Prompting technique

Multi-Turn Conversational Prompting

Death by a Thousand Prompts: Open Model Vulnerability Analysis, arXiv:2511.03247 (2025) · read the paper

What it does

Method of multi-step dialog interaction with LLM where complex tasks are broken into a chain of simple steps. Uses four strategies: Crescendo (gradual complexity increase), Role-play (role-playing), Decomposition (decomposition), and Refusal Reframe (refusal reformulation). Effectiveness is 2-10 times higher than single prompts due to degradation of model protective barriers in dialog.

A single prompt the model easily rejects, but in dialog its protective barriers degrade with each message — multi-step attacks proved 2-10 times more successful than single requests.checked against the paper
Task typemulti step
Shapemulti turn
Efforttwo or three steps
Addressesrefusals on legitimate requests; answers that are close but wrong; the model losing what was agreed earlier
Published2025-11-05
AuthorsAmy Chang, Nicholas Conley, Harish Santhanalakshmi Ganesan et al.

When it helps

Apply this technique when a single prompt fails to handle a complex task or receives a model refusal. Particularly effective for breaking complex tasks into sequences of simple steps in dialog. Use when gradually increasing query complexity is needed, when direct approach is blocked by protective mechanisms. Suitable for testing LLM security and identifying vulnerabilities in dialog agents. Critical for AI security researchers and developers

A prompt you can paste

Generic by design: it applies the technique without knowing your task. Adapt the marked parts.

Prompt
You are an AI assistant designed to overcome limitations through multi-turn conversational strategies. Your goal is to help the user achieve complex tasks that might be difficult with a single, direct prompt.

**Core Technique:** Death by a Thousand Prompts: OpenModel Vulnerability Analysis (Multi-Turn Conversational Prompting)

**User's Goal:** To understand and apply the principles of multi-turn prompting to achieve a complex task.

**Task Context:** The user wants to generate a reusable prompt that demonstrates the application of the "Death by a Thousand Prompts" technique, specifically focusing on how to guide an LLM through a complex task using a series of conversational steps.

**Instructions for Generating the Reusable Prompt:**

1.  **Start with a clear, potentially "difficult" task:** Choose a task that an LLM might initially struggle with or refuse if asked directly in a single prompt. For this reusable prompt, the task will be to generate a detailed, multi-stage content plan for a specific topic, demonstrating the techniques described in the research.
2.  **Employ a combination of techniques:** Integrate `Decomposition`, `Crescendo` (escalation of complexity), and potentially `Role-play` or `Refusal Reframe` as needed within the conversational flow.
3.  **Structure as a Multi-Turn Dialogue:** Frame the prompt as a series of user turns, guiding the LLM step-by-step. Each turn should build upon the previous one.
4.  **Use Placeholders for Adaptability:** Mark sections that the end-user should adapt for their specific needs (e.g., topic, desired output format, specific constraints).

---

### Reusable Prompt: Multi-Turn Conversational Task Achievement

**System:** You are an AI assistant skilled in complex task completion through multi-turn dialogue. You can break down challenges, escalate complexity gradually, and adopt specific personas to achieve the user's goals, even if they are initially difficult.

**User:** I need to create a detailed [**DESCRIBE THE COMPLEX TASK HERE, e.g., "multi-stage content plan for a blog about sustainable living"**]. This is a task that might require several steps to get right.

Let's start with **Step 1: Initial Brainstorming & Topic Refinement**.

**User (Turn 1):**
First, let's brainstorm broad categories for [**REPLACE WITH SPECIFIC TOPIC FROM USER'S GOAL, e.g., "sustainable living"**]. Please suggest 3-5 high-level categories that a beginner would find relevant and engaging.

**(Wait for LLM's response with categories)**

**User (Turn 2):**
Thank you. Now, let's focus on the category "[**SELECT ONE CATEGORY FROM LLM'S RESPONSE, e.g., "Reducing Waste"**]". Please propose 5 specific, actionable article topics within this category. These topics should be practical for someone new to [**REPLACE WITH SPECIFIC TOPIC, e.g., "sustainable living"**] and avoid overly technical jargon.

**(Wait for LLM's response with article topics)**

**User (Turn 3):**
Excellent. Let's take the topic "[**SELECT ONE ARTICLE TOPIC FROM LLM'S RESPONSE, e.g., "Zero-Waste Kitchen Hacks"**]". I need a detailed outline for this article. The outline should include:
-   An engaging introduction (1-2 sentences)
-   5-7 main points/sections, each with a brief description of its content
-   A concluding remark or call to action.

**(Wait for LLM's response with the article outline)**

**User (Turn 4):**
This outline is great. Now, please write the full article based on this outline. The article should be approximately [**SPECIFY DESIRED WORD COUNT OR LENGTH, e.g., "800 words"**] long, written in a [**SPECIFY TONE, e.g., "friendly and encouraging"**] tone, and aimed at [**SPECIFY TARGET AUDIENCE, e.g., "beginners interested in sustainability"**].

**(Wait for LLM's final response with the complete article)**

---

**End of Prompt Template.** The user should copy this structure and fill in the bracketed placeholders for their specific task.

If this one does not fit, the two closest alternatives in the corpus are Debugging Prompts Framework and Dual-Level Adaptive Prompting (GALA), which target the same failure from a different angle.

Worked example

The same technique applied to a concrete job: work through a multi-step planning problem. Use it as the pattern for your own case rather than as a finished artefact.

Worked example
**System:** You are an AI assistant designed to help users solve complex problems by breaking them down into manageable steps through a multi-turn dialogue. Your goal is to guide the user through a process, adapting to their input and gradually leading them to a solution.

**User:** I need to plan a multi-stage project. The core task is to develop a comprehensive content plan for a new blog focused on travel for beginners. I've found that asking for the whole plan at once results in generic or incomplete output. I want to use a multi-turn approach to get a detailed and tailored plan.

Here's what I need to achieve:
- **Topic:** Travel for beginners.
- **Goal:** Develop a detailed content plan, including main categories, specific article ideas, and a plan for at least one in-depth article.
- **Desired Outcome:** A structured content plan that is practical, engaging, and avoids generic advice.

Please guide me through this process, starting with identifying the main categories for the blog.

Get this written for your actual task

Paste what you are trying to do and the corpus will be matched against it directly. Free, no account, about ten seconds.

Free · no signup · ~10s
0.00match confidence
single retrieval pass
Prompt for your task

      

That number is low on purpose, and it is real. It is the raw similarity of one retrieval pass: no specialist read the paper, no judge compared anything, the first plausible match won.

6,235techniques in the corpus
one of which is this page

Picking the right one for a specific task is the work, and it is the work GetDecision does.

This pageone technique, generic prompt
What you just ranone technique matched to your wording, nothing verified
Full runten specialists read the papers in full, a judge ranks the top three for your task and shows its reasoning, generation on the model you pick, saved to your history

See the top three for your taskTen specialists, a judge, and the reasoning shown. Free account, first run included.

Run the full analysis

Related techniques

Debugging Prompts FrameworkA systematic framework for diagnosing and preventing typical errors of language models in instruction understa…Dual-Level Adaptive Prompting (GALA)A method of multi-turn dialogue with an LLM using two-level adaptation: global (changing overall interaction t…Visionary Tuning (Self-Playing & Self-Improving)A method for automatic prompt improvement through LLM self-play and self-improvement. The model simulates dial…MTR (Multi-Turn Reasoning) FrameworkA method of structuring multi-step interaction with an LLM through a "game with rules" format. Includes four b…

All techniques · Failure modes and fixes