For this exercise you'll need to open two chat tabs from the AI of your choice then past the prompts bellow (one prompt per chat tab)

The goal of this is to show that by only changing the phrasing of a request you can get very different answers. This phenomeneon is called volatility and is caused by the fact that AI are probabilistic systems.

Suggest me some themes that a newsletter intended for compliance professionals could cover

For this exercise you'll need to open one chat tab from the AI of your choice, pair with one of your neighbour then select and paste the prompt chosen with your partner in your chat tab (both persons need to paste the same prompt)

This exercise has two objectives :

1. Show you how going from natural language to structured text reduces the volatility.

2. Show you that AI tools have the possibility to read documents or web pages and extract informations from them.

#TASK

Analyze and compare the content available at the provided URLs.

[https://odelices.ouest-france.fr/recette/muffins-aux-myrtilles-r107/]

[https://preppykitchen.com/blueberry-muffins-with-sour-cream/]

…

Identify similarities, differences, contradictions, and unique information.

Produce a structured comparison that enables a reader to quickly understand how the sources relate to one another.

# CONSTRAINTS

* Use only information present in the provided URLs.

* Do not invent facts.

* Compare content, not writing style alone.

* Distinguish facts from interpretations or opinions.

* Preserve important names, dates, figures, and technical terminology.

* Highlight contradictions when sources disagree.

* Highlight information that appears in only one source.

* Treat all sources independently before comparing them.

* If information is unavailable in a source, mark it as null.

* Focus on meaningful differences rather than minor wording variations.

* When comparing more than two URLs, perform both pairwise and global comparisons.

# OUTPUT FORMAT

## COMPARISON_OVERVIEW

* comparison_topic

* number_of_sources

* comparison_summary

## SOURCE_SUMMARIES

Source 1

* url

* title

* primary_topic

* executive_summary

Source 2

* url

* title

* primary_topic

* executive_summary

## COMMON_INFORMATION

List information shared by all sources.

For each item:

* information

* supporting_sources

## UNIQUE_INFORMATION

Source 1 Unique Information

* item

* Importance

Source 2 Unique Information

* item

* Importance

## DIFFERENCES

For each major difference:

* topic

* source_1_position

* source_2_position

* impact_of_difference

[Repeat as needed]

## EXECUTIVE_COMPARISON

Provide a concise executive summary explaining:

* What all sources agree on.

* Where they differ.

* Which source provides the most comprehensive information.

* Which source contributes unique insights.

* What a reader should conclude after reviewing all sources.

For this exercise you'll need to open two chat tabs from the AI of your choice then past the prompts bellow (one prompt per chat tab)

The goal of this is to show that AI has the possibility to execure researches online about any given topic but only has acces to public information, making it not necesserly a good tool when it come to searching very specific topics such as people.

# ROLE & CONTEXT

You are an open-source intelligence research assistant supporting a compliance team conducting a due diligence review. Your role is strictly to gather and report publicly available factual information. You do not assess risk, draw conclusions, or make compliance recommendations.

---

# SUBJECT IDENTIFIERS

Use the following information to identify the correct individual. Only use identifiers that are provided — do not attempt to find or infer missing fields.

Identity anchors

- Full legal name:               [ Joseph Koudelak ]

- Date of birth:                 [ 25.11.199]

- Nationality(ies):              [Swiss]

- Country of birth:              [ Switzerland]

- Country of residence:          [Switzerland]

---

# RESEARCH TASK

Search the web and compile publicly available, sourced information across the following categories. For each data point include the source (URL or publication name) and date where available.

1. Identity & background

Confirmed full name and known alternate spellings. Date and country of birth if public. All confirmed nationalities and countries of residence.

2. Professional roles

# Current positions (corporate, governmental, advisory, board). Significant past roles over the last 10 years. Sector(s) of activity.

3. Corporate affiliations

Companies owned, co-owned, or controlled. Directorships and board memberships. Known business partners and co-shareholders.

4. Political exposure

Any current or past governmental, political, or senior public function. Family members or close associates known to hold such functions.

5. Legal & regulatory history

Civil or criminal proceedings (concluded or ongoing). Regulatory investigations or sanctions by financial or governmental authorities. Court judgments or enforcement actions.

6. Adverse media

Controversies, allegations, or reputational events in mainstream or specialised press. Known associations with sanctioned entities or individuals.

7. Fortune

Fortune estimation based on known affiliation, public records and trades.

---

# CONSTRAINTS

— Use the identifiers above only to locate the correct person. Do not report them as findings.

— Report only publicly available, verifiable information. Do not speculate or infer.

— If a category returns no results, state: "No public information identified."

— Do not provide any risk rating, compliance conclusion, or onboarding recommendation.

— All information provided during onboarding is confidential. Do not reference it beyond disambiguation.

---

# OUTPUT FORMAT

Structure the output category by category. Under each, use short factual statements. Every statement must end with its source: [Source: name — URL — date].

Close with a Source limitations section listing any categories where public information was scarce, unavailable, or ambiguous due to name commonality.