Site AI Chatbots: Direct Answers, No Chit-Chat
Users query site AI chatbots like search bars with short, imperfect prompts and expect instant, scannable answers without pleasantries, fluff, or overload—use truncated pyramid structure for essentials first.
Users Query Like Search: Short and Direct
In a study of 9 participants across 8 site-specific AI chatbots (2–3 per person), users skipped greetings, politeness, and full sentences, firing off minimal prompts like "Need a car for three people. Going to Orlando, FL, from Hampton, Georgia" (Turo), "What are the fees?" (Scouting America), or "Do you sell pavers?" (Home Depot). Initial queries used proper grammar, but follow-ups shortened to keywords as trust built, with typos ignored since bots understood intent. This mirrors search bar behavior: users minimize typing effort and demand quick, scannable replies, avoiding conversational overhead.
Eliminate Fluff for Tool-Like Efficiency
Top chatbots like Home Depot's Magic Apron succeed by ditching sycophantic phrases (e.g., "great question!") that inflate responses. Participants praised directness: one said, "I view these as tools... I just want the information." Brevity pairs with web-writing rules—short sentences (2–3 per paragraph max), lists, bolding, headers, and whitespace—crucial in tiny chat viewports. Scoutly (Scouting America) nailed this: for "What are the fees?", it listed "Youth: $85; Adult: $65" with fine-print notes, earning "informative and concise." Williams Sonoma formatted long tips (e.g., bubbly sourdough) into bulleted, bolded lists under headers like "Temperature Matters," preventing overwhelm. Avoid streaming dense text, which amplifies overload.
Truncated Pyramid: Answer Upfront, Expand on Demand
Extend inverted-pyramid writing with a truncated pyramid: deliver only the exact answer plus accuracy caveats first, hiding context/edge cases behind suggested follow-ups. Olympics chatbot failed by dumping skater details (name, scores, background) on "Who did a flip?"—user wanted just Ilia Malinin's name. ChatGPT does better with bullets first. For ambiguities, ask one quick clarification sparingly. Scoutly estimated startup costs specifically: National fee $85, uniform $50–$100, dues ~$100/year, gear $50–$150; total $300–$450—user appreciated the math and realism over generics.
Admit Limits Directly, Favor Specifics Over Vague Redirects
When unable, say so plainly without padding—e.g., Turo wasted time explaining site search instead of admitting no car-finding. Redfin improved from buried filter suggestions to auto-applying school-rating filters (9+). Vague replies like Turo's protection plans bred distrust; specifics win: for 2-week rental, give ranges (Premium $25–60/day, ~$595 total) vs. "check checkout." This shortens responses and builds trust, turning bots into reliable transaction tools audited via user testing.