Introducing an AI Chatter: A Checklist for Agencies

Updated November 2, 2026 · 4 min read

The best way to introduce an AI chatter is step by step with a single model. Prepare persona and tone, sales scripts, a cleanly labelled content vault and clear pricing rules first, then start a test phase and measure a few simple metrics. Only when that runs smoothly do further models follow.

Step 1: Define persona and tone

The AI is only as convincing as the template it gets. For each model, write down her character (sweet, cheeky, dominant, relaxed; three to five keywords are enough), her language style, her background as far as she wants to share it in the chat, and the taboos. What should never come up? What does the model not answer?

A short persona card might look like this (an invented example): “Cheeky but warm. Short sentences, a 😏 now and then, never says ‘dear’. Works in a bookshop, loves cooking shows. Does not talk about her real name or home town.”

It helps most to provide real example messages. A handful of chats that capture the tone well say more than long descriptions.

Step 2: Prepare sales scripts

An AI chatter guides fans through your scripts, so collect the flows that already work for you: from getting to know each other through the free preview to the offer and the thank you after the purchase. How such scripts are built is covered in Sales scripts in the fan chat.

Make sure each script answers three questions. Which phases exist, and when does it move on to the next one? What are the typical objections (“too expensive”, “later”) and how does the model respond? And what does aftercare look like once a purchase is made?

Step 3: Label your content vault

The AI can only offer what it knows. A properly labelled vault is the basis for fitting offers. Each item needs a short description of what can be seen or heard, in one sentence. Add the theme and mood, so the right piece fits the right fan. Note the type and scope (photo, video or set, length and number) and a price tier: entry, standard or premium. Mark what has already been used, so the same piece isn’t sold twice.

Step 4: Define pricing rules

Decide before you start what the AI may offer and what not. Set price tiers for entry, standard and premium content. An example, with invented numbers: entry 5 €, standard 15 €, premium 40 €. Decide whether and when there are discounts, and how many. Decide how many offers per conversation and per day make sense. And agree on special rules for whales, see Recognising and keeping whales.

Step 5: Test with one model

Don’t start with the whole roster. Pick one model whose tone and content you know well, and run the test phase there:

1. In the first days, read along in many chats to check tone and sales pace.
2. Write down concrete observations, such as “offer came too fast” or “word X doesn’t fit”.
3. Adjust persona, scripts and pricing rules together.
4. Only expand to further models once this one runs stably.

Step 6: Measure a few metrics

Compare the same figures before and after the start, ideally per model. Response time is the first one: how fast does a fan get an answer, including at night? Then look at revenue per fan and per chat, the share of conversations that end in a purchase, and whether regulars get in touch more often. Finally, spot-check some chats and rate them for tone and naturalness.

Measure what matters to you, not everything at once. Compare like with like: same period, same model. For example, compare the four weeks before the start with the four weeks after it.

How onboarding works at WhaleChat

1. You apply for the private beta.
2. We get in touch personally, usually within 48 hours, and show you WhaleChat live in the intro call.
3. Together we set up one model: tone, scripts, content and pricing rules.
4. A test phase with that one model follows, in which we fine-tune.
5. If the test phase goes well, you can add further models.

WhaleChat currently supports 4based and Maloum, and real fan chats run in German today. More background under What is an AI chatter?.

FAQ