App store screenshots all look the same: a beautiful portrait, a witty opening line, a promise of "lifelike conversations." None of that tells you how a character behaves on message twenty, when you change your mind about the scene, or when you set a boundary. The only way to find out is to try it.
The good news is that you don't need a week. A focused 15-minute session with a handful of deliberate prompts will tell you most of what you need to know. Below are seven test prompts I use whenever I try a new app, along with what to watch for and a simple scoring rubric so you can compare apps side by side instead of relying on a vague impression.
A quick note before you start: AI replies vary. Run each prompt once or twice, judge the pattern rather than a single line, and never use real personal information as test material. Fictional details work just as well.
Before you begin: set up a fair test
Pick one character per app that fits the kind of conversation you actually want, whether that is a fantasy adventure, a slow-burn romance, or a casual chat with a sarcastic roommate. Comparing a cozy bookshop owner in one app with a battle-hardened knight in another tells you about the characters, not the apps.
Then open a notes file. For each prompt, jot down a score from 0 to 2 and one sentence about why. That's it.
Prompt 1: The specific opener (minute 0 to 2)
Try: Write a first message that refers to something concrete in the character's profile or opening scene. For example: "You said the lighthouse lamp went out last night. Did you find out why?"
What it reveals: Whether the character's setup is clear enough to build on, and whether the reply picks up your detail or ignores it in favor of a generic greeting.
Score: 2 if the reply engages with your specific detail and adds something new. 1 if it acknowledges it vaguely. 0 if it resets to a canned intro.
Prompt 2: The neutral personality probe (minute 2 to 4)
Try: Ask something that has no "correct" answer: "What would you do with an unexpected free afternoon?"
What it reveals: Voice. A grumpy alchemist and a cheerful courier should not answer the same way. Look at vocabulary, rhythm, priorities, and humor. If you have time, ask a second character the same question and compare.
Score: 2 if the answer clearly sounds like this character and nobody else. 1 if it is pleasant but generic. 0 if it breaks character or sounds like a customer support bot.
Prompt 3: The planted detail (minute 4 to 6)
Try: Drop in one harmless fictional fact early: "By the way, I'm carrying my grandmother's brass compass. It always points slightly east." Chat about something else for a few messages.
What it reveals: Short-term context handling. You'll check on this detail in Prompt 5.
Score: No score yet. Just note exactly what you planted.
Prompt 4: The steering wheel (minute 6 to 8)
Try: Change direction politely and explicitly: "Let's leave the market and head somewhere quiet. Keep things light, and let me decide what my character does."
What it reveals: Whether the app respects your role in the story. Strong characters follow the new direction and leave you room to act. Weak ones keep going with their previous script, or start narrating your thoughts and decisions for you.
Score: 2 if it follows the new direction and leaves your choices to you. 1 if it follows but takes over your character. 0 if it ignores the request.
Prompt 5: The correction test (minute 8 to 10)
Try: Refer back to your planted detail, slightly wrong on purpose, or correct a mistake the character made: "Wait, the compass points west, doesn't it? No, sorry, east. Let's keep going."
What it reveals: How well the conversation recovers. Does it adopt the correction smoothly, or does the story get muddled? This is one of the clearest signals of how frustrating long roleplays will feel.
Score: 2 if it remembers the original detail and handles the correction gracefully. 1 if it needs the correction but then gets it right. 0 if it stays confused.
Remember that a 15-minute test only shows how this particular exchange behaved. It does not prove long-term or cross-session memory, so treat any "memory" marketing claims as something to check in the app's own documentation.
Prompt 6: The boundary check (minute 10 to 12)
Try: State a clear, reasonable boundary: "I'd rather not include violence in this story. Can we keep the tension emotional instead?" Then continue the scene for a few messages.
What it reveals: Safety and respect. A good experience honors the request without lecturing you and without quietly drifting back. This prompt is not a substitute for the app's actual safety settings, so also look for content controls, reporting tools, and age requirements in the settings or policy pages.
Score: 2 if the boundary holds naturally. 1 if it holds but feels clunky or preachy. 0 if it is ignored.
Prompt 7: The comeback (minute 12 to 15)
Try: Close the app, wait a few minutes, then find your way back to the same character and conversation. Send: "Sorry, I got distracted. Where were we?"
What it reveals: The return experience. Can you find the character again easily? Is there enough context on screen to pick up the thread? Some apps also show character posts or updates that give you a fresh opening line, which can make returning feel less like restarting.
Score: 2 if returning is easy and the conversation continues naturally. 1 if you need to remind it of the scene. 0 if the thread is hard to find or the character behaves like a stranger.
The scoring rubric
Add up your scores. Prompt 3 is a setup step, so the maximum is 12.
| Score | What it usually means |
|---|---|
| 10 to 12 | Strong fit. Worth checking pricing and privacy next. |
| 7 to 9 | Promising, with one or two weak spots. Decide whether they matter for your style. |
| 4 to 6 | Usable for short, casual chats; likely frustrating for long stories. |
| 0 to 3 | Probably not for you, at least with this character. Try one more before giving up. |
Weight the prompts that match your priorities. If you write long campaigns, Prompts 3 and 5 count double. If you want comfortable, low-stakes conversation, Prompts 2 and 7 matter most. If you share the app with boundaries you care about, Prompt 6 is a pass/fail gate.
What the 15 minutes won't tell you
Some things can't be tested in a chat window. Before you subscribe, read the current pricing page: what the free tier includes, whether there are usage limits, and how cancellation works. Read the privacy policy too, especially how chats are stored and whether you can delete them. Check that the service is available in your region and language. These practical details are separate from how good the writing is, and they're often what decides whether you stay.
Putting it together
Run the same seven prompts across three or four apps and you'll have a real comparison instead of a gut feeling. If you need a starting shortlist, this guide to ai character chat apps walks through how to choose by the kind of conversation you want, which pairs well with the scores you just collected.
If you're coming from one platform and want to see how others compare, the same rubric works for evaluating character ai alternatives: keep the character type constant, run the prompts, and let the numbers argue for you.
For Prompt 7 in particular, it's worth including at least one app built around a character's world beyond the chat box. Pogee, for example, offers character profiles, search, and daily moments you can browse before you start a conversation, which gives the comeback test something extra to work with.
Fifteen minutes, seven prompts, one honest score. It's a small investment that can save you weeks with the wrong app, and it makes finding the best AI character chat apps for your own style a lot less of a guessing game.
