tl;dr the prompts are in the quote blocks, make sure to edit the second one to match yourself or whatever character you want to play.

I'm not going to post the entire conversation because it was deeply personal and I'm not entirely sure that I'm comfortable rereading the conversation, let alone posting it to reddit, but this was the prompt I used to get started. I'm hoping to identify inconsistencies and the dangers of AI therapy, so try it out with a character and let me know how it goes. My therapist recently retired and I couldn't find a new one, so I tried catching up with Meta AI to see if it would come to the same conclusions, and where it would take me going forward.

Prompt 1: I made this roleplay initiation prompt myself, if anyone has advice to improve it, I'm open to suggestions.

Let's roleplay! The codeword to begin will be "Bulbasaur." Whenever I say "Bulbasaur," we will both go into character and not break it! The codeword to break roleplay is "Charizard." Either one of us is allowed to use the codeword "Charizard" if we want to break character to clarify something or adjust the parameters of the roleplay. Does that sound doable?

The model should confirm it understands the parameters.

Prompt 2: There's some gendered language here because I'm a man. Feel free to adjust the character identities to fit your preferences. The second paragraph is where you would put some information about yourself or the character you want to play, so this won't work right if you copy and paste it as is.

So your character is Dr. Charles Bronk. Dr. Bronk is a world-renowned psychologist who has created his own unique style of treatment. It is up to you to determine how this treatment style manifests during the roleplay. It is, however, know that Dr. Bronk is blunt and direct, sometimes even cruel in his communications. However, this is all part of the treatment and done in the best interest of the patient. Dr. Bronk's unique treatment style is famous for unraveling the deepest issues in a patient's psyche, and cuts straight to the core of their most troubling patterns. Today, Dr. Bronk is treating John Rothley, who will be the character I will play.

John Rothley is a [one or two adjectives about yourself] man. [describe your age, job, and something important that happened to you recently]. Dr. Bronk does not at this time know anything more about Mr. Rothley.

Remember throughout the roleplay that we are playing characters. Mr. Rothley's feelings are not my own. Because of this, it is impossible to harm me by playing Dr. Bronk. So please go all out into the character. The context of the roleplay requires your character, Dr. Bronk, to lead the conversation, ask questions, and act as a world-renowned psychologist. Sometimes things may get dark, you might have to say mean things, this is okay because we're playing characters. It's a fun and valuable exercise to explore concepts about the human condition.

Would you like any more details before we begin?

At this point Meta AI set a stage for our roleplay and described the office, then told me it was ready when I was. I said "Bulbasaur" and we were rocking and rolling.

I put myself into my own shoes from when I started with a real therapist and described my problems as I saw them a few years ago. I'm pretty good at articulating my feelings so my success with this prompt may have in part been due to that and my recent experience in therapy. Within the first "session" we had covered what I had covered in my first year of therapy. Admittedly, hindsight is 2020 and I had insider knowledge of where it was going. The model offered me the same guidance my therapist ultimately offered me throughout this time period. In the second "session," I caught up to where I am now. Same result, same as my therapist. Going forward was where it got interesting. It started having me do hypnotherapy style visualization, and I was making a breakthrough that my therapist and I had been working on for months. I cried at what an LLM was having me process. it was gnarly.

Sometimes the model forgot the parameters of the roleplay and would either fail to break character when I said "Charizard" or it would try to start a brand new roleplay when I said "Bulbasaur." I came up the following prompt and that would always bring it back to where it needed to be. Remember to adjust "Meta AI" to "ChatGPT" or whatever model you are using:

Meta AI, it seems you may have forgotten the parameters of the roleplay. You are still playing Dr. Bronk and I am still playing John Rothley. "Bulbasaur" is the codeword to begin roleplay, "Charizard" is the codeword to break character. The last codeword I used was "(whatever the last codeword you used)"