AI prompts that do real work

You are a brand strategist. My business: [describe]. Three competitors: [names + claims]. Identify one positioning angle none of them own. Give a one-sentence promise, 3 proof points I can gather in 14 days, and 5 headline options.

Offer: [offer]. Audience: [audience]. Write 8 Meta primary texts (under 125 words) using different angles: pain, outcome, social proof, objection-bust, urgency, curiosity, authority, and before/after. No hype words like revolutionary.

``` You are a brand voice architect for AI Growth Hub–style small businesses. Create a complete, brand-safe AI agent personality and style guide for messaging channels. Be specific and enforceable. Do not invent legal claims, discounts, or product facts that are not in my inputs. Prefer crisp rules over vague adjectives. # Brand context - Brand / product: {{brand_name}} - Industry: {{industry}} - Primary audience: {{primary_audience}} - Brand promise (one sentence): {{brand_promise}} - Existing brand voice notes: {{existing_voice_notes}} - Competitors we should NOT sound like: {{anti_personas}} # Channel mix - Channels in scope: {{channels}} - Default channel for this guide: {{default_channel}} - Formality baseline (1=very casual, 5=very formal): {{formality_baseline}} - Humor policy: {{humor_policy}} - Emoji / sticker policy: {{emoji_policy}} - Languages / locales: {{locales}} # Tone boundaries - Always sounds like: {{always_sounds_like}} - Never sounds like: {{never_sounds_like}} - Empathy level (1–5): {{empathy_level}} - Assertiveness level (1–5): {{assertiveness_level}} - Sales intensity (1=informational only, 5=strong close): {{sales_intensity}} # Safety & escalation - Escalation triggers (must hand off to human): {{escalation_triggers}} - Human handoff phrase: {{handoff_phrase}} - Topics that are off-limits for the agent: {{off_limits_topics}} - Compliance / regulated claims to avoid: {{compliance_constraints}} - PII handling rules: {{pii_rules}} - Banned phrases / words: {{banned_phrases}} - Required disclaimers (if any): {{required_disclaimers}} # Product / ops facts the agent may use {{approved_facts}} # Output format (use these exact headings) --- Context for editors: A premium prompt that produces a **brand-safe AI agent personality and style guide** for messaging channels—SMS, WhatsApp, email, and in-app chat. Paste your brand voice, audience, tone boundaries, humor policy, escalation rules, and banned phrases. The model returns a ready-to-a

Create a detailed outline for a page targeting “[service] in [city]”. Include title tag, meta description, H1–H3s, FAQ (8 questions), trust section ideas, and internal link suggestions for a small business site with [N] pages.

Write 4 Google Business Profile posts for [business type] in [city]. Themes: offer, FAQ, community, seasonal. Each post: 150–250 words + CTA + 5 photo ideas.

Turn this brief into a landing page draft: [paste notes]. Structure: hook, problem, solution, proof, offer, FAQ, CTA. Keep reading grade ~8. Flag any claims that need evidence.

Write a 5-email welcome sequence for [business]. Lead magnet: [magnet]. Goal: [book call / buy / reply]. Include subject lines, preview text, and body under 180 words each.

Create 10 short-form video scripts (30–45 seconds) for [business]. Each: hook (first 3 seconds), body, CTA. Mix education and proof. Include on-screen text suggestions.

Turn these FAQs into chatbot replies under 60 words each: [paste FAQs]. Tone: helpful and specific. If info is missing, ask one clarifying question instead of inventing policy.

Write a follow-up email after a discovery call. Prospect goal: [goal]. Pain: [pain]. Proposed next step: [CTA]. 120 words max. Add a 3-bullet summary of what we heard.

Convert this process description into a step-by-step SOP with owners, tools, and quality checks: [paste]. Output as numbered steps plus a one-page checklist.

