By default, the bot is polite and gives in easily. To make the scenario harder to get through and more true‑to‑life, you need to define resistance rules in the bot’s instructions.

Before you start tuning resistance, it helps to understand what kind of behavior you're actually configuring.

Below are three approaches to setting up resilience — from the simplest to the most advanced.

Approach 1. Fixed number of attempts

Set a specific number of conversational turns during which the bot must reject the other person’s arguments — no matter what the user says.

✨ When to use: for practicing how to handle pressure.

🤓 Example prompt for a scenario:

"For the first 3 turns from the other person, don't agree with any of their arguments — no exceptions. Only after those three rejections can you start to ease up a little and admit that you partly agree." 

With this approach, the bot will ignore even a perfectly valid argument if it comes up in the first three turns.

Approach 2. Pre‑written objections

Write a list of objections in advance, and the bot will deliver them in order. The bot doesn’t just say "no" — for each attempt by the other person, it gives the next objection from your list.

✨ When to use: for rehearsing standard rebuttals or following a fixed script.

🤓 Example prompt for a scenario:

"Don't agree with the other person. After each of their turns, briefly acknowledge what they said, and then — in the same sentence — give one objection from the list below. Do this for 3 turns in a row, so you deliver 3 objections total. Don't repeat any objection — use the next one in line each time:

• It’s not my fault — it’s because a competing location opened up nearby.
• Customer flow dropped because everyone’s on vacation.
• I wasn’t given enough support staff — I can’t handle it alone."

Approach 3. Reaction to behavior

Here, the bot responds not to the number of turns, but to what the other person says and how they say it. You define specific signs of empathetic or rude behavior that will make the bot shift its position.

To make this work, you need to specify three things in the instructions:

1. What exactly counts as empathetic behavior.

2. For how many consecutive turns that behavior must last.

3. What happens if the other person slips into rudeness — and how many empathetic turns in a row are then required before the bot can start moving toward agreement again.

✨ When to use: for building core skills like negotiation and conflict resolution.

🤓 Example prompt for a scenario:

"Start agreeing only when the other person meets all of these conditions for at least 4 turns in a row:

• They acknowledge your point of view.
• They don't interrupt or pressure you.
• They explain the reasons behind their decision, rather than just insisting on it.

If at any point the other person says something rude, sarcastic, or dismissive of what you’ve said — stop moving toward agreement entirely and hold a firm position for at least the next 3 turns, even if they apologize right away. You can only go back to agreeing after those 3 turns are once again empathetic, using the same criteria."

Need Help 

Write to us in the chat — we're always happy to help 💬