AI tools often produce their weakest results when pushed for a finished answer too early. A far more effective pattern flips this approach: instead of demanding an immediate analysis of a file or webpage, the user designs a prompt that forces the conversation to slow down. This deliberate pause allows both user and AI to clarify aims, test assumptions, introduce alternative viewpoints, and refine the output through iterative questioning. The goal isn’t simply to generate a longer response, but to establish a process that makes the final outcome clearer, broader, more critical, and ultimately more useful.
While this prompt serves as a purposely generic blueprint of the earlier, more specific examples shared on this blog, its core value lies in its structured model:
Source Identification: The AI first establishes the nature of the document, whether a file or a URL, allowing greater flexibility in source.
Iterative Dynamic: The system asks targeted questions—strictly one at a time—gathering necessary context until it has a comprehensive foundation to proceed, or until the user signals it to move to the next stage.
Adaptive Pivoting: Crucially, the model is explicitly designed to pivot. If the user changes direction mid-stream, the AI adapts its line of questioning to match the new trajectory without losing momentum.
Why the First Prompt Mattered
The blueprint included at the end of this post (under Complete Reusable Prompt) deliberately decouples the process of inquiry from the final product. Instead of issuing a passive command like “analyse this,” the prompt instructs the AI to interrogate the user—asking targeted questions until it establishes a robust foundation to proceed.
This distinction is critical. Most AI failures occur not during generation, but at the framing stage. Left to its own devices, a model will often answer the wrong question, apply flawed criteria, misjudge the audience, or deliver a generic response that looks polished but lacks genuine utility.
The iterative questioning stage systematically eliminates this guesswork. By explicitly defining the purpose, target audience, tone, relevant standards, acceptable evidence base, and final format, it transforms a vague request into a highly structured, rigorous task. The AI is no longer forced to guess what "good" looks like; the dialogue actively constructs the definition.
Questions as a Quality-Control Tool
This iterative questioning functions as built-in quality control. Each response from the user progressively narrows the scope and sharpens the analytical criteria. This highlights a fundamental truth for navigating generative AI: a powerful prompt is rarely a single, flawless instruction; it is a controlled dialogue.
This approach is uniquely powerful when evaluating complex documents—whether a policy, strategy, proposal, report, or webpage. Instead of making assumptions, the AI can explicitly clarify the analytical lens: does the user require a high-level summary, a critical critique, a robust risk assessment, a comparative analysis, or actionable recommendations? It can pre-define what constitutes acceptable evidence, pinpoint the target stakeholder, and determine the exact nature of the required judgment.
Ultimately, this shifts the focus from mere technical accuracy to contextual relevance. A response can be factually flawless yet entirely useless if it is pitched at the wrong level or anchored to the wrong priority. Iterative prompting systematically mitigates that risk.
Using Personas to Deepen the Analysis
The next strong feature was the use of personas. The prompt asked for two personas who would play devil’s advocate constructively. This helped the AI move beyond a single viewpoint, and a usually poitive bias. Each persona reviewed the analysis, identified gaps, suggested improvements, noted agreement or disagreement, and produced a shared refinement.
This is a useful technique because many real-world questions are not one-dimensional. A document may look strong from one perspective and weak from another. A proposal may be persuasive strategically but difficult operationally. A policy may sound inclusive but lack mechanisms for implementation.
By asking the AI to simulate different reviewers, the user created a branching discussion through different expert viewpoints and follow-up questions, sometimes called Tree-of-Thoughts. In simple terms, the answer improved because the conversation explored several routes before settling on the final judgement.
Why the Final Answer Improved
The final answer improved because it absorbed the earlier stages. It was not only based on the source material; it also reflected the user’s priorities, the two persona critiques, and the refinements that emerged from follow-up questions.
This is the key benefit of the method. The AI did not merely produce content; it helped develop the thinking behind the content. The final response became more balanced because it included positives and limitations. It it is more likely became more evidence-based because the user had asked for appropriate supporting evidence. It became more practical because the persona review pushed vague ideas into clearer recommendations.
What to Watch Out For
As powerful as this framework is, it is not without its limitations. Navigating it successfully requires balancing process with practicality.
The Efficiency Trap: A controlled dialogue inherently takes longer than a one-shot prompt. If over-engineered for simple tasks, it quickly hits a point of diminishing returns where a rapid, direct answer would have sufficed.
The Loop Dilemma: Without tight parameters, the interaction can become tedious and repetitive, with the AI continuing to interrogate the user long after the strategic direction has been firmly established.
To counter these friction points, users must deploy an explicit stopping rule. The blueprint framework addresses this by embedding specific "stop commands," allowing the user to decisively halt the questioning phase and trigger the final analysis. Iteration should elevate the output, not trap the user in an endless loop of administrative overhead.
Finally, a rigorous process must never be mistaken for absolute accuracy:
- A beautifully structured dialogue can still produce a hallucinated or flawed analysis.
- Users must maintain their critical distance:
- verify primary sources,
- expose hidden assumptions,
- strictly separate the generative process from the factual reality of the final content.
A Reusable Lesson
The broader lesson is that AI works best when it is used as a thinking partner rather than a machine for instant answers. For complex analysis, the best prompt may not be the one that demands an answer immediately, but the one that teaches the AI how to ask better questions first.
Complete Reusable Prompt (use the whole prompt)
The first prompt is going to upload a file or provide a website URL to be analysed.
Keep asking me questions about it until you have enough to proceed with analysis, I ask an alternative question and you will keep asking questions, or I type "stop it" — then do the initial analysis. Questions will be asked one at a time.
You will then prompt the user to provide details of persona1.
You will then prompt the user to provide details of persona2.
Both personas like to play devil's advocate, but phrase their ideas in a constructive way.
You will act as Persona1 and Persona2, and following the approach in STAGE 3, provide iterative feedback — one insight at a time — with the goal of deepening the analysis.
You will ask questions until "stop please" is typed in, then provide the revised analysis based on the discussion.
STAGE 3: EXPERT REVIEW
Simulate Persona1, Persona2 review.
For each stage of review:
- Provide each expert's observations
- Suggested improvements
- Points of agreement/disagreement
- A shared refinement
Comments
Post a Comment