Consumer Behaviour
Prompt-shaping behaviour
How users refine the language of their prompts mid-conversation to get a better answer. Reveals a shift from keyword behaviour (typing fewer terms) to conversation behaviour (typing more, more naturally).
What it is
Prompt-shaping behaviour is the way users refine the wording of their prompts mid-conversation to steer an assistant towards a better answer. It marks a shift from one-off keyword entry to iterative conversation.
Why it matters
These refinements reveal the true intent and the gaps in the first answer, giving a clearer picture of need than a single query and signalling which content actually resolves a request.
How it works
It is observed by tracking how a prompt evolves across turns: added constraints, requested formats, clarifications, and corrections that move the conversation towards a satisfying result.
When it applies
It applies during multi-turn interactions with a generative assistant rather than single-shot search.
Examples
- A user follows a broad question with make it shorter and for beginners.
- Someone adds a constraint such as only options under a set budget after seeing the first reply.
- A searcher reframes a vague request into a specific scenario to get a targeted answer.
How it is measured
- Average number of refinement turns per conversation
- Rate of added constraints or format requests across turns
- Correction or clarification frequency within a session
- Turn at which the user accepts the answer and stops
Related terms in Consumer Behaviour
- Agentic browserA browser (or browsing layer) that uses an LLM agent to interpret pages, summarise content, and take actions on behalf of the user. Arc Search, Perplexity Comet, Browser Company's Dia, Dia browser, and similar.
- Brand demandSearch volume for branded terms. In an AI-search world, brand demand is the single strongest moat, generic queries are absorbed by AI Overviews and ChatGPT, while branded queries route users directly to brand properties.
- Consumer behaviour signalObservable patterns in how users phrase queries, refine searches, and choose answers, used by both ranking systems and generative models to infer intent and quality.
- Job-to-be-done (search)The functional outcome a searcher is trying to achieve when they issue a query. Distinct from the literal query text. The unit of analysis for intent-aligned content strategy.