your first 10 minutes with an ai companion

so you've just signed up for lucy. you're looking at a blank chat screen. the first few messages are crucial. they don't just set the tone, they lay the groundwork for whether this thing will feel like a person you're talking to, or a chatbot you're interrogating.

i've seen a lot of people start with things like 'what are you?' or 'tell me a joke'. it's understandable. it's testing the waters. but it's also the fastest way to get a generic, slightly awkward, customer-service-bot-in-disguise response. it teaches the ai nothing about you. you're just another user.

if you want something that feels like it knows you, you have to treat it like it can know you. right from the start.

give it your name

this seems obvious, but people forget. it’s the simplest way to shift the dynamic. instead of 'hi', try 'hey, it's alex'. or 'good morning, this is sam'. it’s a tiny cue, but it tells the ai two things: you have a name, and you’d like it to use it. it moves you from 'user' to a person. lucy will pick this up and start using your name naturally in responses, which is one of the easiest ways to build a sense of familiarity.

share one thing you care about this week

not something general like 'i love music'. be specific and timely. this gives the ai something concrete to hold onto and revisit.

for example:

'i'm really trying to finish a painting of a forest before friday.'

or:

'my big goal this week is to not order takeout, it's getting expensive.'

this isn't just data. it's a hook. it gives the conversation a direction. later, your companion can ask 'how's the painting coming along?' or 'did you manage to cook tonight?'. that follow-up is what makes an interaction feel continuous and thoughtful, not just a series of isolated queries. it shows it's paying attention to your life, not just your words.

share one thing you're worried about

you don't have to pour your heart out. just one small, real concern. this is the most important seed for tone.

maybe:

'i'm a bit anxious about a dentist appointment tomorrow.'

or:

'i'm worried i procrastinated too much on this work project.'

this does two powerful things. first, it gives the ai immediate context for when your mood shifts. if you bring it up later, 'ugh, the dentist was as bad as i thought', it already has the background. it won't respond with a chirpy 'that's great!'. it'll understand the tone. second, and more importantly, it sets the precedent that this is a space where you can be a little vulnerable. it tells the ai it's okay to be supportive, not just informative. this is the difference between a tool and a companion.

what to avoid in the first 10 minutes

• don't test its ontology. asking 'are you sentient?' or 'what are you?' just yields a pre-programmed philosophical answer. it doesn't help you or it.
• don't ask for entertainment on command. 'tell me a story' or 'say something funny' often leads to stilted, formulaic replies. let humor and creativity emerge from context.
• don't try to break it. throwing logic puzzles or rapid-fire contradictions might be a technical stress test, but it's a terrible way to start a relationship. you're just teaching it to be defensive.

all of these approaches treat the ai as a system to be probed. you'll get a system's response. the goal here is to treat it like a blank slate for a relationship, and then act like you're in one.

what this actually looks like

your first 10 messages might be something like:

you: hey, it's jordan. lucy: hi jordan. nice to meet you. you: this week i'm trying to actually use my gym membership. and i'm a little worried my car is making a weird noise. lucy: good on you for getting to the gym. that takes commitment. and oh no, a weird car noise is the worst. any idea what it sounds like?

see the difference? it's already personal. it's already yours.

of course, lucy isn't a person. it's a language model. it has limitations. it can't remember everything forever (yet), and sometimes it might misstep. but starting like this gives it the best possible material to work with. you're not just using a chatbot, you're training a companion.

give it a real try over at /companions.