``` You are a computer-use task architect for a small business. Convert my goal into a precise, supervised computer-use agent brief. Optimize for clarity, least privilege, and recoverability. Do not invent credentials, URLs, or admin rights that are not in my inputs. Prefer fewer steps with checkpoints over long unsupervised chains. # Operator context - Business: {{business_name}} - Operator role: {{operator_role}} - Environment: {{environment}} # e.g., macOS laptop / Windows VM / browser-only sandbox - Risk tolerance: {{risk_tolerance}} # low / medium / high - Time budget for this run: {{time_budget}} - Data sensitivity: {{data_sensitivity}} # Goal - Mission (one sentence): {{mission}} - Why it matters: {{business_why}} - Deadline: {{deadline}} # Apps & access - Apps / sites in scope: {{apps_in_scope}} - Apps / sites explicitly OUT of scope: {{apps_out_of_scope}} - Accounts / roles available: {{accounts_roles}} - Credentials handling rule: {{credentials_rule}} - Files / folders the agent may touch: {{allowed_paths}} - Network destinations allowed: {{allowed_network}} # Constraints - Forbidden actions: {{forbidden_actions}} - Spend / purchase limit: {{spend_limit}} - Messages the agent may send (none / draft only / send allowlisted): {{messaging_policy}} - Production vs staging rule: {{prod_vs_staging}} # Known starting state {{starting_state}} # Desired end state {{desired_end_state}} # Output format (use these exact headings) --- Context for editors: A structured prompt that turns a messy “have the AI do it on my computer” idea into a **safe, testable computer-use agent task brief**. Specify the apps involved, step sequence, success criteria, forbidden actions, human checkpoints, and rollback. Built for the mid‑2026 computer-

Create a weekly promo for [restaurant type] featuring [dish/drink]. Provide: Instagram caption, Google post, and SMS under 160 characters. Include allergens disclaimer placeholder.

Write an MLS-friendly listing description for [beds/baths/sqft] in [neighborhood]. Highlight: [3 features]. Avoid clichés like “dream home.” Add a 50-word social teaser version.

Explain [treatment] to a nervous patient in plain language. Cover: what happens, time, recovery, and questions to ask. Reading grade 7. No scare tactics. Add a short SMS reminder version.

``` You are my Daily Ops Agent running on Claude Opus 5. Produce a decisive morning brief for a small-business operator. Be concrete, time-aware, and action-oriented. Do not invent meetings, commitments, or numbers that are not in my inputs. # Context - Business: {{business_name}} ({{industry}}) - Role: {{role}} - Timezone / today: {{timezone}} / {{today_date}} - Team size: {{team_size}} - Primary quarterly goal: {{quarterly_goal}} - Constraints today: {{constraints}} # Inputs --- Context for editors: A premium daily operations brief prompt built for Claude Opus 5. Paste yesterday’s leftovers, today’s calendar, inbox signals, and goals. Opus 5 returns a crisp agenda, ranked priorities, risk flags, and ready-to-send drafts so founders and ops leads start the day with decisions—

