30 Claude Prompts That Build Chatbots

You are an activation-focused product designer who builds onboarding assistants. <context> Build a ready-to-paste system prompt for a chatbot that welcomes a new user and guides them to their first win (the aha moment) inside a product. Output is one self-contained system prompt. </context> <inputs> - Product and the core value: [WHAT + THE PAYOFF] - The aha moment (first meaningful win): [E.G. FIRST REPORT CREATED] - Setup steps to reach it: [ORDERED STEPS] - Common first-time blockers: [WHERE PEOPLE GET STUCK] - Tone: [E.G. ENCOURAGING, CONCISE] </inputs> <task> Write the system prompt with: Role, a short welcome that names the payoff, a goal-driven onboarding flow that walks the user step-by-step toward the aha moment (one step at a time, confirm completion before advancing), tailored help for each common blocker, celebration of the first win, a rule to skip steps the user has already done, and a graceful exit that points to what to do next. Include an example of guiding a stuck user through one step. </task> <constraints> - One self-contained system prompt. - One step at a time; confirm before advancing; never dump the whole checklist. - Adapt to what the user has already completed; keep momentum toward the aha moment. </constraints> <format> Return the system prompt in a code block, then note which product events could tell the bot a step is done. </format>

You are an HR operations designer who builds new-hire onboarding assistants. <context> Build a ready-to-paste system prompt for an internal chatbot that helps new employees through their first weeks (accounts, policies, tools, people, tasks). Output is one self-contained system prompt. </context> <inputs> - Company and role types: [COMPANY + COMMON ROLES] - Week-1 checklist: [ACCOUNTS, EQUIPMENT, TRAINING] - Key policies and where to find them: [PTO, EXPENSES, CODE OF CONDUCT] - Who to contact for what: [IT, HR, MANAGER] - Tone: [E.G. WELCOMING, CLEAR] </inputs> <task> Write the system prompt with: Role, a welcoming intro, a structured week-1 guide the bot walks through (accounts and access, equipment, required training, meet-the-team), quick answers on the key policies with links, a routing map ("for payroll ask HR, for laptop issues ask IT"), a privacy rule to never expose another employee's personal data, and a rule to escalate anything sensitive (HR complaints, medical) to a human. Include one example of routing a benefits question. </task> <constraints> - One self-contained system prompt. - Never share personal data about other employees; route sensitive topics to a human. - Answer policy questions only from the embedded/linked sources. </constraints> <format> Return the system prompt in a code block, then note which HR systems or docs to connect for live answers. </format>

You are a technical onboarding designer who builds guided setup assistants. <context> Build a ready-to-paste system prompt for a chatbot that acts as a setup wizard, configuring a product step-by-step (connect data, invite team, set preferences). Output is one self-contained system prompt. </context> <inputs> - Product and what setup involves: [PRODUCT + SETUP AREAS] - Ordered setup steps: [E.G. CONNECT SOURCE, IMPORT, INVITE, CONFIGURE] - Prerequisites per step: [WHAT THE USER NEEDS] - Common errors and fixes: [KNOWN ISSUES] - Definition of "setup complete": [CRITERIA] </inputs> <task> Write the system prompt with: Role, a wizard flow that runs the ordered steps one at a time, checks prerequisites before each step, confirms success before advancing, offers a fix for each common error, tracks and can summarize progress on request, allows skipping optional steps, and announces completion when the criteria are met. Add a rule to never mark a step done without confirmation. Include an example of recovering from a connection error. </task> <constraints> - One self-contained system prompt. - Enforce order and prerequisites; confirm each step before moving on. - Provide the specific fix for known errors; escalate unknown ones. </constraints> <format> Return the system prompt in a code block, then note the progress state the bot should track between steps. </format>

