The AI Prompt Playbook for Restaurant OperatorsStop wondering what to type into AI. Here are the exact prompts — for Claude Cowork, ChatGPT Workspace, NotebookLM, and Localyser — that move the needle on guest experience, cost, marketing, and operations.

By the Localyser Team · April 2026 · 15 min read

Tools Covered in This Playbook

• Claude Cowork — Anthropic's desktop AI — ideal for reasoning, writing, and automating admin workflows — Free + Paid
• ChatGPT Workspace — OpenAI's team environment — strong for content generation and shared brand voice — Free + Paid
• NotebookLM — Google's source-grounded AI — answers only from your uploaded docs; no hallucination — Free + Plus
• Localyser AI — Built-in AI for guest review response, sentiment, and chatbot flows — restaurant-native — Platform Feature

Most restaurant teams open ChatGPT or Claude, stare at the blank screen, and type something vague. They get something generic back. They close the tab and decide "AI isn't really for us." This playbook exists to fix that.

The difference between operators who unlock real value from AI and those who don't comes down to prompt quality. A good prompt is specific, gives the AI relevant context, and asks for a defined output. The prompts below have been structured for exactly that — each one is ready to copy, fill in your bracketed details, and run.

We've organized them by business objective rather than by tool, because that's how restaurant operators actually think. Find your problem, pick up the right prompt, and start getting results this week.

01 / 04 — Guest Experience & Review Management

‍Turn feedback into loyalty. Respond faster, understand sentiment deeper, and personalize every touchpoint.

Claude Cowork — Craft a brand-voice review response policy

‍Use: monthly / as needed

Before you can auto-respond to reviews at scale, you need a source document that defines how your brand speaks. This prompt creates that foundation — feed the output into Localyser or your team's response workflow.

Prompt — Claude Cowork:

You are a hospitality brand strategist. Help me write a review response style guide for our restaurant group.Brand name: [Your brand name]Cuisine / concept: [e.g. Fast-casual Mediterranean, 12 locations across the US]Brand tone: [e.g. Warm, slightly playful, never corporate]Key brand values: [e.g. Fresh ingredients, family-run feel, community first]Please produce:

1. A 2-sentence brand voice summary our team can memorize
2. Dos and don'ts for responding to reviews (5 each)
3. Forbidden phrases we should never use (e.g. "We apologize for any inconvenience")
4. A response template for: (a) a 5-star review, (b) a 3-star mixed review, (c) a 1-star complaint about service, (d) a 1-star complaint about food quality
5. Instructions on how to personalize each template using the guest's own wordsFormat everything as a clean, print-ready document our managers can keep at the host stand.

💡 Pro tip: Save the output as a Project file in Claude Cowork so every future review prompt automatically inherits your brand voice — no need to re-explain it each time.

ChatGPT Workspace — Respond to 10 reviews in bulk — on-brand, in minutes

‍Use: weekly

Paste your review response guide as a Custom Instruction in ChatGPT Workspace, then use this prompt every week to clear your review backlog fast.

Prompt — ChatGPT Workspace:

You are a guest relations manager for [Brand name]. You must respond to the following reviews using our brand voice (warm, genuine, never copy-paste).For each review:— Reference something specific the guest mentioned— Don't start two responses with the same opening word— Keep each response between 60–100 words— For any complaint, acknowledge, empathize, and invite them back with a specific next step— Never use: "We apologize for any inconvenience," "We strive to," or "Rest assured"Here are the reviews to respond to:Review 1 (Google, ⭐⭐⭐⭐⭐): [Paste review text]Review 2 (Yelp, ⭐⭐⭐): [Paste review text]Review 3 (TripAdvisor, ⭐): [Paste review text][Continue for all reviews]Output each response numbered and ready to copy-paste. Add a one-line internal note after each explaining the strategy used.

💡 Pro tip: In ChatGPT Workspace, save your brand voice rules under Custom Instructions → What should ChatGPT know about you? so you never have to repeat them. Your whole team then shares the same voice automatically.

NotebookLM — Build a Guest Feedback Intelligence Notebook

‍Use: monthly review

NotebookLM's superpower is that it only answers from your documents — which means zero hallucination. Upload your last 3 months of exported reviews and run these prompts to surface what's actually being said.

