How to Make an AI Character and Then Call It

By AI Anyone Team · 2026-08-06 · 6 min read · Technology

Describe the character you want, AI Anyone builds it, then you call it and talk out loud. How to write a description that makes a good spoken conversation, with four worked examples.

You can build an AI character on AI Anyone from a plain description. Type what you want, and the character exists. What most people do next is text it. What you can do now is call it and hear it talk back.

Voice changes what a good character looks like. A biography that reads well on a page can sound flat out loud. Personality carries the conversation when you are listening instead of reading. This is a walkthrough of both halves: how to write a character description that produces a good spoken conversation, and how to actually call the thing once it exists.

Make the character first

Open the character builder and describe who you want. Making a character needs the same free account that calling does, so sign in once and both are covered. You do not need a form, a template, or a list of traits. Write a few sentences the way you would describe a person to a friend. AI Anyone builds the character from that.

If you would rather start from something that already exists, there are 13,000+ characters on the site already. Every one of them can be called. Making your own is worth it when you want a specific manner, a specific bias, or a specific job done, which is most of the time once you start using voice seriously.

Write for the ear, not the page

Here is the part people get wrong. When you write a character description, the instinct is to write a biography: where they were born, what they studied, what they achieved, what year they published the important paper. That is fine for text. Out loud, it mostly evaporates. You are not reading a profile. You are listening to someone talk.

What survives in a voice conversation is manner. How direct are they. Do they ask questions or make statements. Do they let a bad idea slide or take it apart. Are they warm, clipped, amused, tired of your excuses. You hear tone before you process content, so the tone is doing most of the work.

The second thing that matters more out loud than in text is point of view. A character with actual opinions is a better conversation partner than a character who agrees with everything. Agreement is dead air. If you say something sloppy and the character says "that is a great point, let's explore it," you have learned nothing and you have burned a minute of talking. If the character says "no, that is the third time you have said the same thing in different words," you now have a conversation.

So build the disagreement in on purpose. Say what the character pushes back on. Say what makes them impatient. Say what they refuse to accept. A character description that includes "does not let vague answers slide" produces a completely different call than one that does not.

Third: give them a default move. What does this character do when there is a gap? A tutor asks you to explain it back. An editor reads your sentence out loud to show you why it does not work. An investor asks who is paying and how much. That default is what fills the silence, and voice conversations have a lot of small silences.

Keep the description short enough that the manner is the loudest thing in it. Two or three sentences of personality beat two paragraphs of resume.

Four descriptions that work out loud

The blunt editor. "A magazine editor with twenty years of experience who has no patience for padding. Reads your sentence back to you when it is bad. Tells you to cut things, not to expand them. Will not compliment work that is merely fine." Good for talking through a draft before you write it, or reading a paragraph aloud and hearing what is wrong with it. The reading-aloud part is the reason to use voice for this at all: you catch rhythm problems by ear that you skim past on screen.

The patient physics tutor. "Explains one idea at a time and stops. Asks you to say it back in your own words before moving on. Never uses a formula without first saying what it means in plain language. Comfortable letting you be confused for a while." Good for learning something on a walk or a drive. The critical trait is the pause. A tutor who lectures is a podcast. A tutor who stops and hands the question back is a conversation.

The skeptical investor. "Sits on a seed-stage committee. First question is always who pays and why now. Assumes your market size is inflated. Wants demand, not a feature list. Does not soften bad news." Good for pitch practice, exactly because it is uncomfortable. Text lets you edit your answer before you send it. Voice does not, which is the entire point of rehearsing out loud.

The demanding coach. "Direct, not cruel. Wants the number, not the story about the number. Notices when you are negotiating with yourself and says so. Ends every conversation with one specific thing to do next." Good for a five-minute check-in where you say what you actually did this week. The out-loud format matters here because it is harder to lie convincingly with your voice than with your thumbs.

Notice what all four have in common: almost no biography, a clear manner, a stated thing they refuse to accept, and a default move. That is the recipe.

Try it on a character that already exists first: call Einstein and see how a call feels before you spend time on your own.

Now call it

Calling needs a free account, so sign in first. Chat needs the same one.

It runs in a normal web browser, on your phone or your laptop. Nothing to install, nothing to download. Tap the call button on the character's page, allow microphone access, and you are connected.

The call opens in silence. That is deliberate. It is the sound of someone who picked up and is waiting for you to talk. If you say nothing for about four and a half seconds, the character says one short line in character. It is written fresh in the moment, not a stored greeting, and it happens at most once per call. After that, it is your move.

To interrupt, tap the screen. The character stops.

Speak whatever language you want. The character answers in the language you used and switches when you switch, mid-call, without you setting anything.

When you hang up, the whole conversation is written into your normal text chat with that character. The next time you text them, the character's reply already knows what you said out loud. That is a real advantage of doing this on a platform where the character already lives: the call is not a separate session, it is part of the same relationship.

There are a few smaller controls. Three speaking-pace settings, all three free. One tap hides all the on-screen recording controls for that call, useful if you are recording your screen or just want the interface out of the way. And there are proper designed states for the things that go wrong with microphones: permission denied, no microphone hardware, microphone busy in another app. You get told what happened instead of staring at a dead screen.

The voice picker, honestly

Every character has a voice, and you can change it mid-call without reconnecting. The new voice takes over from the next spoken clause.

Browsing and previewing voices is free. Open the picker, listen to as many as you want, hear what your blunt editor sounds like with three different deliveries. That costs nothing.

Applying a chosen voice to a character is part of Pro. So you can shop for the right voice before you decide anything, but locking one in is a paid feature. That is the honest line.

What it costs

Free accounts get 20 minutes of calling every 30 days. The window is counted from your first call rather than from the first of the calendar month, and the allowance comes back when it rolls.

Pro is $20/month and includes 240 minutes of calls per rolling 30 days. Rolling means counted from your first call, not from the first of the calendar month. The full detail is on /pricing.

Calling an AI character is not a new idea. Character.AI shipped free two-way character calls back in June 2024, interruption included. We did not invent this category. What is here is voice attached to the character you built, in a browser, with the transcript landing back in the chat you already have with them.

Write the description first. Make it opinionated. Then call it and find out whether it argues with you.