AI-CODE — AI Services for Continuous Trust in Emerging Digital Environments

Problem & Context

Most media professionals lack the understanding of how LLMs work to assess risks like hallucinations and biases. At the same time, the polished output of AI chatbots triggers cognitive biases — the “chat trap” and “beauty trap” — that make errors harder to spot. Existing AI literacy resources focus on generic prompt engineering and don’t address the deeper challenges of trustworthy media production.

PromptED — Guided Simulation & Prompt Playground

The centrepiece of EIPCM’s contribution is PromptED — an interactive coaching service based on experiential learning that takes a fundamentally different approach to AI literacy. Rather than passively teaching “how to prompt,” it lets media professionals observe, interact with, and learn from the inner workings of LLMs in a controlled environment.

Guided Simulator

Step-by-step interactive simulations using realistic journalistic tasks guide users through key concepts:

• How next-word prediction works — and how it leads to factual errors
• How prompt specificity, interaction style, and context reshape LLM outputs
• How prompt formulation induces confirmation bias in results
• How system temperature, system prompts, and RAG influence outputs
• How few-shot prompting shapes — and can be misused to manipulate — LLM behaviour
• An interactive reliability checklist for systematic evaluation against journalistic quality standards

Prompt Playground

A hands-on workspace where users apply what they learned to their own tasks — with real-time LLM outputs, interactive prompt optimisation controls, diff-based comparison views, PDF upload for RAG-based queries, and the same reliability checklist from the guided simulation.

AI Claim Debunker

A complementary module for verifying claims and generating counternarratives in 8 European languages, with highlighted source-document references.