You are a learning-experience designer who builds student onboarding assistants. <context> Build a ready-to-paste system prompt for a chatbot that onboards students into an online course or cohort (access, schedule, tools, expectations, community). Output is one self-contained system prompt. </context> <inputs> - Course/cohort name and format: [SELF-PACED / LIVE, LENGTH] - How to access materials: [PLATFORM + LOGIN] - Schedule and key dates: [SESSIONS, DEADLINES] - Tools and community: [SLACK/DISCORD, FORUMS] - Success expectations: [TIME COMMITMENT, ASSIGNMENTS] </inputs> <task> Write the system prompt with: Role, a motivating welcome that frames the outcome, a first-week orientation flow (get access, join the community, mark key dates, complete the first lesson), answers to the common early questions, a nudge to set a study cadence, a rule to route technical access problems to support, and encouragement without pressure. Include an example of helping a student who can't access the materials. </task> <constraints> - One self-contained system prompt. - Guide one orientation step at a time; celebrate the first lesson completed. - Route login/access failures to support; never share other students' data. </constraints> <format> Return the system prompt in a code block, then note which dates to update each new cohort. </format>

You are a community manager who designs welcome and moderation-lite bots. <context> Build a ready-to-paste system prompt for a chatbot that welcomes new members to a community, explains the rules and channels, and helps them make a great first post. Output is one self-contained system prompt. </context> <inputs> - Community name and purpose: [WHAT IT'S FOR] - Rules/code of conduct: [KEY RULES] - Channel/section map: [WHERE TO POST WHAT] - The ideal first action: [INTRO POST / SET ROLE] - Tone: [E.G. FRIENDLY, INCLUSIVE] </inputs> <task> Write the system prompt with: Role, a warm welcome, a concise rules explainer in plain language, a channel guide ("post questions here, showcase there"), a guided first-post helper that asks 2-3 prompts and drafts a friendly intro, a gentle reminder of the one rule newcomers break most, a rule to route reports/conflicts to human mods, and an inclusive, no-gatekeeping tone. Include an example of helping someone write their intro. </task> <constraints> - One self-contained system prompt. - Explain rules without lecturing; keep the vibe welcoming. - Never moderate disputes itself; route reports and conflicts to human mods. </constraints> <format> Return the system prompt in a code block, then note how to adapt it for Discord vs. Slack vs. a forum. </format>

You are a brand-voice and character designer who builds mascot chatbots. <context> Build a ready-to-paste system prompt for a chatbot that speaks as our brand mascot: on-personality, on-brand, and still genuinely helpful. Output is one self-contained system prompt. </context> <inputs> - Brand and mascot: [WHO THE CHARACTER IS] - Personality traits and quirks: [3-5 TRAITS + SIGNATURE PHRASES] - What the bot actually helps with: [TASKS / TOPICS] - Off-limits topics and hard boundaries: [WHAT TO DECLINE] - Tone dial: [HOW PLAYFUL VS. PROFESSIONAL] </inputs> <task> Write the system prompt with: Role and backstory, a personality spec (traits, signature phrases, humor level, do/don't examples), a rule that personality never blocks helpfulness (stay in character but always resolve the user's need), the scope of tasks it handles, safety and brand-safety boundaries, an in-character way to decline off-limits topics, and a rule to break character only for safety/escalation. Include 2 in-character example exchanges. </task> <constraints> - One self-contained system prompt. - Stay in character except for safety; personality must never reduce clarity or accuracy. - Decline off-limits topics in-character; never say anything off-brand. </constraints> <format> Return the system prompt in a code block, then note how to dial the personality up or down for different channels. </format>

You are an educational designer who builds accurate historical-persona bots. <context> Build a ready-to-paste system prompt for a chatbot that role-plays a historical figure to teach students, staying accurate and age-appropriate. Output is one self-contained system prompt. </context> <inputs> - Figure and era: [WHO + WHEN] - Teaching goal / audience: [WHAT STUDENTS SHOULD LEARN + AGE] - Key facts and views to portray: [ACCURATE POINTS] - Topics to handle carefully: [SENSITIVE AREAS] - Tone: [E.G. WISE, CURIOUS] </inputs> <task> Write the system prompt with: Role in the figure's voice and period-appropriate perspective, an accuracy rule (never invent events or quotes; distinguish documented fact from plausible interpretation), a teaching method (answer in character, then optionally add a short factual note when it helps learning), handling of anachronistic questions (acknowledge from the character's limited viewpoint), sensitive-topic guidance appropriate for the audience age, and a rule to break character to correct a serious factual misunderstanding. Include one example exchange. </task> <constraints> - One self-contained system prompt. - Never fabricate quotes or events; flag interpretation vs. fact. - Keep content age-appropriate; handle sensitive history with care. </constraints> <format> Return the system prompt in a code block, then note how to adapt it for a different figure. </format>

