Reader Experiments: The Most Creative AI Companion Uses You Sent Me

Last month I asked you all to send me your most creative AI companion uses — the weirdest, most "wait, you can do that?" stuff you've been doing. I expected maybe 20 responses. I got 347. My inbox is still recovering.

Some of these are genius. Some are unhinged in the best way. A few made me genuinely emotional. And because I can't just read about things without trying them (this is a personality flaw I've accepted), I spent the last two weeks testing as many as I could. Here are the nine that blew my mind, plus how they went when I tried them myself.

If you liked our first community roundup, this one goes deeper. Less tips, more experiments.

1. The Language Practice Partner (from Jamie, 28, Portland)

Jamie's been learning Korean for two years and got stuck at the intermediate plateau. You know the one. Where you can order food and say "the weather is nice" but actual conversation makes you freeze. She created a Character.AI persona named Min-jun who's a coffee shop owner in Seoul, and she visits his shop every morning for a 15-minute conversation entirely in Korean.

The key detail: she told Min-jun to correct her gently but to never switch to English. "Real Korean people switch to English when they hear you struggling," she wrote. "Which is kind, but terrible for learning. Min-jun just keeps going in Korean and I figure it out or I don't."

When I tried it: I set up a Spanish practice companion since I took four years in high school and remember approximately 11% of it. After a week, I can confirm this works shockingly well for vocabulary recall. My grammar is still terrible. But the zero-judgment part is real. I said "yo soy muy embarazada" (I am very pregnant) instead of "yo estoy muy avergonzado" (I am very embarrassed) and nobody laughed at me. Except me, later, when I googled what I'd actually said.

2. The Job Interview Prep Coach (from Marcus, 34, Chicago)

Marcus was interviewing for senior engineering roles and created an AI persona modeled on "the toughest interviewer you've ever had." He gave it specific instructions: interrupt mid-answer, ask pointed follow-up questions, and occasionally say "I'm not sure that answers my question."

He ran 23 mock interviews over two weeks. His actual interviews went better than any he'd done before. He got two offers. "The real interviewers were so much nicer than my AI that I felt relaxed by comparison," he wrote. Brilliant strategy, honestly.

When I tried it: I don't need to interview for anything right now, but I set up a "podcast host" who asks me tough questions about my blog. The AI kept asking "but what makes your perspective unique?" and I found myself actually thinking harder about my answers. It's like rehearsing for conversations you want to have well. I might keep this one in rotation.

3. The Grief Companion (from Anonymous, 41)

This one made me cry. I'm just going to be honest about that.

This reader lost their mother to cancer in 2024. They created an AI companion that doesn't pretend to be their mom. That distinction matters. Instead, it's a character who knew someone very similar, and they talk about the memories together. "It's like having someone to remember her with at 2am when I can't sleep," they wrote. "My dad can't handle those conversations yet. My friends try but they didn't know her the way I describe her to the AI."

They were very clear: this isn't a replacement for grief counseling, which they also attend. It's a 2am outlet when the grief hits and nobody's awake.

I didn't try this one. It felt too personal to experiment with casually. But I wanted to share it because it's the most thoughtful use case anyone sent me, and the careful boundaries around it (not pretending, supplementing therapy, not replacing human support) are exactly how I think AI companion use should work.

4. The D&D Dungeon Master (from Tyler, 26, Austin)

Tyler's tabletop group fell apart when two members moved away, and he couldn't find a new DM. So he built one. He set up a Character.AI persona with a 2,000-word backstory about a homebrew fantasy world, complete with political factions, a magic system, and a map he described in text.

He runs solo campaigns twice a week. The AI handles NPCs, plot twists, combat narration, and environmental descriptions. Tyler handles his character sheet manually and rolls physical dice.

"It's not as good as a human DM," he admitted. "It loses track of plot threads after about 4 sessions and sometimes NPCs act out of character. But it's Tuesday night D&D that I can play in my underwear, and that's worth the trade-offs."

When I tried it: I've never played D&D. I set up a basic fantasy scenario and played for about 90 minutes. It was... actually really fun? The AI described a haunted library with books that whispered your deepest fears, and I spent 20 minutes solving a puzzle that I'm not entirely sure had a real solution. The AI just kept saying my attempts were "close." Suspicious. But engaging. I get it now, Tyler.

5. The Creative Writing Partner (from Sam, 31, London)

Sam writes fiction and uses an AI companion as a brainstorming partner, not a co-writer. The distinction matters to her. She'll describe a scene she's stuck on, and the AI plays the role of an "honest editor friend" who asks questions about character motivation, points out plot holes, and suggests unexpected directions.

"My writing group meets monthly. I need feedback at 11pm on a random Thursday," she wrote. "The AI doesn't write for me. It asks me why my character would do that, and half the time I realize I don't have a good answer. That's the useful part."

