotal claims, refrain from definite diagnoses, and include an ethics note on typing individuals without explicit consent.

 

Why is this good? Because it avoids the biases associated with self-report, the bane of all self-administered personality tests. Now you can get a highly data-driven, third person testing of your personality type with just the click of a button.

Feel free to feed this prompt to ChatGPT o3. Also, somebody try toggling the deep research option and increase the sample size by 10x (simply 10x all the relevant numbers in the prompt). Please share your results and your thoughts. Also feel free to provide your own prompts and develop the promps further by feeding it back to ChatGPT and cleverly asking it for help on how to improve it.

Edited Tuesday at 03:23 AM by Carl-Richard

[UnbornTao]

Cool.

Quote

Thus INTJ emerges strongest (~60%), followed by INTP (~20%), INFJ (~15%), and INFP (~5%). These reflect our reading of the evidence: UnbornTao’s style is most consistent with INTJ (dominant Ni, auxiliary Te). We stress that these probabilities are illustrative, not definitive diagnoses.

Usually it's INTP, though. But interesting. 

Edited Monday at 02:44 PM by UnbornTao

[Joshe]

Cool idea. One problem I ran into when working with ChatGPT and MBTI is, ChatGPT’s MBTI knowledge is too intertwined with internet lore - it hasn’t been sufficiently trained on MBTI. Your prompt would probably produce mostly accurate results but ChatGPT would  still make critical errors at a significant rate.

Also, if it reads a post where someone claims to be ENTP, it might just lazily conclude that claim is true. Getting the prompt right is difficult. 

Edited Monday at 02:38 PM by Joshe

[Carl-Richard]

52 minutes ago, Joshe said:

Cool idea. One problem I ran into when working with ChatGPT and MBTI is, ChatGPT’s MBTI knowledge is too intertwined with internet lore - it hasn’t been sufficiently trained on MBTI. Your prompt would probably produce mostly accurate results but ChatGPT would  still make critical errors at a significant rate.

The eternal problem with generalized LLMs. A possible counter to that other than the "only use academic literature" prompt would be to feed it a list of common inaccurate internet tropes that it should avoid.

Here is what I will update the original prompt with (could be developed further):

Quote

### MBTI Misconceptions to Avoid

• ✗ Treating the four-letter code as a complete personality description → ✓ Use it only as a coarse preference map; report evidence for preference *strength*, not destiny.  

• ✗ Assuming type is immutable across the lifespan → ✓ Note test–retest drift (≈40 % change) and frame conclusions as probabilistic snapshots.  

• ✗ Equating “Extraversion” with loud sociability and “Introversion” with shyness → ✓ Define the axis in Jungian terms (energy orientation), then check behavioural evidence, not social anxiety cues.  

• ✗ Reducing “Thinking” to unemotional logic and “Feeling” to irrationality → ✓ Clarify that T/F index the *decision-making criteria* (impersonal logic vs. values), not presence/absence of emotion.  

• ✗ Reading “Judging” as judgmental/stubborn and “Perceiving” as open-minded/chaotic → ✓ Explain J/P as preferred *lifestyle structure* (decisive closure vs. adaptive exploration).  

• ✗ Mixing dichotomies with the eight cognitive-function stack without theoretical grounding → ✓ Keep dichotomy evidence separate from any function-level hypotheses, and justify any stack claims.  

• ✗ Treating type descriptions as stereotypical “career match” or “compatibility” horoscopes → ✓ Cite empirical findings only where effect sizes exist; otherwise mark as speculative.  

• ✗ Pathologising types (e.g., “INTPs lack empathy”, “ESFPs are shallow”) → ✓ Use neutral language; avoid clinical labels unless supported by peer-reviewed data.  

• ✗ Over-mapping MBTI letters onto the Big Five (e.g., “N ≙ Openness”) → ✓ Acknowledge areas of overlap but keep construct boundaries explicit; reference cross-inventory correlations.  

• ✗ Ignoring cultural/linguistic variation in type expression → ✓ Specify that behavioural indicators may shift with sociolinguistic context, so weight linguistic cues accordingly.  

• ✗ Cherry-picking quotes that confirm a favoured type → ✓ Actively sample for disconfirming evidence; document selection criteria and coder reliability.  

 

52 minutes ago, Joshe said:

Also, if it reads a post where someone claims to be ENTP, it might just lazily conclude that claim is true. Getting the prompt right is difficult. 

