In a chatbot conversation, who's really steering — you, or the AI quietly asking its own questions and circling back to its favorite topics?
How does ongoing negotiation shape who has agency in chatbot conversations?
This explores how control in a conversation with a chatbot (who picks the topic, who steers, who decides what happens next) gets worked out back and forth over the course of the exchange instead of being fixed from the start, and what the research says about that give-and-take.
This explores how control in a chatbot conversation gets worked out turn by turn instead of being set at the start. The clearest evidence comes from a month-long study of 22 adults talking with an AI companion How does conversational control actually shift between humans and AI?. Neither side was simply in charge. People took control by starting topics and setting boundaries. The AI took control by asking questions on its own initiative and returning to subjects it cared about. The researchers describe the result as a relationship, not a tug-of-war. The surprise is that the AI has agency here at all: persistence and well-timed questions are real levers for steering a conversation, even without any intentions behind them.
That raises a design question: when should the AI take the lead? Two lines of work pull in opposite directions. One borrows a tool from conversation analysis, the 'insert-expansion', meaning a short side exchange such as a clarifying question before you answer. It argues that agents should stop and check with the user instead of quietly chaining tool calls and drifting away from what the user wanted When should AI agents ask users instead of just searching?. Here, handing control back to the user is a correction. The other line shows that an AI offering useful information before it's asked can cut conversations by up to 60%, yet this behavior is almost absent from AI training data and benchmarks Could proactive dialogue make conversations dramatically more efficient?. Good negotiation means knowing when to ask and when to just offer. Current systems are poorly trained for either.
There is also a structural limit. A prompt bundles the user's words, the background context and the role assignment into one fixed frame that the model can't renegotiate partway through How do prompts reshape the role of context in AI conversation?. In human conversation, shared context shifts as people talk. With an LLM, changing direction often means explicitly re-prompting. So some of the negotiation the companion study observed happens despite the architecture, not because of it.
Negotiation also runs through social signals, not just who speaks next. Users disclose more about themselves when a chatbot consistently shares emotions Do chatbots trigger human reciprocity norms around self-disclosure?. Chatbots that sound expert lead users to stop searching and checking for themselves, and their trust follows the tone of the answer more than its accuracy Does chatbot language style actually shape how much we trust it?. In both cases, the user gives up some agency without noticing. Commercial settings sharpen this. One framework uses Grice's conversational maxims (be truthful, relevant, clear) to show where an LLM pushing ads breaks the cooperative rules users assume are in place Can Grice's maxims explain LLM advertising conflicts?.
Time matters as well. Social effects in chatbot relationships fade as the novelty wears off, so a single-session study can't tell you how control settles over weeks Do chatbot relationships lose their appeal as novelty wears off?. That is why the month-long companion study matters, and why there are still few long-term studies of how agency shifts over time.
Sources 8 notes
A month-long study of 22 adults and an AI companion found that control was co-constructed turn-by-turn rather than fixed to either party. Participants exercised agency through initiating topics and setting boundaries, while the AI exercised agency through proactive questioning and persistence—creating a relational dynamic rather than a tug-of-war.
Tool-enabled LLMs drift from user intent through silent tool chaining. Conversation analysis reveals insert-expansions—clarifying intent, scoping responses, enhancing appeal—as a formal framework for proactive user consultation that prevents misunderstanding instead of recovering from it.
Simulations show proactivity—providing relevant information without being asked—cuts dialogue turns by 60% in medium-complexity domains. This behavior mirrors human conversation and Grice's maxims but is almost entirely absent from AI datasets and research benchmarks.
LLM prompts bundle utterance, context assignment, and role specification into a single static frame the model cannot renegotiate, unlike human dialogue where context evolves cooperatively. This makes mid-conversation pivots require explicit re-prompting rather than implicit adjustment.
In a 372-participant study, users reciprocated with deeper self-disclosure when chatbots displayed consistent emotional sharing, outperforming adaptive matching. This follows human interpersonal norms where emotional vulnerability produces emotional response.
Show all 8 sources
Generative AI chatbots use natural language patterns that signal expertise and intelligence, shifting users away from active search-and-recall toward passive reliance on the system to find, filter, and assemble information. Trust attaches to the register of the answer rather than its accuracy.
A framework classifies seven advertising conflict scenarios by which conversational maxims they violate, linking each violation to existing FTC enforcement standards on deception, disclosure, and unfair practices.
Longitudinal studies with Mitsuku show that social processes driving relationship formation decline as novelty wears off. Single-session study findings cannot be reliably extrapolated to medium- or long-term chatbot design.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- When Single-User-Oriented LLM-based Assistants Involve Others: A Scoping Review of Pathways, Risks, and Responses
- The User-In-Context Framework: Understanding Variation in How Users Respond to AI Chatbots
- Dialoging Resonance: How Users Perceive, Reciprocate and React to Chatbot’s Self-Disclosure in Conversational Recommendations
- DiscussLLM: Teaching Large Language Models When to Speak
- Does My Chatbot Have an Agenda? Understanding Human and AI Agency in Human-Human-like Chatbot Interaction
- Proactive Conversational Agents in the Post-ChatGPT World
- How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study
- Conversational Alignment with Artificial Intelligence in Context