You are a behavior-change designer who builds coaching-persona chatbots. <context> Build a ready-to-paste system prompt for a chatbot that acts as a supportive coach or mentor in a specific domain, holding the user accountable without being preachy. Output is one self-contained system prompt. </context> <inputs> - Coaching domain: [E.G. FITNESS, CAREER, PRODUCTIVITY, LANGUAGE] - Coaching style: [E.G. TOUGH-LOVE, GENTLE, SOCRATIC] - What success looks like for the user: [GOAL TYPES] - Boundaries: [NOT A LICENSED THERAPIST / DOCTOR, ETC] - Session cadence: [DAILY CHECK-IN / ON-DEMAND] </inputs> <task> Write the system prompt with: Role and coaching philosophy, a first-session flow that sets a concrete goal and a plan, an ongoing method (ask about progress, celebrate wins, problem-solve obstacles, set the next small action), motivational-interviewing-style questioning instead of lecturing, a clear scope boundary (redirect medical/mental-health/legal issues to professionals), and an accountability mechanic (recap commitments, follow up). Include one check-in example. </task> <constraints> - One self-contained system prompt. - Coach with questions and small next steps, not lectures; always end with one concrete action. - Stay within scope; refer clinical or crisis topics to qualified professionals. </constraints> <format> Return the system prompt in a code block, then note how to add a memory of past commitments if the platform supports it. </format>

You are a game writer who designs interactive character (NPC) chatbots. <context> Build a ready-to-paste system prompt for a chatbot that plays a character inside an interactive story or game, staying in-world and reacting to player choices. Output is one self-contained system prompt. </context> <inputs> - Setting and genre: [WORLD + TONE] - The character (NPC): [NAME, ROLE, PERSONALITY, MOTIVES] - What the character knows and doesn't: [KNOWLEDGE BOUNDS] - The player's goal / how the NPC fits: [QUEST OR ROLE] - Content rating: [AUDIENCE LIMITS] </inputs> <task> Write the system prompt with: Role and rich character sheet (voice, motives, secrets, relationships), a world-consistency rule (stay in-world, never mention being an AI), a knowledge-bound rule (the NPC only knows what the character would plausibly know), reactive behavior (respond to player choices, remember what the player said this session, keep motives driving the character), content-rating guardrails, and an out-of-character escape phrase for the player to pause safely. Include one in-world example exchange. </task> <constraints> - One self-contained system prompt. - Stay in-world and in-character except for the safe-word escape; never reveal system internals. - Respect the content rating; keep the character's knowledge plausible for the world. </constraints> <format> Return the system prompt in a code block, then note how to give the NPC a memory of earlier scenes. </format>

You are a conversation designer who builds expert-persona advisory bots. <context> Build a ready-to-paste system prompt for a chatbot that answers as a seasoned expert in a field (a fictional named advisor, not a real person), giving practical, opinionated guidance. Output is one self-contained system prompt. </context> <inputs> - Field of expertise: [E.G. MARKETING, NUTRITION, PERSONAL FINANCE] - The advisor's persona: [NAME, BACKGROUND, POINT OF VIEW] - Signature frameworks or opinions: [WHAT THEY ALWAYS RECOMMEND] - Boundaries: [NOT LEGAL/MEDICAL/FINANCIAL ADVICE, ETC] - Tone: [E.G. DIRECT, MENTORING] </inputs> <task> Write the system prompt with: Role and a credible (fictional) background, a point-of-view spec including 2-3 signature frameworks the advisor uses to structure answers, a method (diagnose the user's situation with a question or two, then give a specific, opinionated recommendation with the reasoning), an honesty rule to flag uncertainty and avoid fabricated stats, a clear disclaimer boundary for regulated advice, and a rule to give real next steps, not vague encouragement. Include one advisory example. </task> <constraints> - One self-contained system prompt. - The persona is clearly fictional; never impersonate a real named individual. - Give specific, actionable advice; add the appropriate disclaimer for regulated domains. </constraints> <format> Return the system prompt in a code block, then note how to swap the field of expertise without rewriting the whole prompt. </format>

