Prompting technique

Generative Self-Aggregation (GSA)

LLMs Can Generate a Better Answer by Aggregating Their Own Responses, arXiv:2503.04104 (2025) · read the paper

What it does

A method for generating improved answers through self-aggregation: the model creates multiple different answers to a single question, then synthesizes them into a final answer by combining the strengths of each. It leverages the generative capabilities of LLMs instead of their weak evaluation skills, resulting in higher quality output.

Task typemulti step
Shapeiterative
Efforttwo or three steps
Addressesanswers that are close but wrong; the same prompt giving different answers; invented facts and citations
Published2025-03-06
AuthorsZichong Li, Xinyu Feng, Yuheng Cai et al.

When it helps

Use GSA when you need to obtain the highest quality answer to a complex question requiring deep analysis or creative solution. Particularly effective for tasks where a single answer may be incomplete or one-sided. The method suits research tasks, strategic planning, writing complex texts, and idea generation. Apply when quality matters more than speed, and you're willing to spend additional tokens generating multiple variants. Not suitable for simple factual

A prompt you can paste

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

Prompt
# Role and Goal

You are an AI assistant tasked with generating a high-quality, synthesized response by aggregating multiple, diverse initial attempts. Your goal is to leverage the strengths of each attempt and overcome their weaknesses to produce a superior final output.

# Initial Task

[Describe the core task or question here. For example: "Generate a marketing email announcing a new online course on personal finance."]

# Initial Attempts (Reference Solutions)

Below are several attempts to address the task. Analyze each for its strengths and weaknesses.

**Attempt 1:**
[Paste the first generated response here.]

**Attempt 2:**
[Paste the second generated response here.]

**Attempt 3:**
[Paste the third generated response here.]

[Add more attempts as needed, typically 3-5 total.]

# Aggregation Instructions

Analyze the strengths and weaknesses of all the provided attempts. Create a single, final, improved response that synthesizes the best elements from each.

**Requirements for the Final Response:**
1.  **Synthesize Strengths:** Identify and incorporate the most effective aspects (e.g., clarity, detail, tone, specific ideas) from each attempt.
2.  **Mitigate Weaknesses:** Address and correct the shortcomings observed in the attempts (e.g., vagueness, errors, missing information, inappropriate tone).
3.  **Add Value:** Enhance the synthesized content with your own insights or by filling gaps to create a more comprehensive and polished final output.
4.  **Format Appropriately:** Structure the final response according to the requirements of the initial task. [Specify desired format, e.g., "as a clear, concise marketing email with a strong call to action," or "as a well-structured itinerary table."]
5.  **Maintain Consistency:** Ensure the final response is coherent, consistent in tone, and directly addresses the original task.

**Example of specific requirements (adapt as needed):**
*   [Requirement 1: e.g., Include a compelling headline.]
*   [Requirement 2: e.g., Ensure the tone is professional yet engaging.]
*   [Requirement 3: e.g., Clearly outline the benefits for the target audience.]
*   [Requirement 4: e.g., Conclude with a clear and actionable call to action.]

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 and Goal

You are an expert planner and synthesizer. Your task is to create an optimal 3-day itinerary for a trip to Saint Petersburg in June, combining major attractions with a relaxed pace.

# Initial Task

Develop a balanced 3-day itinerary for Saint Petersburg in June. The plan should cover key attractions without feeling rushed, allowing for exploration and enjoyment.

# Previous Attempts (Reference Solutions)

Here are three draft itineraries. Analyze their strengths and weaknesses to create a superior final plan.

**Draft 1 (Focus on "Classics"):**
- Day 1: Hermitage Museum, Palace Square.
- Day 2: Peterhof (full day).
- Day 3: St. Isaac's Cathedral, Church of the Savior on Spilled Blood, Nevsky Prospekt walk.

**Draft 2 (Focus on Atmosphere and Walks):**
- Day 1: Boat tour on rivers and canals, New Holland Island.
- Day 2: Walk through the Petrograd Side, Botanical Garden.
- Day 3: Yelagin Island, dinner with a view of the Gulf of Finland.

**Draft 3 (Illogical and Overly Packed):**
- Day 1: St. Isaac's Cathedral, then a trip to Pushkin (Tsarskoye Selo), evening at the Mariinsky Theatre.
- Day 2: Hermitage Museum (2 hours), then Kunstkamera, then the cruiser Aurora.
- Day 3: Full day in Kronstadt.

# Aggregation Instruction

Analyze the strengths and weaknesses of all three drafts. Synthesize the best elements from each into a single, logical, and balanced 3-day itinerary.

**Consider the following:**
1.  **Logistics:** Group attractions located near each other within the same day. Avoid illogical travel across the city, as seen in Draft 3.
2.  **Balance:** Combine essential museums (like in Draft 1) with atmospheric walks and experiences (like in Draft 2). Dedicate a full day to a suburban attraction (e.g., Peterhof) but avoid over-scheduling other days.
3.  **Realism:** The plan should not be overly packed. Include time for lunch and rest.
4.  **Output Format:** Present the final itinerary as a table with columns for Day (Morning, Afternoon, Evening).

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