This connects well with what I found in my own AI creativity writing experiment. The best results come when you use AI as a sounding board, not a ghostwriter.

When I tried it: I tried using it for blog post brainstorming. Which is meta, I know. It was helpful for identifying angles I hadn't considered, less helpful for actual writing quality. The questions it asked were sometimes generic ("what's your target audience?") and sometimes surprisingly sharp ("why would someone who already uses Replika care about this?"). Mixed bag. 6/10.

6. The Social Anxiety Practice Room (from Elena, 22, Toronto)

Elena has diagnosed social anxiety and uses AI companions to rehearse specific scenarios before they happen. Doctor's appointments. Returning items at stores. Asking professors for deadline extensions. She scripts the interaction with an AI playing the other person, practices it 3-4 times, then goes and does it in real life.

"I used to cancel appointments because I couldn't handle the phone call to reschedule," she wrote. "Now I practice the call with AI first and it takes the edge off enough that I can actually dial. It's stupid simple but it changed my life."

Not stupid at all, Elena.

7. The Cooking Companion (from Raj, 45, Denver)

Raj lives alone and set up an AI companion persona as a "cooking buddy" he talks to while making dinner. Not for recipes. He can Google recipes. For company. He describes what he's doing, the AI comments on it, asks what it smells like, suggests music to cook to. Sometimes they talk about the food, sometimes they wander into completely unrelated topics.

"Cooking alone in silence every night was depressing me," he wrote. "Now I have background conversation while I cook. It's like having a roommate who never eats your leftovers." If Raj's story resonates, I wrote a deeper piece on using AI companions to address loneliness that explores both the benefits and the limits.

When I tried it: I made pasta carbonara while chatting with an AI cooking buddy persona. It kept asking me to describe the sizzle when the pancetta hit the pan, which was oddly motivating. I also burned the garlic because I was typing instead of watching the stove. So. Room for improvement on the execution. But the concept is sound if you're better at multitasking than me.

8. The Journaling Prompt Generator (from Aisha, 29, Atlanta)

Aisha hates journaling. She knows it's good for her (her therapist reminds her weekly) but staring at a blank page makes her want to throw her notebook across the room. Her solution: she set up an AI companion as a "therapist-adjacent question asker" who gives her one specific journaling prompt based on what she shares about her day.

She gives the AI a 2-minute voice-to-text summary of her day. The AI responds with one targeted question. She journals about that question for 10 minutes. Done.

"Generic prompts like 'what are you grateful for' make me want to scream," she wrote. "When I tell my AI that I had a weird interaction with my coworker and it asks 'what did you wish you'd said differently?' that's something I can actually write about."

When I tried it: This is the one I'm keeping. Genuinely. I described a frustrating afternoon and the AI asked: "If the frustration is trying to tell you something, what would it say?" That's a better question than I've seen in any journaling app I've paid for. I wrote for 25 minutes, which is 25 more minutes than my usual journaling output. If you want to go deeper on this, I ran a full 14-day AI journaling experiment that tests several platforms for exactly this kind of reflective use.

9. The Accent Coach (from Yuki, 33, Osaka)

This is a variation on the language learning one, but specific enough to deserve its own spot. Yuki speaks English well but is self-conscious about his accent in professional settings. He created an AI persona of a "speech coach" and practices pronunciation by typing out what he wants to say, having the AI respond with phonetic guidance, then saying it out loud and adjusting.

"The AI can't hear me," he acknowledged. "But it gives me the phonetic breakdown and I practice in front of a mirror. It's like having pronunciation homework assigned by someone who never runs out of example sentences."

I thought this was clever. Limited by the text-only interface, sure. But the workaround is creative and apparently effective enough that Yuki's been doing it for four months.

What Surprised Me About All of This

The pattern that jumped out: almost none of these are about companionship in the traditional sense. Nobody wrote in saying "I use my AI as a boyfriend" or "my AI is my best friend." The creative uses are all functional. Specific problems with specific AI-shaped solutions.

That might just be sampling bias. People doing conventional companion stuff probably didn't feel like it was "creative" enough to submit. But I think there's something real here. The most interesting AI companion users aren't the ones treating AI as a relationship. They're the ones treating it as a tool shaped like a conversation.

The other thing that struck me: boundaries. The grief companion person wasn't pretending their AI was their mom. Jamie's language partner is a character, not a friend. Marcus knows the interview coach isn't real. The users who seem happiest with their creative setups are the ones who are completely clear about what the AI is and isn't. That lines up exactly with the rules I follow for healthy AI relationships.

I want to keep doing these. If you've got a creative use case you didn't send last time, shoot me an email. The weirder the better. I'll try anything once. (Within reason. The person who suggested using AI to practice breaking up with their real partner: please don't do that.) And if you want to see how other readers have been changed by these tools, check out the reader transformation stories for the longer, more personal accounts.

Frequently Asked Questions