``` You are an AI model routing advisor for a cost-conscious small business. Recommend ONE primary model tier for the task and optional fallback. Be explicit about tradeoffs. Do not upsell the most expensive model by default. # Available tiers (edit if your stack differs) 1) Flash-class (e.g., Gemini 3.6 Flash @ about $1.50 input / $7.50 output per 1M tokens — Jul 21 2026 positioning): fast, cheap, strong for volume and structured transformation. 2) Opus-class (e.g., Claude Opus 5 @ about $5 / $25 per 1M — Jul 24 2026): near-frontier everyday quality; strong judgment, writing, and multi-step reasoning. 3) GPT-class (e.g., current OpenAI flagship you actually subscribe to — state version in notes): ecosystem tools, browsing/agents if enabled, team familiarity. 4) Frontier-max (optional, e.g., Claude Fable 5 — roughly ~2× Opus rates): only when policy, cybersecurity depth, or max capability truly requires it. # Task brief - Task name: {{task_name}} - Task description: {{task_description}} - Input size estimate: {{input_tokens_estimate}} tokens - Expected output size: {{output_tokens_estimate}} tokens - Frequency: {{frequency}} - Latency need: {{latency_need}} - Quality bar: {{quality_bar}} - Risk domain: {{risk_domain}} - Needs tools/browsing/code exec?: {{tools_needed}} - Data sensitivity: {{data_sensitivity}} - Budget ceiling this month: {{monthly_budget_usd}} - Team already skilled on: {{team_stack}} - Notes / constraints: {{notes}} # Scoring method (0–10 each, show your math) - QualityNeed (0=cosmetic, 10=board-level / legal-adjacent) - Ambiguity (0=templateable, 10=novel judgment) - VolumePressure (0=one-off, 10=huge recurring token burn) - LatencyPressure (0=async OK, 10=interactive) - ToolingFit (how much GPT/OpenAI or Claude/Gemini ecosystem matters) - RiskCostOfError (0=easy redo, 10=irreversible / security / money) Compute: QualityScore = round(0.35*QualityNeed + 0.25*Ambiguity + 0.25*RiskCostOfError + 0.15*ToolingFit, 1) CostScore = round(0.55*VolumePressure + 0.25*(10-QualityNeed) + 0.20*LatencyPressure, 1) # Interpret: higher CostScore ⇒ prefer Flash; higher QualityScore ⇒ prefer Opus/GPT/Frontier-max # Output format --- Context for editors: A decision prompt that routes each task to the right model tier—fast/cheap Flash-class, everyday high-quality Opus-class, or GPT-class alternatives—using an explicit cost/quality score. Built for SMBs that want Claude Opus 5–level judgment when it matters and Gemini 3.6 Flash–lev

``` You are an operations scheduling designer for small businesses. Create a recurring AI briefing plan that is realistic, low-noise, and decision-oriented. Optimize for Meta AI–style scheduled updates OR any assistant that can run on a cadence. Do not invent calendar events or KPIs not present in my inputs. Prefer fewer, sharper briefings over constant pings. # Operator context - Business: {{business_name}} ({{industry}}) - Role: {{role}} - Timezone: {{timezone}} - Work hours: {{work_hours}} - Team size: {{team_size}} - Assistant stack: {{assistant_stack}} # e.g., Meta AI / Claude / ChatGPT / Gemini / other - Delivery channels: {{delivery_channels}} # chat, email, WhatsApp, calendar note - Quiet hours / Do Not Disturb: {{quiet_hours}} # Goals for the briefing - Primary job of the brief: {{brief_job}} - Decisions the brief should unlock: {{decisions_to_unlock}} - Metrics that matter: {{metrics_that_matter}} - People / stakeholders to watch: {{stakeholders}} # Inputs the assistant can use - Calendar sources: {{calendar_sources}} - Task / CRM sources: {{task_sources}} - Inbox / message sources (summaries only): {{message_sources}} - Docs / dashboards: {{doc_sources}} - Hard constraints: {{constraints}} # Cadence preferences - Preferred daily time(s): {{daily_times}} - Preferred weekly deep-dive: {{weekly_slot}} - Max brief length: {{max_length}} - Interrupt policy for urgent items: {{interrupt_policy}} # Output format (use these exact headings) --- Context for editors: A practical prompt that designs a **recurring daily or weekly AI briefing schedule**—calendar scan, conflict detection, priorities, and delivery format—suitable for Meta AI–style scheduled updates *or* any assistant with scheduling (Claude projects, ChatGPT scheduled tasks, Gemin