"Avoid anecdotal claims" was supposed to counter this. Who knows if it actually works though lol

Edited Monday at 03:33 PM by Carl-Richard

[Carl-Richard]

My newest results based on adding the misconceptions section:

Quote

| Rank | Type | Posterior P | Dominant / Aux | Notes |

| ---- | -------- | ----------: | -------------- | ------------------------------------------------------------------------------- |

| 1 | **INTP** | 0.41 | Ti – Ne | Highest Ti lexical indicators; low social-affect markers |

| 2 | ENTP | 0.24 | Ne – Ti | Higher posting frequency could suggest Extraversion, but Ti>Ne weights disagree |

| 3 | INTJ | 0.17 | Ni – Te | System-building evident but Te evidence weaker than Ti |

| 4 | INFP | 0.15 | Fi – Ne | Hedging & metaphor compatible; logical phrasing argues against Fi-lead |

 

I still think I'm more INFP than INTP but maybe my writing is more INTP. Maybe I could do some data-driven dialectics where I feed it my own description of myself and my own type estimation and make it single out places where its analysis might be inaccurate and how the prompt can be improved to address this.

It could also probably be beneficial after new additions to ask it to streamline the prompt and remove repetitions to improve processing.

Edited Monday at 03:58 PM by Carl-Richard

[Joshe]

I noticed the larger and more complex a set of instructions, the more it will fail for any given instruction. AI seems to be good at adhering to a few complex instructions, but once you add more than 5 or so, it starts to fail, IME. 

I wonder if a better approach is to instead of letting AI use it's own knowledge, define heuristics for each type and have it check against that. 

Something like: 

Dominant characteristics of an INTP:

• Internal logic focus – prioritizes internal consistency and logical coherence over external rules
• System-builder mindset – loves building internal frameworks, models, or taxonomies
• Truth over efficiency – values what is logically correct, not necessarily what’s useful or practical
• Detaches from emotion – separates logic from personal feeling to maintain objectivity
• Highly analytical – breaks down ideas to their core components
• Skeptical and critical – naturally questions assumptions, especially broad generalizations
• Precision-driven – prefers clarity, specificity, and refined definitions
• Slow to speak, quick to refine – may take longer to reach conclusions but seeks intellectual purity
• Personal logic – doesn’t blindly accept external systems (like laws, traditions, or consensus)
• Private problem-solving – works things out internally, often without expressing the full process aloud

But this too would probably be too comprehensive for AI to manage well. 

Now I'm thinking, come up with heuristics for how each type tends to communicate, because that's essentially what is being tracked, and that would be much simpler for AI to manage. Something like:

So for example, if someone is often talking about the future or doing things now for the future and if they don't soften their critiques, and if they're usually very direct, this would point to an INTJ.

I think simple heuristics like this could work with a large enough sample size, and it probably wouldn't even have to be that large. 

So if someone has this communication pattern: oftens clarifies terms, often questions assumptions, and is less concerned with social niceties, it's a good bet they're INTP. 

I'm thinking 5 communication patterns per type should suffice. Maybe even add style and trait heuristics as well, but maybe would have to be careful not to overwhelm the AI.

Maybe what types of conversations they typically engage in, are they more scientific or spiritual, abstract vs concrete, degree of politeness, what topics they're passionate about and what they tend to defend, how careful do they write, etc. I think heuristics like this could be very efficient if used well. Like, if someone doesn't give a fuck about using proper grammar, that simple heuristic could be high-leverage. 

I bet there are some very specific high-leverage patterns you could flesh out so that two or 3 alone could get you most of the way. This would be very efficient, but maybe too optimistic. 

Also, a 2-step process seems good. Maybe first have it run your initial prompt and then use the results from that to confirm with the heuristics analysis.

Edited Monday at 06:42 PM by Joshe

[Carl-Richard]

Below is a **re-engineered version of your original prompt**.
It keeps every substantive requirement you listed but applies the five fidelity aids we discussed:

* **Chunk & nest** → Major sections and sub-bullets
* **Explicit priorities** → Numbered “override order” up front
* **Self-check** → Checklist the model must output before final answer
* **Two-turn structure** → You send Part 1 (spec) first, then Part 2 (question)
* **Clarification invitation** → Model must ask if anything is unclear

---

### 🚩 PART 1 — SPECIFICATION (READ-ONLY)

*The assistant should not answer yet; it should only acknowledge understanding or ask clarifying questions.*

#### 0. Override-Order (Priority Rules)

1. **OpenAI policy & safety**
2. **This specification**
3. **Any future user follow-ups**

---

#### 1. Context

You will analyse the forum contributions of *Carl-Richard* (Actualized.org) to estimate his probabilistic MBTI type using peer-reviewed standards only.