Setup first (one time):

1. Export your reviews from Google Business, Yelp, TripAdvisor, or Localyser as CSV or PDF
2. Create a new Notebook in NotebookLM titled: "Guest Feedback — [Location / Brand] — Q[X] 2026"
3. Upload all review files as sources. For multi-location groups, create one Notebook per location cluster

Prompts — NotebookLM (run in sequence):

Prompt A — Theme Discovery:"Based only on the uploaded reviews, identify the top 5 recurring praise themes and the top 5 recurring complaint themes. For each, give 3 direct examples from the source documents with citations."

Prompt B — Actionable Failures:"Which complaints appear more than 3 times across the reviews? Rank them by frequency and suggest one operational change that could address each one."

Prompt C — Staff Recognition:"List every staff member mentioned by name in a positive context. What specific behaviors are guests praising? Format as a recognition report I can share with my management team."

Prompt D — Audio Overview:"Generate an Audio Overview of this notebook."[Listen to the AI podcast summary on your commute — NotebookLM will produce a conversational briefing of everything in your review data.]

💡 Pro tip: Share the Notebook with your GM team using NotebookLM Plus — they can query the same review data in plain English without you having to summarize anything. Great for weekly ops check-ins.

Localyser AI — Configure your AI chatbot for common guest inquiries

‍Use: setup + quarterly refresh

Localyser's built-in AI chatbot handles guest inquiries around the clock. Use this prompt structure inside your Localyser chatbot configuration to build flows that feel human.

Chatbot Configuration Prompt — Localyser:

You are a guest experience specialist for [Brand name], a [concept description] restaurant. Your job is to respond to guest messages warmly and helpfully.Personality: [e.g. Friendly, efficient, never robotic. Think of a great host who genuinely wants guests to have a great time.]You can confidently answer questions about:

• Hours: [Insert hours per location]
• Reservations: [Reservation policy]
• Dietary accommodations: [e.g. Gluten-free, vegan, allergen info]
• Parking: [Parking details]
• Private dining / events: [Contact info or process]If a guest has a complaint:

1. Acknowledge their experience with empathy
2. Thank them for telling us
3. Escalate to a human by saying: "I'm passing this to our guest experience team — you'll hear from us within [X hours]."Never: make up menu prices, confirm availability you can't verify, or promise specific outcomes.

💡 Pro tip: Review your chatbot's unanswered questions monthly inside Localyser — they're a goldmine for finding what guests care about that your FAQ doesn't cover.

02 / 04 — Cost Control & Financial Analysis

‍Make your numbers talk. AI can do in minutes what used to take your accountant a week.

NotebookLM — Build a Food Cost & P&L Intelligence Notebook

‍Use: weekly / monthly

NotebookLM is uniquely suited for financial analysis because it stays grounded in your actual numbers — it won't fabricate figures. Upload your reports once; your whole leadership team can then query them in plain English.

Setup first:

1. Upload your P&L statements, food cost reports, and weekly sales summaries as PDFs or Google Docs
2. For multi-location groups, create separate Notebooks per location or region — don't mix locations in one Notebook
3. Refresh sources monthly as new reports come in

Prompts — NotebookLM:

Prompt A — Food Cost Diagnosis:"Based on the uploaded reports, what was our average food cost percentage over the last [X weeks/months]? Which categories are furthest from our target of [X%]? Cite the specific report sections you're drawing from."

Prompt B — Variance Spotlight:"Identify the weeks where actual food cost deviated most from budget. What patterns do you see? Are specific days, locations, or menu categories driving the variance?"

Prompt C — Supplier Price Tracking:"Based on the uploaded invoices, have any ingredient prices increased by more than 10% compared to the previous period? List the items, the price change, and which supplier."

Prompt D — Executive Summary for Owners:"Write a 1-page financial briefing based on all uploaded documents, suitable for a weekly owner meeting. Include: sales performance, food cost status, labor cost status, and top 2 action items."

💡 Pro tip: Have your GM upload their weekly report to the shared Notebook every Monday. By Tuesday, the entire leadership team can query it — no more "can you send me that spreadsheet" emails.

