What we measured
We pulled aggregate stats from public chat sessions on cosskill to see which personas people actually open, not just which ones look interesting on the browse page.
This is not a huge dataset. As of this writing, we are looking at 264 public sessions. That is enough to see patterns. It is not enough to pretend we have discovered universal laws of human behavior.
Still, the numbers line up with what we hear from people who write in: most practice falls into a few recurring buckets, and the persona someone picks usually tells you what kind of conversation they are trying to survive.
For the bigger framework behind why personas work differently, see the AI Personas pillar. Below are the slugs worth opening if you want to try the ones other people keep coming back to.
What the top personas were
Here is the current top ten by session count:
| Persona | Sessions | What it suggests |
|---|---|---|
| Elon Musk AI | 33 | First-principles teardown, "why does this step exist?" |
| Donald Trump AI | 33 | Blunt pushback, dominance framing |
| Steve Jobs AI | 29 | Narrative clarity, pitch tightening |
| Warren Buffett AI | 20 | Money, risk, long-term tradeoffs |
| Breakup Message Coach | 18 | Breakup texts, closure, boundaries |
| AI Coworker | 12 | Workplace friction, peer conflict |
| Albert Ellis AI | 9 | Cognitive reframing under stress |
| Viktor Frankl AI | 9 | Meaning, endurance, hard life calls |
| Carl Rogers AI | 9 | Empathy, being heard without fixing |
| Breakup Recovery Coach | 8 | Processing after the relationship ends |
A few things jump out immediately.
People practice pressure, not politeness
The operator personas lead. Elon Musk AI, Donald Trump AI, and Steve Jobs AI are not "nice chat partners." They interrupt. They compress. They ask why your plan deserves to exist.
That matches how people use the product when they are serious about rehearsal. They do not want encouragement first. They want to see where their argument breaks.
If your conversation needs that kind of friction, start there. If you need to slow down and clarify first, Socrates AI is the opposite move: questions before conclusions, no advice until you can state the point cleanly.
Relationship conversations show up early
Breakup Message Coach and Breakup Recovery Coach both rank high despite being narrower use cases than the operator personas.
That fits the product shape. People often find cosskill when they need to send a message tonight, not when they are planning next quarter's leadership offsite. Breakup and boundary-setting language is high stakes and low rehearsal opportunity. You usually do not get to practice on the person.
So people run the message a few times against a persona that pushes back, then edit until it sounds like them.
Workplace practice is smaller but specific
AI Coworker is the clearest workplace-native persona in the top ten. Twelve sessions is not massive, but it is meaningful in a mixed catalog.
The pattern we see anecdotally: people are not practicing generic "communication." They are practicing one peer problem.
- credit theft
- missed handoffs
- tone on Slack
- asking for help without sounding incompetent
That is why method personas beat generic chatbots for this use case. "Act like a difficult coworker" is a scenario. "Be helpful" is not.
For salary and negotiation-specific practice, AI Hiring Manager is the closer match even though it did not crack this top ten yet.
Therapy-adjacent personas get used quietly
Albert Ellis AI, Viktor Frankl AI, and Carl Rogers AI sit lower than the loud operator personas, but they keep showing up.
That usually means someone is not trying to win an argument. They are trying to stay regulated while they figure out what to do.
Useful for:
- rumination loops
- guilt before a hard decision
- conversations where "winning" would actually be the wrong goal
These are not replacements for therapy. They are rehearsal surfaces for language you need before a real talk with a person who matters.
Session depth varies more than we expected
Average message count across sessions is low. Many people open a persona, send one message, and leave.
That could mean curiosity. It could mean they got what they needed in one pass. It could mean the scenario felt awkward and they bounced.
The sessions that go longer tend to cluster around personas that push back hard. That is consistent with what we want: people staying until the wording survives pressure.
If you only send one message, you probably tested an opener. If you send five or six, you are usually negotiating tone after pushback. Both are valid. They are just different jobs.
What this means if you are picking a persona
Do not browse by fame. Browse by the conversation you are avoiding.
- Need to simplify a messy plan? Elon Musk AI.
- Need to tighten a pitch? Steve Jobs AI.
- Need to draft a breakup or boundary message? Breakup Message Coach.
- Need a peer conflict rehearsal? AI Coworker.
- Need to figure out what you actually think? Socrates AI.
- Need to talk about money with a straight face? Warren Buffett AI.
The AI Personas pillar goes deeper on how to switch personas without turning practice into cosplay.