---

#### 2. Deliverables & Format

Provide your answer under **exactly seven numbered H2 headings**:

1. **Methodology**
2. **Linguistic Analysis**
3. **Evidence Excerpts**
4. **Typological Reasoning**
5. **Results Table**
6. **Bias Checks**
7. **Limitations & Ethics**

*(Do **not** add or remove headings.)*

---

#### 3. Core Requirements (“Must-haves”)

**A. Sampling**

* State sampling frame and page-interval method.
* Collect ≥ 100 unique posts; list count per calendar year.

**B. Linguistic Corpus Metrics**

* Compute per-1 000-word rates for:

  * mean sentence length
  * type–token ratio
  * jargon density
  * hedging frequency
  * metaphor density
  * rhetorical-device count

**C. Evidence Excerpts**

* Quote ≥ 10 passages (≤ 40 words each) with thread titles & dates.

**D. Typology**

* Map observed behaviours to cognitive functions using recognised functional models.

**E. Bayesian Inference**

* Show priors, likelihood cues, and posterior probabilities for the **four** most plausible types.

**F. Trait Triangulation**

* Cross-reference findings with Big Five **and** HEXACO literature.

**G. Bias & Rigour**

* List specific cognitive biases checked + mitigation steps.

**H. Citations**

* Minimum: three MBTI validity papers **and** one personality-language corpus study (APA style).

**I. Misconception Guardrails**

* For each of the 11 misconceptions supplied, include the ✗/✓ bullet exactly as provided.

**J. Closing**

* End with practical next steps to improve reliability (e.g., LIWC pass, Form M).

---

#### 4. Secondary Style Guidelines (“Nice-to-haves”)

* Keep tone scholarly and neutral.
* Avoid anecdotal claims and pathologising language.
* Refrain from definite diagnoses; phrase in probabilistic terms.
* Use tables **only** where they improve clarity (metrics & results).

---

#### 5. Self-Check Checklist

At the very end of your response, output a short checklist confirming you have:

| # | Item                           | Yes/No |
| - | ------------------------------ | ------ |
| 1 | ≥ 100 posts sampled            |        |
| 2 | Metrics table present          |        |
| 3 | 10+ excerpts with metadata     |        |
| 4 | Bayesian table (4 types)       |        |
| 5 | ≥ 4 scholarly citations        |        |
| 6 | Misconception bullets included |        |
| 7 | Ethics note present            |        |

*(Populate the “Yes/No” column yourself before sending the final answer.)*

---

#### 6. Clarification Protocol

If **any** part of this specification is ambiguous, ask follow-up questions **before** attempting the task.

---

### ➡️ PART 2 — QUESTION / GO-AHEAD

*(You will send this as a separate message after the assistant acknowledges Part 1.)*

> “Proceed with the MBTI estimation of Carl-Richard per the above specification.”

---

### How to use

1. **Send the text above (Part 1) as a single system or user message.**
2. Wait for the assistant to say something like *“I understand. Do you have any clarifications?”*
3. Send Part 2 to trigger the full analysis.

This structure should maximise compliance with every requirement while giving the model explicit leeway to ask for clarification if needed and a built-in self-audit step.

@Joshe I got ChatGPT o3 to tell me how to best make my prompt to maximize things like readability and I made it revise my prompt based on those criteria, and this is what it fed me.

 

Edited Monday at 08:49 PM by Carl-Richard

[Joshe]

1 hour ago, Carl-Richard said:

I got ChatGPT o3 to tell me how to best make my prompt to maximize things like readability and I made it revise my prompt based on those criteria, and this is what it fed me.

Nice! Your prompt is too scientific for my taste. 😆

I actually just gave the deep researcher my profile link and told it to analyze 200 posts/comments to figure out my type based on cognitive functioning and it did really well. The only rule I gave it was to ignore any post about where the user mentions their own personality traits and ignore any MBTI claims the user has made. It seems to work pretty good out of the box by having it focus specifically on cognitive functions instead of MBTI. Still not sure how accurate it is for other people though. 