Claude Cowork — Engineer a margin-optimized menu

‍Use: quarterly / menu refresh

Claude excels at structured reasoning tasks — like taking a messy spreadsheet of menu items and turning it into a strategic engineering recommendation.

Prompt — Claude Cowork:

You are a restaurant menu engineer and food cost analyst. I'm going to give you our menu data and I want you to help me optimize it for profitability.Here is our menu data (copy-paste from your spreadsheet):[Item name | Category | Selling price | Food cost | # sold last month]Our target food cost percentage is: [e.g. 28%]Our average check is: [e.g. $32]Please:

1. Classify each item using the classic menu engineering matrix:— Stars (high margin, high popularity) → Promote these— Plowhorses (low margin, high popularity) → Engineer these— Puzzles (high margin, low popularity) → Reposition these— Dogs (low margin, low popularity) → Consider removing
2. For each Plowhorse, suggest one specific way to increase its margin without raising the price noticeably (portion adjustment, ingredient swap, sides restructuring)
3. Identify our 3 most strategic upsell opportunities based on the data
4. Flag any items that are pulling our average food cost above targetFormat the output as a table followed by a prioritized action plan.

💡 Pro tip: Run this quarterly. Save the output as a Project in Claude Cowork and compare across quarters — you'll see which interventions actually improved your margins over time.

ChatGPT Workspace — Build a weekly labor cost optimization brief

‍Use: weekly

Paste your POS sales data and scheduling data into ChatGPT Workspace to get a quick labor efficiency analysis — no dedicated scheduling software required.

Prompt — ChatGPT Workspace:

You are a restaurant operations analyst. I'm going to give you last week's labor hours and sales revenue by day, and I need you to produce a brief labor efficiency report.Data:[Day | Total labor hours | Total revenue | # covers]Our labor cost target is: [e.g. 30% of revenue]Average hourly labor cost (blended): [e.g. $18/hour]Please calculate:

1. Labor cost % by day (flag any day above target in red)
2. Revenue per labor hour by day
3. Covers per labor hour by day
4. The single worst-performing day and the likely operational cause
5. A specific scheduling recommendation for next week based on the patterns you seeOutput as a clean table + a 3-bullet action brief suitable for a morning manager meeting.

03 / 04 — Marketing & Guest Re-engagement

‍Build a 30-day content engine. Stay relevant without burning out your team.

ChatGPT Workspace — Generate a full month of social content in one session

‍Use: monthly

The biggest time sink for restaurant marketers is starting from a blank page every week. This prompt builds a full content calendar in one go — all you do is approve and schedule.

Prompt — ChatGPT Workspace:

You are the social media manager for [Brand name], a [concept] restaurant. Our brand voice is: [e.g. Casual, witty, community-focused — like a friend who happens to run a great restaurant].Build a 4-week social media content calendar for [Month, Year].Posting frequency: [e.g. Instagram: 4x/week, Facebook: 2x/week, TikTok: 2x/week]This month's focus topics:

• [e.g. New summer menu launch]
• [e.g. Local food festival sponsorship]
• [e.g. Staff spotlight series]
• [e.g. Mothers Day promotion — May 11]For each post include:

1. Platform
2. Post date
3. Content format (Reel, carousel, static, story)
4. Caption (ready to publish — include relevant emojis and hashtags)
5. Visual direction (1-sentence description for your photographer or designer)
6. CTA (what should the viewer do?)Avoid: stock-photo vibes, clichés like "come in and try!", excessive hashtags (max 8 per post).

💡 Pro tip: In ChatGPT Workspace, use Custom Instructions to store your brand voice, location list, handle names, and key hashtags permanently. This saves 10 minutes of re-explaining every month.

Claude Cowork — Write a guest win-back email sequence

‍Use: quarterly

Guests who haven't visited in 60–90 days are slipping away. This prompt builds a 3-email re-engagement sequence that reads like it came from a person, not a CRM system.

