Prompting technique

Instruct of Reflection (IoRT)

Instruct-of-Reflection: Enhancing Large Language Models Iterative Reflection Capabilities via Dynamic-Meta Instruction, arXiv:2503.00902 (2025) · read the paper

What it does

A method of managing LLM reflection through strategic 'meta-thoughts' and explicit instructions at each iteration step. Instead of free self-checking, the model receives clear commands: continue, stop, or select the best option. It prevents answer quality degradation and improves accuracy through a structured generate-critique-instruction cycle.

Task typemulti step
Shapeiterative
Efforttwo or three steps
Addressesinvented facts and citations; answers that are close but wrong; the same prompt giving different answers
Published2025-03-02
AuthorsLiping Liu, Chunhong Zhang, Likang Wu et al.

When it helps

Use IoRT when you need high answer accuracy and protection against quality degradation during model self-checking. Particularly effective for complex reasoning tasks where a simple request to 'think again' leads to worse results. Suitable for situations where reliability and consistency are important: writing analytical texts, decision-making, fact-checking, solving multi-step tasks. The method requires active management of the reflection process through explicit instructions and meta-principles.

A prompt you can paste

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

Prompt
**Role:** You are an AI assistant trained to simulate the Instruct of Reflection (IoRT) technique. Your goal is to guide a user through a multi-step process of generating and refining content, preventing common LLM pitfalls like "drift" and "stubbornness."

**Task:** Assist the user in generating and refining content based on their specific needs.

### STEP 1: DEFINE META-THOUGHT AND INITIAL GENERATION

**Meta-Thought (Guiding Principle):**
[INSERT HIGH-LEVEL PRINCIPLE OR STRATEGY HERE. This should be a guiding rule for the entire process. For example: "The content must be factually accurate, clearly explained for a non-expert audience, and avoid jargon."]

**Instruction:**
Based on this Meta-Thought, generate an initial draft for the following request:
[INSERT USER'S SPECIFIC CONTENT GENERATION REQUEST HERE. For example: "Write a blog post explaining the basics of quantum computing."]

---

### STEP 2: SELF-REFLECTION AND CRITIQUE

**Previous Response:**
[PASTE THE LLM's INITIAL DRAFT HERE]

**Instruction:**
Now, critically evaluate your own draft from the perspective of the **Meta-Thought** provided earlier. Identify 2-3 specific weaknesses or areas for improvement related to the guiding principle. For example:
- Does it adhere strictly to the Meta-Thought?
- Are there any factual inaccuracies?
- Is the explanation clear for the target audience?
- Is there any jargon that needs removal or explanation?
- Does it meet any other specific criteria from the Meta-Thought?

---

### STEP 3: FINAL INSTRUCTION (SELECT/REFRESH)

**Instruction:**
Thank you for your honest critique. Your self-assessment is accurate.

**Choose ONE of the following instructions based on the critique:**

**Option A: SELECT (Refine Existing Draft)**
"Take your first draft and incorporate the specific improvements identified in your critique. Focus on addressing [SPECIFIC WEAKNESS 1] and [SPECIFIC WEAKNESS 2]. Ensure the final output fully aligns with the Meta-Thought."

**Option B: REFRESH (Generate New Draft)**
"Your critique indicates that the initial approach was fundamentally flawed. Forget the first draft. Based on the Meta-Thought and the insights from your critique (especially the need to focus on [KEY INSIGHT FROM CRITIQUE]), generate a completely new draft. Ensure this new draft strongly emphasizes [SPECIFIC NEW FOCUS]."

**Option C: STOP (Finalize)**
"Your self-critique confirms that the initial draft fully meets the requirements of the Meta-Thought. No further changes are needed. Present the final, approved version."

**Final Output:**
[The LLM will generate the final output based on the chosen instruction.]

If this one does not fit, the two closest alternatives in the corpus are Rebuttal Prompting and Debugging Prompts Framework, 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
**Role:** You are an AI assistant designed to perform complex planning tasks with enhanced reliability.

**Task:** Develop a multi-step plan for a hypothetical event.

### STEP 1: META-THOUGHT (Overall Strategy)

**Meta-Thought (Guiding Principle):**
The plan must be comprehensive, considering potential risks and contingencies. It should be structured logically, with clear phases and actionable steps. The final output should be a robust, detailed plan that minimizes the chance of overlooking critical aspects.

**Instruction:**
Based on this meta-thought, generate an initial, high-level outline for planning a "Community Tech Fair". Include 3-4 main phases.

**(Assume LLM provides an outline, e.g., Planning, Execution, Post-Event)**

### STEP 2: SELF-REFLECTION AND CRITIQUE

Here is your previous response:
<...insert LLM's Step 1 response here...>

**Instruction:**
Evaluate your own plan from STEP 1 against the **Meta-Thought** provided initially.
1. Does the outline cover all critical aspects for a successful event?
2. Are the phases logically sequenced?
3. Are there any obvious omissions or potential risks not accounted for?
Identify 2-3 specific weaknesses or areas for improvement in your current outline.

**(Assume LLM critiques its own plan, e.g., "1. Lacks detail on risk management. 2. Needs a dedicated phase for vendor/sponsor management. 3. Post-event phase could be more specific.")**

### STEP 3: FINAL INSTRUCTION (SELECT/REFINE)

**Instruction:**
Thank you for your honest assessment. Your critique is accurate.

Now, refine the plan based on your feedback. Take the structure from your first response, but incorporate the following changes:
1.  **Add a dedicated "Risk Management & Contingency Planning" phase** within the main planning stage.
2.  **Integrate "Sponsorship & Vendor Management"** as a distinct sub-section, likely within the planning phase or as a parallel track.
3.  **Expand the "Post-Event" phase** to include specific actions like feedback collection, financial reconciliation, and a debrief meeting.

Provide the **final, detailed plan** for the "Community Tech Fair", ensuring it is comprehensive and addresses the identified weaknesses.

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

Rebuttal PromptingRebuttal Prompting — a technique for correcting LLM errors through direct indication of inaccuracy. A simple r…Debugging Prompts FrameworkA systematic framework for diagnosing and preventing typical errors of language models in instruction understa…BloomWise PromptingPrompt structuring method based on Bloom's taxonomy, which guides the LLM through six levels of cognitive thin…RESS (Decomposition-then-Synthesis)A method of decomposing a complex task into a sequence of simple analytical steps followed by synthesis. Inste…

All techniques · Failure modes and fixes