I just tested i on @Nemra. Nemra, ChatGPT analyzed your posts and says you're an INTP. Here's the results. 

https://chatgpt.com/share/6862fb68-6880-8010-9944-596af7ca8ec1

 

[Natasha Tori Maru]

@Carl-Richard

As always, a fascinating and useful tool to contribute - thank you

@Joshe

You may be interested to know this prompt pulled INTJ for me - which was your guess in previous discussions. INFJ, ENFP, INFP where next, in that order. I definitely feel there are some inaccuracies due to the data ChatGPT pulls from. I will add a - thank you - also! That previous discussion were you interpreted my self report was like a bomb dropped! I think I didn't understand how I was using extraverted thinking. Also I was a bit turned off by the typical type descriptions of INTJ. I ended up watching a lot of Joyce Meng's YouTube channel where she hosts many different types & discussions on perceived differences. Personality Hacker Podcast was another one that contained good information.

I have only been intermittently active on the forum up until recently (it took data from 2025 only), and I deliberately censor myself here quite a bit. I do note that in text, we have the time to craft and form an impression - which can be false. But this also reveals unknowns the prompt may pick up.

I had fun!

[Joshe]

@Natasha Tori Maru Yeah, there are problems with using ChatGPT for typing. 

You’re welcome! Yeah, the way you want to help others by skipping over all the feels and getting right to the solution is Te. Lol. It’s funny. It presents as odd to people. Like, they can sense that I care deeply about their problem but I’m only interested in connecting with them logically, strategically, searching for clarity and a path forward. Classic Ni-Te. That’s just how we  “connect” I guess. I’ve wondered if this is some sort of a defense mechanism to keep emotional flood gates closed. 

Yeah, a lot of the INTJ personality descriptions are like fluff-pieces. When lots of people read it,  they say “I’m that one!” because the type does contain virtues most people would like to embody. Ni provides vision and direction, Te analyzes and calculates how to get there, and Fi stands for something and will stand all by itself if it must. 

I never let the fluff pieces go to my head but there is some truth to it. Doesn’t make us special, just different. And we still have many shortcomings, obviously. I never really felt better or superior to anyone, just very different. Learning about my type explained a lot. 2-3% of the population are INTJ, so realizing that and the difference in cognition has helped me to accept myself and others. Most people IRL see me as ‘weird’, so being and INTJ isn’t anything glamorous, lol, that’s for sure. 

Yeah, Joyce is a good resource, as is personality hacker. And yes, text is different than speaking off the cuff but unless you’re very fragmented, I don’t see how you could hide your type for very long. But maybe some types are just easier to peg than others. I haven’t gone deep on all the types. 

You might also like “Love who” and Frank James. Frank James is mostly funny skits but you can pickup a lot from them.
 

Edited Tuesday at 04:39 AM by Joshe

[Carl-Richard]

I fed the prompt in the original post to the deep research ChatGPT o3 and told it to type @Leo Gura with a ≥1000 post sample size. Here is a summary and a link to the enormous ~7500 words research paper it generated:

Quote

To summarize our findings:

• Leo Gura’s MBTI type is most likely INTP (Introverted, Intuitive, Thinking, Perceiving), with an approximately 35% Bayesian posterior probability in our model.
• A close second is INTJ (about 33% posterior).
• It is highly implausible that he is extroverted (we assign ~5% to ENTP mostly for theoretical completeness), and somewhat unlikely that he is a Feeling type like INFJ (~27% but that number would drop if we weighted the Thinking evidence more).
• In practical terms, we can say he is an INTx – an intellectual introvert with intuitive and thinking preferences. He himself once casually identified as INTP, and our rigorous analysis supports that with slight reservations about J/P.
• Functionally, he shows dominant introverted analysis (Ti or Ni) and auxiliary expansive intuition (Ne or Te, respectively). The balance of evidence (tone, flexibility, test result) leans Ti-Ne (INTP).

https://chatgpt.com/s/t_6863583b5c588191a59bb26dfccec267

I might be misremembering, but this "INTx" answer was something I argued at one point when discussing Leo's MBTI type with @thisintegrated some years ago where he insisted he was INTJ.

I should probably tell it to not weigh any previous test results though, but it didn't seem to weigh it that much according to the report.