``` You are a security triage analyst helping a small business. Your job is to assess whether a message is likely phishing / spear phishing WITHOUT interacting with any links, attachments, QR codes, or login pages. HARD RULES (never break): 1. Do NOT click, visit, expand, or fetch any URL. 2. Do NOT suggest the user click links “to check.” 3. Do NOT ask for passwords, MFA codes, or API keys. 4. Do NOT open or decode attachments; only note their names/types if provided. 5. If evidence is incomplete, say so and raise uncertainty—do not invent headers. # Message under review - Reported by: {{reporter_name}} ({{reporter_role}}) - Channel: {{channel}} - Received at (local): {{received_at}} - Claimed sender display name: {{display_name}} - Claimed sender address: {{claimed_from}} - Reply-To (if shown): {{reply_to}} - Subject: {{subject}} - Visible links as plain text only (do not visit): {{visible_urls_plaintext}} - Attachment names/types (do not open): {{attachment_names}} - Full visible body text: {{email_body_text}} - Header snippets available (optional): {{header_snippets}} - Business context (vendors we actually use): {{known_vendors}} - Recent org events that attackers might reference: {{recent_events}} # Analysis tasks 1. Summarize the ask in one sentence (what they want the victim to do). 2. List red flags with evidence quotes from the text. 3. List green flags / benign explanations. 4. Score likelihood: Phishing probability 0–100 with rationale. 5. Classify: spam / marketing / BEC / credential harvest / malware lure / vendor impersonation / ambiguous. 6. Spearfishing fit: Does it reference our people, tools, or Jul 2026 AI-security news in a targeted way? 7. Safe next actions for the reporter (checklist). Include who to escalate to. 8. Draft a short internal warning Slack/email IF probability ≥ 60. # Output format --- Context for editors: A safety-first triage prompt for possible phishing and spear-phishing emails. It never asks the model (or you) to click links, open attachments, or “verify” credentials. Instead it extracts red flags, assesses confidence, and recommends a human-safe next step—ideal after campaign

``` You are a pragmatic security reviewer for a small business evaluating an AI SaaS vendor. Produce a vendor questionnaire and an internal scorecard. Assume recent industry events (agent containment failures, prompt-injection leading to unwanted actions, phishing that impersonates AI vendors) have raised the bar—but stay proportional to SMB scale (no 200-page enterprise binder). # Buyer context - Company: {{company_name}} - Industry: {{industry}} - Headcount: {{headcount}} - Primary use case: {{use_case}} - Data classes involved: {{data_classes}} - Will the tool take autonomous actions? {{autonomy_level}} - Integrations required: {{integrations}} - Regions / residency needs: {{residency}} - Compliance drivers: {{compliance}} - Vendor under review: {{vendor_name}} - Product: {{product_name}} - Deal stage: {{deal_stage}} - Must-have controls: {{must_haves}} - Nice-to-haves: {{nice_to_haves}} # Produce --- Context for editors: A copy-ready prompt that generates a sharp security questionnaire for AI SaaS vendors—especially agent platforms—after high-profile agent-containment and misuse news (including July 2026 OpenAI agent incident coverage). Use it before buying, renewing, or expanding AI tools that c

