Does being polite to AI change what you get back? ??
Our paper, “Cooking Up Politeness in Human-AI Information Seeking Dialogue,” accepted at CHIIR 2026, investigates this question.
In a cooking-assistant scenario, we investigated how user politeness affects AI responses.
? What we did
We analyzed 30 user–assistant dialogues collected in a Wizard-of-Oz experiment and identified four user politeness clusters (from hyperpolite to hyperefficient).
We scaled up to 18,000 simulated conversations across five politeness profiles (incl. impolite input) and three open-weight models.
? What we found
Politeness is not just cosmetic, it systematically affects:
- Response length
- Informational gain
- Efficiency
- Energy consumption
Engagement-seeking prompts produced longer and information-richer replies than hyper-efficient prompts — but with lower informational density.
Impolite prompts generated less efficient answers with fewer information per watt-hour compared to polite input.
? Why this matters
This highlights politeness as a fairness and sustainability issue.
Different conversational styles may advantage or disadvantage users.
And “being polite” may come with hidden energy costs.
?? Find the paper incl. design implications for inclusive and resource-aware information agents here: https://arxiv.org/abs/2601.09898 (externer Link, öffnet neues Fenster)
Authors:
By David Elsweiler (Information Science Regensburg) in collaboration with linguist Christine Elsweiler and her talented student worker Anna Ziegner (Department of English, Universität Innsbruck)
This interdisciplinary work bridging human-centered computing and the humanities shows that linguistic theory can help us better understand human-AI interactions and inform the design of AI-based information agents.
We are looking forward to presenting this work at CHIIR 2026 in Seattle. ?
#HumanAIInteraction #Politeness
#FairAI #SustainableAI
#ConversationalAI #GenerativeAI
#InformationInteraction #InformationSeeking
#InformationSearch #InteractiveSearch
#EmpiricalStudy
#SearchResearch #AIResearch
#Interdisciplinarity
#Linguistics #AI #LinguisticsMeetsAI
#ACM #SIGIR #ConferenceOnHumanInformationInteractionAndRetrieval #CHIIR #CHIIR2026
#ResearchSuccess #ResearchPaperAccepted
#InformationScienceRegensburg #StayInformed