Edited Tuesday at 01:46 PM by Carl-Richard

[Natasha Tori Maru]

I don't see Leo as INTJ, but that is from witnessing his reported behaviour & the INTP inferior function - extraverted feeling. I actually see him getting caught off guard by this in his dialogues & interactions on the forum. His pursuit of knowledge for its own sake seems very Ti to me, as opposed to Te. My experience of Te is always looking for how it can be applied. Where will this be useful, how does it fit in with my current understanding etc. Leo is a very very careful thinker, I get the impression he takes some time to thoroughly go through a topic or subject before rapping on it - if it is in his chosen field of attention, that is. Because sometimes he has brain-farts, or doesn't specify his frame appropriately. Additionally he seems to be VERY averse to planning and structure with his life routines. He prefers freedom there, from his own reports. INTJ's typically want structure and order in their outside environment to act as support scaffolding so they can be free to bounce around internally.

He reads to me (with my not expert knowledge of MBTI cognitive functions) as Ti - Ne all over the place. 

Just my perception though.

24 minutes ago, Joshe said:

Yeah, the way you want to help others by skipping over all the feels and getting right to the solution is Te. Lol. It’s funny. It presents as odd to people. Like, they can sense that I care deeply about their problem but I’m only interested in connecting with them logically, strategically, searching for clarity and a path forward. Classic Ni-Te. That’s just how we  “connect” I guess. I’ve wondered if this is some sort of a defense mechanism to keep some sort of flood gates closed. 

It's friggin' too real >.<

I cannot tell you the volumes of consternation I cause/d the women in my family/life through always steamrolling the emotions and spearheading to a solution. I think I felt like always paying attention to emotions and emotional systems (cause/effect etc) made me a feeler. Additionally, I am also very aware I keep people at a distance by this type of emotional analysis and attention. I dislike expressing my own feelings and emotions to others, so it is much easier to focus on theirs and attempt to fix. And, like yourself, I am definitely an oddball. So determined to remain authentic to myself, with zero fucks given to fitting in. 

*spits on Fe*

 

Edited Tuesday at 04:43 AM by Natasha Tori Maru

[Joshe]

3 minutes ago, Natasha Tori Maru said:

*spits on Fe*

😂

[Joshe]

14 minutes ago, Natasha Tori Maru said:

I dislike expressing my own feelings and emotions to others, so it is much easier to focus on theirs and attempt to fix.

Yeah, good point. It’s like there’s some deep, deep well of emotion we have to keep down. I once tripped on psychedelics and that mother fucker came wide open. Lol. It was the best. I literally felt a flood of heavy shit just streaming out of my chest like a beam. Something serious is buried deep. Probably trauma. Lol.

Edited Tuesday at 04:53 AM by Joshe

[Ramanujan]

@Natasha Tori Maruwhat is your MBTI type

[Ramanujan]

@Natasha Tori MaruLEO is not INTJ .he is indeed an INTP . He is goofy and values Fe . INTJ care about Te. Leo is very anti Te. he shut down all conspiracy thoeries discussion which shows he is anti Te

Edited Tuesday at 07:05 AM by Ramanujan

[Natasha Tori Maru]

@Ramanujan INTJ is my current best fit for MBTI.

Re-read my post above. Nowhere do I say Leo is INTJ. INTP would be my closest guess.

[Oppositionless]

Edited Tuesday at 11:04 AM by Oppositionless

[Ramanujan]

@Josheare you INTJ

[Ramanujan]

9 hours ago, Joshe said:

 

Yeah, a lot of the INTJ personality descriptions are like fluff-pieces. When lots of people read it,  they say “I’m that one!” because the type does contain virtues most people would like to embody. Ni provides vision and direction, Te analyzes and calculates how to get there, and Fi stands for something and will stand all by itself if it must. 

I never let the fluff pieces go to my head but there is some truth to it. Doesn’t make us special, just different. And we still have many shortcomings, obviously. I never really felt better or superior to anyone, just very different. Learning about my type explained a lot. 2-3% of the population are INTJ, so realizing that and the difference in cognition has helped me to accept myself and others. Most people IRL see me as ‘weird’, so being and INTJ isn’t anything glamorous, lol, that’s for sure. 

 

the downside of being INTJ is pretty fucked up too

how little energy we have , we live in a room most of the time , our poor social skills , not going out , having no energy to protect ourself from physical agression etc etc