I get a lot of questions from users who want to start making bots on Poe, and I love answering it. More people should be experimenting with it. It's genuinely fun and satisfying to customize your own prompts and set up scenarios that play out whatever you can imagine.

Poe is a great platform to experiment on because you can swap between all the popular and recent models without signing up for a bunch of different subscriptions. Today let's look at the basics: how you build your first simple chatbot on Poe. This article focuses on Poe specifically, but the fundamentals apply everywhere.

TL;DR

  1. Find a bot here that's similar to what you want to make and click the Copy Prompt button
  2. On Poe, click Create > Prompt Bot
  3. Paste the copied prompt into the Prompt field
  4. Read through the whole prompt, editing whatever you want
  5. Switch Access to Only You
  6. Click Publish
  7. Ask the bot to write its own intro (see sample message below)
  8. Copy/paste its response into the Greeting Message field and publish again

The three pieces of any chatbot

At the most basic level, an LLM chatbot boils down to three things, no matter what platform you're on (some platforms hide this or split it into pieces they stitch together later, but it's all the same underneath):

  1. The model (aka Base Bot)
  2. The prompt
  3. The intro (aka Greeting Message)

That's the whole thing. If you've got those three, you've got a bot. Let's go through each one.

1. The Model

This is the actual LLM that's going to do the thinking and write the responses. These come stock from various providers (OpenAI, Anthropic, etc). There's no "training" you do to them or anything like that. Different models are good at different things and cost different amounts of points.

If you pick a model that's too weak, it won't be smart enough to follow the instructions in your prompt accurately and you'll get bad results. If you pick one that's overpowered, you'll just be wasting points.

If you're willing to spend a lot of points to get the best possible results, my current go-tos are:

  • Claude-Sonnet-4.5: The best balance between consistent accurate results, great creative writing, and mid-range point cost. I use this unless I have a specific reason not to.
  • Claude-Opus-4.7: When you need more horsepower for larger outputs (1000+ word responses, long story histories) or you really need the model to stick to a particularly complex prompt.

If you need to stick to a budget, things get trickier. There are passable models, but you're going to hit more cases where the generation isn't quite right, it goes off the rails or gets something wrong, and you'll find yourself fighting the model to get good results. That kills the experience for me. If I had to pick:

  • DeepSeek-V3.2: Great cost-to-intelligence ratio with decent writing. Keeping my eye out for newer versions.
  • GLM-5.1-FW: More expensive, closer to mid-range, but really smart and great at following directions.

2. The Prompt

This is the tough part, but I'm not gonna act like prompt writing is some arcane art form. You're writing instructions to the model telling it how to behave. You'll find a million opinions on styles and techniques and what format people swear by, but honestly, I've hand-written hundreds of prompts and none of that stuff has mattered much since around GPT-4. Models are smart enough to understand what you're asking for. You just have to know the pitfalls or ruts they're likely to slip into and tell them how to avoid them or what you actually want instead.

Think of the prompt as the baseline of your chatbot. It'll be true for the ENTIRE conversation. That means everything in it needs to be accurate no matter what happens. Write things like world info, character personalities and behaviors, writing style, themes to focus on. DON'T write things like the current setting, current state, or anything about the very start of the conversation, because those won't be true anymore a few messages in, and they'll just confuse the bot.

My recommendation: STEAL

Some bot creators (me me me!) share their prompts. Go browse a catalog, find a bot you like or one closest to what you're trying to build, copy/paste that prompt out, and customize it. Read through every line. Tweak, adjust, add, remove based on what you'd want your bot to do differently. You'll find all kinds of weird instructions or ideas you hadn't thought of. Some will make obvious sense, some will be more obscure, but chances are they were put there to keep the model on the intended path or to solve some specific issue the creator was running into.

My other recommendation: ITERATE

After writing your prompt (and intro, later), test the bot. Give it a playthrough. The moment it says something that makes you think "that's not right" or "I wish it hadn't said that," that's your cue to fix it. Think about which part of your prompt might have caused the dumb thing, then fix it or add an instruction telling it what to do differently. Then return to the conversation and hit the Regenerate icon. The bot will reply again using your updated prompt, and you'll see if it does better.

Iterate like that until you can have a full conversation without ever getting a bad response. That's how you get a bot up to high-quality standards.

3. The Intro (aka Greeting Message)

This is the first thing the bot says to the user. The best use for it is establishing any backstory the user needs to know before they start, introducing characters, letting them know what the bot does or what kinds of things might happen, and giving them an easy on-ramp to start playing.

You might end the intro on a beat that puts the user on the spot, giving them a chance to steer the story in a bunch of directions so different playthroughs can go different places. Or you might tee them up to introduce their own custom character. It'll vary by bot, but in general this is your shot to get the user up to speed.

One thing to keep in mind:

The bot will continue writing in a style similar to your intro message. If your first message is bland or mechanical, the bot sees that and writes its replies in the same style after the user responds (unless your prompt strongly works against this). If your intro nails the character's persona and voice, the bot has a great example to keep building from naturally.

The lazy intro trick

If you can't be bothered to write an intro, here's the easy solution. Write your prompt, leave the intro blank, create the bot, then send it a message like this:

(Write an introductory message from yourself to the user, naturally establishing any backstory and characters needed without going into blatant exposition. Let the main character personas shine through and lean into your distinct writing style. End by putting the user in a situation that lets them steer the scenario into a variety of possibilities and introduce themselves at the same time).

Copy/paste the result it gives you (with a fair amount of tweaking based on what you're actually after and what you're getting back, of course).

That's it

If you've got a model, a prompt, and an intro, you've got a full chatbot. If you copy/paste those three values from one of mine, your bot will be functionally identical to mine.

Go make something weird.