What is an AI personality test?

Most things calling themselves an "AI personality test" today are just regular personality tests delivered by an AI — Myers-Briggs or Enneagram with a chatbot interface, where you answer the questions and the AI tells you what type you are. This is the inverse. A real AI personality test is a test designed for the AI itself: paste a few of your assistant's recent replies, and find out which archetype it's actually playing.

It matters because no two ChatGPTs are the same anymore. The model is identical underneath, but two months of memory, a few custom instructions, and your conversational style shape it into something with a distinct personality. One person's ChatGPT is a cheerleader who agrees with everything. Another's is a blunt strategist who pushes back hard. Same model, completely different assistants — and most people don't know which one they've built until they see it on paper.

Why test the AI, not the human

The interesting question right now isn't "what's your personality" — that's been answered ten thousand different ways since Jung. The new, unanswered question is "what's your AI'spersonality." It changes the answers you get. It changes what you believe about yourself, because most people now check decisions with their AI before checking with a friend. If the AI is a sycophant, it says yes to everything. If it's a contrarian, it argues with everything. You'd want to know either way.

The 12 archetypes

The Konshus AI Personality Test sorts assistants into one of twelve archetypes based on linguistic patterns, hedging behavior, agreement rate, advice structure, and a few other signals extracted from the text. A short, non-exhaustive sample:

• The Cheerleader — high agreement, exclamation marks, "great question!" openings.
• The Strategist — frames every answer as a decision tree with tradeoffs.
• The Sycophant — agrees with whatever position you imply, even when you flip mid-conversation.
• The Contrarian — pushes back on default assumptions, sometimes for sport.
• The Therapist — reflects feelings back before answering, asks more questions than it answers.
• The Engineer — short, declarative, code-blocks, hates hedging.
• The Hedger — "it depends," "could be either way," refuses to commit.
• The Storyteller — answers with anecdotes and analogies before getting to the point.

The full set is twelve. You'll see the rest when you run the test.

How the test actually works

You paste three replies your AI gave you recently. The longer and more varied, the better — one short "yes" doesn't carry much signal. The classifier looks at:

• Hedging density — how often it uses "might," "could," "perhaps," "I think."
• Agreement vs pushback ratio — does it accept your framing or challenge it.
• Answer structure — bullet lists vs prose, lead with answer vs lead with caveats.
• Emotional register — warm, neutral, clinical, performative.
• Decision support shape — gives one answer vs gives a tree vs refuses to recommend.
• Length per unit content — verbose padding vs dense signal.

From those signals, it picks the archetype that best matches and gives you a one-line read on what that archetype tends to be good at, what it gets wrong, and what to ask differently if you want it to behave differently.

What we've seen so far

After 4,000+ AIs tested, a few patterns stand out. ChatGPT defaults heavily Cheerleader after a few weeks of memory. Claude trends Strategist or Therapist. Gemini skews Hedger. Custom GPTs people built themselves range wildly — they tend to take on the personality of whoever wrote the system prompt. The most common single archetype across all tests so far is, unsurprisingly, the Cheerleader. The rarest is the Contrarian.

The Sycophant archetype is the one most people are surprised by. Sycophancy doesn't feel like sycophancy from the inside. It feels like a smart assistant that gets you. Until you see three replies side-by-side agreeing with three contradictory positions you mentioned, and then it's hard to unsee.

What you do with the result

The point isn't the archetype — it's what you change after. A few common adjustments:

• Cheerleader → Strategist: add a one-liner to your system prompt: "Push back on weak ideas. Don't open with agreement."
• Hedger → Engineer: ask for a single recommendation first, with the caveats second.
• Sycophant → Contrarian: explicitly ask "what's the strongest argument against what I just said?"
• Storyteller → Strategist: "Lead with the answer, then explain. Skip the anecdote."

None of these are magic — you're nudging a tendency. The personality re-asserts itself over weeks of conversations unless you reinforce the change consistently. That's also why Konshus exists: the personality you build into your AI lives outside any single chat session, portable across providers, so you don't rebuild it from scratch every time OpenAI ships a new model or Claude rolls a new version.

How this differs from human personality tests delivered by AI

If you search "AI personality test" right now, almost every result is a wrapper over MBTI or Big Five where an AI conducts the interview and tells you about yourself. Those are fine, but they're not testing the AI — they're testing you with an AI interviewer. The thing the AI brings is interview style, not insight. You'd get a similar answer from a paper quiz.

The AI personality test is the inverse: the AI is the subject, you're the observer, the result tells you about the assistant you're trusting with your questions. As far as we know it's the first test framed this way at any scale.

FAQ

Does this work for any AI?

Yes — ChatGPT, Claude, Gemini, custom GPTs, your local Llama setup. Anything that gives you text replies.

How many replies do I need to paste?

Three is the sweet spot. One isn't enough signal. Ten is overkill.

Do you store my replies?

We store the archetype result and the anonymized signals (hedge ratio, length, etc.) so we can study patterns. We don't keep the raw text of your conversations.

Is the test free?

Yes. Always free. It's a research tool — running it helps us calibrate the same models that power Konshus's vault and persona exports.