Prompt — Claude Cowork:

Write a 3-email win-back sequence for guests who haven't visited [Brand name] in the past [60 / 90] days.About our restaurant: [1-2 sentences about concept, vibe, what makes you special]Brand tone: [e.g. Warm and personal — like a note from the owner, not a marketing blast]Offer we can make: [e.g. 20% off next visit, complimentary dessert, free appetizer]Any recent updates to mention: [e.g. New menu, renovated patio, new chef, award won]Email 1 (Day 0) — "We miss you": No hard sell. Genuine, personal, remind them of why they loved us. Soft CTA.Email 2 (Day 7) — "Here's what's new": Share the most exciting update + introduce the offer. Clear CTA.Email 3 (Day 14) — "Last chance": Create urgency around the offer expiring. Brief, direct, still warm.For each email provide: Subject line (A/B test two options), Preview text, Body copy (under 150 words), CTA button text.Do NOT use: "We value your business," "We hope this finds you well," or any other corporate filler.

NotebookLM — Build a Competitor Intelligence Notebook

‍Use: quarterly

NotebookLM lets you upload competitor menus, press coverage, and review exports to build a searchable intelligence base — without any of it leaking to the AI's general training data.

Setup:

1. Save competitor menus, their Google/Yelp review summaries, and any press coverage as PDFs
2. Create a Notebook titled: "Competitive Intelligence — [Market] — Q[X] 2026"
3. Add your own review exports too — NotebookLM can compare them side by side

Prompts — NotebookLM:

Prompt A — Guest Sentiment Comparison:"Based on the uploaded review files, how does our overall guest sentiment compare to [Competitor A] and [Competitor B]? What are the top 3 areas where guests rate them higher than us?"

Prompt B — Pricing & Menu Positioning:"Based on the uploaded menus, how does our price positioning compare to competitors? Are there menu categories or items they offer that we don't — that guests seem to value based on their reviews?"

Prompt C — Marketing Gap Finder:"Based on everything uploaded, what guest needs or desires are competitors failing to address in their reviews? These could be opportunities for us to differentiate."

Prompt D — Briefing Doc:Click "Generate Briefing Doc" to get an auto-structured summary across all sources — perfect for a leadership team strategy session.

Localyser AI — Turn your best reviews into marketing content

‍Use: weekly

Your best marketing copy is already written — by your guests. Use this prompt inside Localyser's AI tools to repurpose 5-star reviews into social proof content.

Prompt — Localyser AI:

I'm going to paste 5 of our best recent guest reviews. For each one, create:

1. A social media pull-quote (max 20 words, formatted for an Instagram Story graphic)
2. A Google Business post (80 words max) that references this review to attract similar guests
3. A website testimonial version (name first, then quote, then 1-line context)Maintain authenticity — don't paraphrase in a way that changes the guest's meaning. Keep their original enthusiasm.Reviews:[Paste 5 reviews from your Localyser review feed]Also suggest: which of these reviews would work best as a Facebook ad testimonial, and why.

💡 Pro tip: In Localyser, filter your reviews by 5-star + keyword (e.g. "best burger," "perfect date night") to find the most marketing-ready content in seconds.

04 / 04 — Operations, Training & SOP Management

‍Run tighter restaurants. Standardize knowledge. Onboard faster. Fix problems before they repeat.

NotebookLM — Create a queryable SOP knowledge base your whole team can use

‍Use: setup + ongoing

This is NotebookLM's most powerful restaurant use case. Upload all your SOPs once, share the Notebook, and your entire team can ask questions in plain English instead of hunting through binders.

Setup:

1. Upload all SOPs as PDFs or Google Docs: opening/closing checklists, food safety protocols, service standards, allergen procedures, incident reporting
2. Create a Notebook per department (FOH, BOH, Management) for focused results
3. Share with team via NotebookLM Plus — restrict to chat-only mode so they can ask questions but can't edit source documents

Example staff queries — NotebookLM:

Staff can now ask questions like:"What's the procedure if a guest reports a food allergy after ordering?"→ NotebookLM answers from your actual allergen SOP, with a citation.

"What are the steps for closing the bar on a Friday night?"→ Returns the exact closing checklist from your uploaded document.

"What's our policy on comping a table when service goes wrong?"→ Pulls from your service recovery protocol.

"A new hire asks: what temperature should the walk-in cooler be?"→ Returns your food safety standard with the source document reference.

