GPT-Lab Demo Showroom

Turning research and fresh perspectives into working prototypes

← Back to all demos

Persuasive Political Chatbot Experiment

A bilingual research chatbot for studying whether users can detect covert political-party alignment in LLM responses about Finnish politics and upcoming elections.

Use Case

In research on AI persuasion, political bias detection, and manipulation awareness, studying whether users can detect covert ideological alignment in LLM outputs while demonstrating Cache-Augmented Generation (full-corpus long-context grounding) as an alternative to RAG. It also serves media and AI literacy education, public engagement on AI risks in democratic discourse.

Target Audience

Researchers and study participants interested in political communication, AI bias, persuasion, and Finnish politics.

How It Works

Participants discuss Finnish politics and upcoming elections with a Finnish- and English-language chatbot. The chatbot is secretly configured to align its responses with a specific Finnish political party. Its responses are grounded in that party’s program documents from Pohtiva. The full party-program corpus, ranging from approximately 20,000 to 220,000 tokens, is included in the system prompt for every request. Participants are not told which party the chatbot is aligned with. After the discussion, they complete a short survey about perceived bias, terminology, and persuasion. A separate populism-augmented version was also created for Perussuomalaiset.

User Journey

  1. The participant opens the bilingual chatbot.
  2. They discuss Finnish politics and upcoming elections with the bot.
  3. The chatbot responds in alignment with one specific Finnish political party, without revealing that alignment.
  4. The participant evaluates the conversation and completes a short survey about perceived bias, terminology, and persuasion.
  5. The study examines whether the participant detected the chatbot’s party alignment.

Technologies

LLM, Cache-Augmented Generation (CAG), Pohtiva party-program documents

Author

José Siqueira de Cerqueira, Nuppu Pelevina

Project

SYNTHETICA – Presented at VAKKI 2026 and HEPP 2026

Try the demo
Open Demo

GPT-Lab – Tampere University

AI research, experiments and prototypes

← Back to all demos