Applying prompts to GenAI/LLMs

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—gathe...

I recently found myself watching a recording of Jean-Michel Jarre’s 2025 concert in Seville. Alongside the music and spectacle, there was a lot of discussion around AI and AI-generated content. The concert can, at the time of writing, be found on  https://www.arte.tv/en / well worth a look. That set me thinking, not in a grand theoretical way at first, but in a simple, curious way: what would ChatGPT visualise if I asked it to turn Oxygène into a single cartoon image? Not “make a copy of the album cover”. Not “recreate Jean-Michel Jarre’s visual style”. Just: take the word, the title, the atmosphere it suggests, and see what happens. That distinction mattered to me. Oxygène already has a famous visual identity, and the point of the experiment was not to imitate it or produce a substitute. It was more about asking what an AI system does when given a culturally loaded musical reference, a single evocative word, and a loose creative brief. Would it lean towards the known album im...

B rief recap In an earlier post, “Same prompt, four AIs: why answers aren’t always the same” , I looked at what happened when the same prompt was given to four different LLMs. Unsurprisingly, perhaps, the answers were not identical. That raised an interesting follow-on question: what was the prompt actually trying to do? The answer is that it was trying to support a form of thematic analysis. In that case, the object of analysis was an academic profile, including my Google Scholar profile. The version I discuss here is a modified prompt, used with Claude.ai, where I uploaded a CV and asked the system to find other appropriate public resources connected with a named person and institution. The aim was not simply to summarise the CV, but to identify visible and less visible themes across a whole career. Why use thematic analysis? Thematic analysis is widely used by academics, especially in qualitative research. Braun and Clarke’s well-known paper, Using thematic analysis in psy...

Same Prompt, Four AIs — Why the Answers Aren’t the Same The differences aren’t just in the answers—they’re in the thinking Generative AI tools are often discussed as if they were interchangeable—different interfaces delivering broadly similar outputs. However, when applied to complex intellectual tasks, meaningful differences begin to emerge. To explore this, I ran the same academically rigorous prompt through four leading systems—Claude, ChatGPT, Google Gemini, and Copilot. The task required a full thematic analysis of a researcher’s career using the framework developed by Virginia Braun and Victoria Clarke . What followed was not simply variation in output, but variation in how each system approached the act of analysis itself. Same Input, Different Interpretations At a high level, the experiment is simple: One prompt → Four models → Four distinct approaches What changes is not the instruction, but how each system: Interprets the task Handles uncertainty Applies methodology Defines ...

Analysing Documents with AI: A Multi-Stage Prompting Approach What happens when a data scientist and a statistician are asked to challenge each other's reading of the same paper? The coding-focused prompting technique described in  a previous post  has a natural sibling: the same multi-stage, dual-persona approach works remarkably well for document analysis. Instead of building software through iterative expert review, you are analysing a piece of work — a research paper, a dataset report, a literature review — and subjecting it to exactly the same kind of structured, adversarial scrutiny. This post walks through how that adapted prompt works, why the underlying techniques make it more than a glorified summarisation tool, and what happened when it was tested on a social network analysis of co-authorship patterns in an academic repository. Why Not Just Ask for a Summary? A single-shot summary prompt is fine if you want a précis. But analysis is different. Analysis requires aski...

Building Software with AI: A Multi-Stage Prompting Approach What if you could have two seasoned 'experts' sitting beside you as you built software (you can add more if you like) — questioning your assumptions, challenging your logic, and pushing your code toward something genuinely better? That's exactly what this prompting approach tries to recreate. This post walks through a structured, multi-stage prompt technique for AI-assisted software development. It combines three well-established prompt strategies — iterative questioning, persona simulation, and collaborative reasoning — into a single, coherent workflow. Whether you're new to prompt engineering or looking to level up your practice, this approach offers a repeatable pattern worth trying. Why This Approach? It's tempting to use AI for code in a single shot: describe the problem, get some code, move on. The results are often functional, but sometimes shallow — the AI has no real understanding of your cont...

In a companion post on context stacking, I came across an idea that stayed with me — and I wanted to explore it further. This piece isn’t just about the final blog produced, but about the process behind creating it using generative AI (specifically, Claude.ai). Rather than using AI as a writing shortcut, I used it as a thinking partner — one that could challenge my assumptions, test my reasoning, and help strengthen the argument before anything was finalised. What emerged was a structured workflow (shown below) that others can adopt when using AI to improve the rigour, not just their output. And it all starts with setting the context and the audience and telling the generative AI to pick it apart. The process: Before using Generative AI: Two drafts were produced, and the second draft went through this final process, as described by claude.ai here. Here's the workflow we followed: 1. Critical Reading and Initial Diagnosis We started with a close reading of your original draft, ...