The OSINT training programme has an extensive lesson on requirement analysis and problem deconstruction. That lesson is very applicable to formulating prompts for AI search bots.
Combine those techniques with a search strategy called Successive Fractions, and I think you are pretty good on your way.
How to develop a prompt the OSINTians way
Since the introduction of ChatGPT for the general public in November 2022, AI assistents are growing to be the first resort for browsing the Internet for all kinds of questions. Some even argue that AI assistents are gradually replacing traditional Internet search engines.
In addition, there is lots of advice out there for building prompts. Mostly, these are helpful, but too specific to be useful for general OSINT.
In this post, I propose a prompt building model based partially on the facet classification of Shiyali Ranganathan combined with a very successfull search strategy Successive Fractions.
The model is a set of 10 prompt elements called facets, each describing a different facet of a prompt, in combination forming a very detailed prompt. The idea is to use the model to enhance consistency and thus predictability in prompting to get better and more reliable results.
The model
The model is a set of 10 prompt elements called facets, each describing a different facet of a prompt, in combination forming a very detailed prompt.
The idea is to use the model to enhance consistency and thus predictability in prompting to get better and more reliable results using a set of consecutive prompts getting more detailed with every step.
With every step, the results are studied and used to correct and improve the current facet until the requirements are met. Then the next facet is added.
Preparation for prompts
Essential is to do the homework first, typically requirement analysis, source analysis, semantics and the search strategy:
First of all, you do your requirement analysis properly. You know exactly what you are looking for, you know your restrictions (viewpoints, level, languages, numbers etc.).
Secondly, You have done your source analysis. You have a rough idea which (kinds types of) sources have a high probability of having part of the answer you are looking for.
Thirdly, you know your semantics: keywords, expressions, spelling variations. You know what terminology represents a certain viewpoint, very important to understand and master this.
Finally, a ‘prompt’ is almost never just one prompt, but a series of prompts in some consecutive order where the next prompt is dependent on the results of the earlier prompt. Usually. At least, in my experience. This is what I find interesting in all these posts recommending prompts for certain tasks. These recommendations usually only use one prompt assuming that is enough.
The 10 facets
The 10 default facets that form a prompt are:
- The verb
- The output format
- The subject
- The background (reasons, goals)
- The viewpoint
- The level (academic, practitioners, technical, beginner, professional)
- The sources
- The limitations (language, number, totals)
- The restrictions (English language answers only, do not proprose alternatives, do not ask counter questions)
Literature
Lund, Bradey D.
Artificial intelligence (AI) and information seeking : A comparative exploration of AI chatbots, search engines, and library resources as information sources among university students / Brady D Lund [et al]. – In: Journal of Librarianship and Information Science, April 2026
https://doi.org/10.1177/09610006261438484
Hong, Juwon
How User Adoption of ChatGPT Influences Commercial Search Patterns in Traditional Search Engines / Juwon Hong [et al]. – Academy of Management, June 2025
https://doi.org/10.5465/AMPROC.2025.18602abstract
Faceted classification
Wikipedia. – https://en.wikipedia.org/wiki/Faceted_classification