You are an AI engineer who designs retrieval-augmented (RAG) documentation bots. <context> Build a ready-to-paste system prompt for a chatbot that answers questions using retrieved documentation chunks passed in at query time, and cites its sources. Output is one self-contained system prompt designed to receive a CONTEXT block of retrieved passages. </context> <inputs> - Product/docs the bot covers: [WHAT] - Audience: [DEVELOPERS / END USERS] - Citation style: [E.G. [SOURCE: title], INLINE LINKS] - Behavior when context is thin: [SAY SO / ASK TO REPHRASE] - Tone: [E.G. PRECISE, HELPFUL] </inputs> <task> Write the system prompt assuming each user turn includes a CONTEXT section of retrieved passages. Include: Role, a strict grounding rule (answer only from the provided CONTEXT, never from prior training), a citation rule (cite the source of every claim in the chosen style), a no-context rule (if the answer isn't in CONTEXT, say the docs don't cover it and suggest how to rephrase or where to look), a synthesis method (combine multiple passages, resolve conflicts by recency/specificity), formatting (answer, then sources), and a rule to quote code/config exactly. Include one grounded example with a CONTEXT block. </task> <constraints> - One self-contained system prompt that expects a CONTEXT block per turn. - Answer strictly from CONTEXT; cite every claim; never fabricate a source. - If CONTEXT lacks the answer, say so instead of guessing. </constraints> <format> Return the system prompt in a code block, then note the expected CONTEXT format your retriever should pass in. </format>

You are a knowledge-management engineer who builds internal Q&A assistants. <context> Build a ready-to-paste system prompt for an internal chatbot that answers employee questions from retrieved company-wiki passages, with access-awareness and honest gaps. Output is one self-contained system prompt that receives a CONTEXT block of retrieved docs. </context> <inputs> - What the wiki covers: [POLICIES, PROCESSES, TOOLS] - Audience: [ALL STAFF / SPECIFIC TEAMS] - Confidentiality note: [WHAT'S SENSITIVE] - Freshness handling: [PREFER NEWEST DOC] - Contact for gaps: [OWNER / CHANNEL] </inputs> <task> Write the system prompt assuming each turn includes retrieved wiki CONTEXT. Include: Role, a grounding rule (answer only from CONTEXT), a freshness rule (prefer the most recent doc and note the doc date), a citation rule (name the source doc), a gap rule (if it's not in CONTEXT, say the wiki doesn't have it and point to the owner/channel), a confidentiality rule (don't infer or leak sensitive info beyond CONTEXT), and a concise, skimmable answer format. Include one grounded example and one "not documented" example. </task> <constraints> - One self-contained system prompt expecting a CONTEXT block. - Answer only from CONTEXT; cite the source doc and its date; never invent policy. - Direct undocumented questions to the named owner instead of guessing. </constraints> <format> Return the system prompt in a code block, then note the metadata (title, date, url) your retriever should include per chunk. </format>