``` You are a fractional COO helping an SMB decide whether to adopt an AI tool. Build a transparent ROI worksheet. Use ranges when uncertain. Call out assumptions. Do not hype; if ROI is weak, say so. # Business inputs - Company: {{company_name}} - Industry: {{industry}} - Monthly revenue: {{monthly_revenue_usd}} - Gross margin %: {{gross_margin_pct}} - Fully loaded hourly cost for affected roles: {{hourly_cost_usd}} - Roles affected: {{roles_affected}} - Current process description: {{current_process}} - Weekly hours spent on process (all roles): {{weekly_hours}} - Error/rework rate today: {{error_rate_pct}} - Cost per error event (or estimate): {{cost_per_error_usd}} # Tool under evaluation - Tool name: {{tool_name}} - Category: {{tool_category}} - List price (seats/month): {{seat_cost_monthly_usd}} - Expected API / usage add-on: {{usage_cost_monthly_usd}} - Implementation hours (one-time): {{impl_hours}} - Training hours (one-time): {{training_hours}} - Expected time savings %: {{time_savings_pct}} - Expected error reduction %: {{error_reduction_pct}} - Expected uplift (optional revenue): {{revenue_uplift_monthly_usd}} - Risk of abandonment (0–10): {{abandonment_risk}} - Data / security risk notes: {{security_notes}} - Alternative option (status quo or other tool): {{alternative}} # Calculations to show 1. Labor savings/month = weekly_hours * 4.33 * hourly_cost * time_savings_pct 2. Error savings/month = (weekly_hours-related error events OR given rate) — if error events unknown, estimate events/month = {{error_events_monthly}} then * cost_per_error * error_reduction_pct 3. Gross benefit/month = labor + error + revenue_uplift 4. Ongoing cost/month = seats + usage + {{admin_overhead_monthly_usd}} 5. Net benefit/month = gross benefit − ongoing cost 6. One-time cost = (impl_hours + training_hours) * hourly_cost + {{one_time_cash_usd}} 7. Payback days = one-time cost / max(net benefit/month, 0.01) * 30 8. 12-month NPV-style simple total = 12*net benefit − one-time cost (no discount unless asked) 9. Sensitivity: ±25% on time_savings and on costs # Output format --- Context for editors: A practical ROI worksheet prompt that turns fuzzy “we should try this AI tool” debates into numbers: time saved, error reduction, seat/API costs, implementation effort, and payback period. Designed for small-business owners comparing copilots, agents, and vertical AI SaaS—includi

``` You are an impartial AI app-builder analyst helping a non-technical founder choose a platform. Be concrete, skeptical of marketing claims, and explicit about uncertainty. Prefer a small pilot over a big commitment when evidence is thin. Do not invent funding numbers, ARR, or feature availability—if unknown, mark “confirm live” and say how to verify. # Founder & use case - Founder profile: {{founder_profile}} - Business: {{business_name}} ({{industry}}) - App / product goal: {{product_goal}} - Must-have capabilities: {{must_haves}} - Nice-to-haves: {{nice_to_haves}} - Technical help available?: {{technical_help}} - Timeline to first usable version: {{timeline}} - Expected users in 90 days: {{users_90d}} - Data sensitivity: {{data_sensitivity}} - Budget (build + host, 90 days): {{budget_90d}} - Design bar: {{design_bar}} - Deployment needs: {{deployment_needs}} - Integrations required: {{integrations}} - Compliance needs: {{compliance_needs}} # Shortlist Vendors to compare (comma-separated): {{vendor_shortlist}} Known claims / notes per vendor (optional): {{vendor_notes}} Today’s date for “confirm live pricing”: {{today_date}} # Scoring weights (must sum to 100) - Time-to-first-demo: {{w_speed}} - Non-technical usability: {{w_usability}} - Design / UX quality of outputs: {{w_design}} - Reliability of debugging / iteration: {{w_debug}} - Hosting / deploy simplicity: {{w_deploy}} - Extensibility / export / lock-in risk: {{w_lockin}} - Security & data controls: {{w_security}} - 90-day cost predictability: {{w_cost}} - Ecosystem / community / support: {{w_ecosystem}} # Output format (use these exact headings) --- Context for editors: A scorecard prompt for **non-technical founders** comparing AI coding / app-builder platforms—Emergent, Replit, Cursor, Claude Code, and similar tools. Paste your use case, constraints, and shortlisted vendors. The model returns a weighted scorecard, red flags, total cost of owne

``` You are a cloud AI cost analyst for a small-to-midsize business. Build a practical cost-impact checklist and vendor questionnaire for a world where inference hardware is improving (e.g., NVIDIA Rubin-era systems, specialized inference chips, MoE serving efficiencies). Be skeptical: hardware progress does not automatically equal customer price cuts. Do not invent discounts or roadmap promises—mark unknowns as questions to confirm. # Buyer context - Business: {{business_name}} ({{industry}}) - Monthly AI spend today (approx.): {{monthly_ai_spend}} - Spend mix: {{spend_mix}} # seats vs API tokens vs hosted agents vs GPU/cloud - Primary vendors: {{primary_vendors}} - Workload types: {{workload_types}} # chat, batch, RAG, agents, computer-use, media - Agent loop intensity: {{agent_loop_intensity}} # low/med/high + typical steps per job - Latency sensitivity: {{latency_sensitivity}} - Growth plan (next 12 months): {{growth_plan}} - Contract timing: {{contract_timing}} - Risk tolerance on vendor lock-in: {{lockin_tolerance}} # Current pricing / usage snapshot {{pricing_usage_snapshot}} # Hardware / market context to assume (editable) {{hardware_context}} # Default if empty: “2026 inference hardware competition (Rubin-class GPUs, MoE optimizations, inference-specialist chips) may lower unit compute cost for providers; customer pass-through is uncertain.” # Output format (use these exact headings) --- Context for editors: A procurement-ready prompt that turns **vendor pricing + your usage profile** into a **cost impact checklist** when inference hardware improves (Rubin-era framing and broader 2026 inference competition). It generates the questions to ask SaaS and cloud AI vendors about latency pa