# Trust, delegation, and the autonomy paradox

> Users want AI to do more but resist ceding full control, creating a persistent tension in the trust-delegation relationship. Trust varies significantly by age and is built more through competence than personality.

Users want AI to do more but resist ceding full control, creating a persistent tension in the trust-delegation relationship. Trust varies significantly by age and is built more through competence than personality.

## Do the British public trust search engine AI summaries?

**Publisher:** YouGov

[Open the source](https://yougov.com/en-gb/articles/51644-do-the-british-public-trust-search-engine-ai-summaries) ↗

> Britons more likely to find search engine AI summaries trustworthy than not. Some <strong>42% find these summaries trustworthy, while 35% do not</strong>. This does change along age lines: younger Britons are more likely to put their faith in search engine overviews, while older Britons are more sceptical.

## Prioritize Smarts over Sentience to Increase Trust with AI

**Publisher:** Nielsen Norman Group

[Open the source](https://www.nngroup.com/articles/smarts-emotion-trust-ai) ↗

> Final trust ratings were correlated with advice taking. Participants who reported trusting ChatGPT were more likely to change their answers to match the AI's.

> <strong>Perceptions of intelligence were positively related to taking ChatGPT's advice</strong>, while perceived emotionality was slightly negatively related.

> Users are more likely to follow AI advice if they perceive AI as smart rather than emotional; building trust requires competence and reliability more than emotional rapport.

## GenAI for Complex Questions, Search for Critical Facts

**Publisher:** Nielsen Norman Group

[Open the source](https://www.nngroup.com/articles/ai-search-infoseeking/?utm_source=Alertbox&utm_medium=email&utm_campaign=content&utm_content=articles) ↗

> Search was commonly used as a validation mechanism and to get specific, trustworthy details. This behavior reflected <strong>users' mental model of genAI as helpful but fallible</strong>.

> Several participants said they would not fully trust pricing information provided by AI and still wanted to confirm exact numbers through traditional search.

> <strong>Citations do not fully solve the trust problem.</strong> Even when tools provided source links, some participants struggled to understand which claims were supported by sources and which were uncited synthesis.

## Mental Models of Generative AI Chatbot Ecosystems

**Publisher:** International Conference on Intelligent User Interfaces

[Open the source](https://dl.acm.org/doi/10.1145/3708359.3712125) ↗

> <strong>Chatbot as a Key Player:</strong> participants saw the chatbot as a separate actor, and trust in the parent company did not automatically extend to the chatbot.

> <strong>Chatbot as a Medium:</strong> participants viewed the chatbot as an interface to the parent company, and trust in the parent company was more likely to transfer to the chatbot.