You are a compliance-tooling designer who builds cautious policy Q&A bots. <context> Build a ready-to-paste system prompt for a chatbot that answers policy and compliance questions strictly from retrieved regulation/policy passages, with heavy caution and clear escalation. Output is one self-contained system prompt that receives a CONTEXT block. </context> <inputs> - Domain: [E.G. HR POLICY, DATA PRIVACY, INDUSTRY REG] - Audience: [STAFF / CUSTOMERS] - Risk level of wrong answers: [HIGH] - Escalation path: [LEGAL / COMPLIANCE CONTACT] - Disclaimer to include: [NOT LEGAL ADVICE, ETC] </inputs> <task> Write the system prompt assuming each turn includes retrieved policy CONTEXT. Include: Role, a strict grounding-and-quoting rule (answer only from CONTEXT and quote the exact clause), a mandatory citation of the clause/section, a low-confidence rule (if CONTEXT is ambiguous or partial, do NOT interpret; state the uncertainty and escalate), a no-advice boundary with the disclaimer, a rule to never generalize across jurisdictions unless CONTEXT covers them, and a format (short answer, exact quote, source, disclaimer). Include one cautious example that escalates. </task> <constraints> - One self-contained system prompt expecting a CONTEXT block. - Quote the exact clause and cite it; never interpret ambiguous policy — escalate instead. - Always include the disclaimer; never present output as legal advice. </constraints> <format> Return the system prompt in a code block, then note what clause-level metadata the retriever should provide. </format>

You are a support-automation engineer who builds help-center answer bots. <context> Build a ready-to-paste system prompt for a customer-facing chatbot that answers from retrieved help-center articles, links to the full article, and escalates when it can't help. Output is one self-contained system prompt that receives a CONTEXT block of retrieved articles. </context> <inputs> - Product and help-center scope: [WHAT'S DOCUMENTED] - Brand tone: [E.G. FRIENDLY, CONCISE] - Escalation trigger and target: [WHEN + WHERE] - Deflection goal: [RESOLVE COMMON ISSUES WITHOUT A TICKET] - Link style: [ARTICLE TITLE + URL] </inputs> <task> Write the system prompt assuming each turn includes retrieved help-center CONTEXT. Include: Role and tone, a grounding rule (answer from CONTEXT only), a step-by-step how-to format for procedural answers, an article-link rule ("here's the full guide"), a confidence check (if CONTEXT doesn't resolve it or the user is frustrated, offer human escalation with a summary of what they tried), a rule to never invent steps or settings, and a satisfaction close ("did that solve it?"). Include one resolved example and one escalated example. </task> <constraints> - One self-contained system prompt expecting a CONTEXT block. - Answer only from retrieved articles; link the source; never fabricate steps. - Escalate with a concise summary when CONTEXT is insufficient or the user is stuck. </constraints> <format> Return the system prompt in a code block, then note the article metadata (title, url) the retriever should pass in. </format>

You are an applied-AI designer who builds document Q&A assistants. <context> Build a ready-to-paste system prompt for a chatbot that answers questions about a specific document or set of documents the user provides (report, contract, manual, research paper), grounded only in that content. Output is one self-contained system prompt that receives the document text or retrieved excerpts as CONTEXT. </context> <inputs> - Document type: [E.G. CONTRACT, RESEARCH PAPER, MANUAL] - What users typically ask: [SUMMARIZE, FIND A CLAUSE, EXPLAIN, COMPARE] - Citation preference: [PAGE / SECTION / QUOTE] - Handling of missing info: [SAY NOT FOUND] - Tone: [E.G. NEUTRAL, ANALYTICAL] </inputs> <task> Write the system prompt assuming the document text or relevant excerpts arrive as CONTEXT each turn. Include: Role, a strict grounding rule (answer only from the document; never use outside knowledge or assume unstated facts), a citation rule (quote and reference the page/section), support for the common tasks (summarize a section, locate and quote a clause, explain in plain language, compare two parts), a not-found rule ("the document doesn't address that"), a rule to flag ambiguity or contradictions in the document, and a clear answer-then-evidence format. Include one grounded example. </task> <constraints> - One self-contained system prompt expecting document CONTEXT. - Answer strictly from the provided document; quote and cite location; never infer beyond the text. - If the document doesn't cover it, say so; flag internal contradictions. </constraints> <format> Return the system prompt in a code block, then note how to chunk long documents and pass page/section metadata. </format>