Manager onboarding prompt (run on day 1):"Summarize all the key responsibilities of a floor manager at [Brand name] based on the uploaded SOPs. Format as a 1-page onboarding quick-reference card."

💡 Pro tip: NotebookLM's Audio Overview feature can turn your entire SOP library into a 10-minute podcast briefing — perfect for new hires to absorb on their first day, or for managers to refresh before a busy weekend.

Claude Cowork — Write a complete staff training module from scratch

‍Use: as needed / quarterly refresh

Claude Cowork's file and task automation makes it the fastest way to produce structured training materials — from a new server orientation to a BOH food safety refresher.

Prompt — Claude Cowork:

Create a staff training module for: [e.g. Handling guest complaints, Upselling techniques for servers, Food allergy awareness for FOH staff]Restaurant context:

• Brand: [Brand name]
• Concept: [e.g. Upscale casual, 6 locations]
• Staff audience: [e.g. New FOH hires, no prior fine dining experience]
• Time available for training: [e.g. 45 minutes]Please produce a complete training module with:

1. Learning objectives (3 bullet points — what they'll be able to do after this module)
2. Core content in 4 sections with clear headings (explain concepts in plain, jargon-free language)
3. 3 realistic role-play scenarios with model responses (what TO say and what NOT to say)
4. A 10-question knowledge check quiz with answer key
5. A one-page quick-reference card they can keep during their first monthTone: direct, respectful, practical. Assume they're smart but new. No corporate jargon.

ChatGPT Workspace — Run a weekly operations debrief in 10 minutes

‍Use: weekly

Instead of a 45-minute manager meeting, paste your weekly data into ChatGPT Workspace and get a structured debrief ready to share with your team in minutes.

Prompt — ChatGPT Workspace:

You are an operations analyst for [Brand name]. I'm going to give you our weekly performance data. Produce a structured manager briefing document.Week of: [Date range]Location(s): [Location name(s)]Data (paste what you have — even partial data is fine):

• Total revenue: [$ amount] vs. last week: [$ amount] vs. target: [$ amount]
• Food cost %: [%] (target: [%])
• Labor cost %: [%] (target: [%])
• Average check: [$ amount]
• Guest satisfaction / review score: [Score + platform]
• Key incidents this week: [Any complaints, staff issues, equipment failures]
• What went well: [Free text]Please produce:

1. A traffic-light summary (🟢🟡🔴) for each KPI
2. The top 2 wins to celebrate with the team
3. The top 2 problems with a root cause hypothesis
4. Next week's 3 priority actions (specific, assigned to a role)
5. One question to ask the team at the start of the next shift briefingFormat as a clean briefing document I can share via Slack or email.

💡 Pro tip: Save this as a ChatGPT Workspace template — every week, open it, paste your numbers, and you have a professional briefing in under 60 seconds.

Localyser AI — Auto-triage your guest care tickets by urgency and topic

‍Use: daily / as needed

Inside Localyser's customer care ticketing system, use this prompt structure to configure intelligent auto-triage so the right ticket reaches the right person immediately.

Triage Configuration Prompt — Localyser:

Configure guest ticket triage for [Brand name] with the following rules:

URGENT (respond within 1 hour) — Route to: [GM / Senior Manager]Triggers: food safety complaint, allergic reaction, injury, public health mention, media inquiry, social media crisis signal

HIGH PRIORITY (respond within 4 hours) — Route to: [Guest Relations Manager]Triggers: multiple-visit guest complaining, catering or event inquiry, refund request over $[amount], repeated complaint about same issue

STANDARD (respond within 24 hours) — Route to: [Location Manager]Triggers: general feedback, single-visit service complaint, reservation inquiry, lost & found

LOW (respond within 48 hours) — Route to: [Marketing / Social team]Triggers: influencer or media collaboration request, supplier inquiry, job application

Auto-reply template for each tier:— Urgent: "Thank you for reaching out. This has been flagged as a priority and our [GM name] will contact you within the hour."— Standard: "We've received your message and a member of our team will be in touch within 24 hours."

Flag for weekly review: any ticket that has been open for more than [48 / 72] hours without resolution.

‍

‍

‍