Imagine waking up to a flood of new reviews — some glowing, some scathing, but all demanding a timely response. For one coffee chain, the old way meant a dedicated team spending up to 48 hours to craft replies. Today, they do it in 10 minutes flat. The secret? Not a magic wand, but a carefully engineered set of AI prompts for multimodal analysis. In this case study, you'll see exactly how they turned a chaotic feedback loop into a streamlined, brand-boosting machine — and how you can replicate it.
The Problem: When Customer Feedback Becomes a Bottleneck
With over 120 locations, the chain was drowning in reviews across platforms like Yelp, Google, and Instagram. Each review often included photos — a burnt latte, a messy table, or a beautifully presented pastry. Manually sorting through text and images was slow, error-prone, and inconsistent. The result? Angry customers, missed opportunities to showcase praise, and a bruised brand image.
The core challenges:
- Delayed responses: Average time to reply was 48 hours, exceeding customer expectations.
- Inconsistent tone: Each team member had a different voice, leading to a fragmented brand persona.
- Blind spots: Visual cues in photos were often overlooked, missing crucial context for complaints or compliments.
The Solution: Multimodal AI Prompts
The chain deployed a custom AI workflow using multimodal LLMs (like GPT-4 with vision) and a set of carefully crafted prompts. These prompts didn't just analyze text; they also 'read' images, extracted sentiment, and generated empathetic, on-brand responses. Here's the breakdown of the prompts they used — you can adapt them to your own business.
1. The Initial Triage Prompt
This prompt sorts incoming reviews by urgency and sentiment, flagging those that need immediate attention.
You are a customer service triage assistant. Analyze the following review, including any images, and classify it as: 'urgent', 'negative', 'neutral', or 'positive'. Provide a one-sentence reason for your classification. If an image is present, describe any visual issues (e.g., burnt food, messy table).
Review text: [paste text here]
Image URLs: [if any]
Example output: urgent - The photo shows a broken chair, and the text mentions a safety hazard.
2. Sentiment Analysis with Context
This prompt digs deeper into the emotional tone, considering both text and visual cues.
Analyze the sentiment of this review (text + image) on a scale of 1 (very negative) to 5 (very positive). Consider the text tone, emojis, and any visible elements in the image. Output: 'score' (1-5), 'emotion' (e.g., angry, happy, disappointed), and 'key drivers' (specific aspects mentioned or shown).
Review text: [paste text here]
Image URLs: [if any]
3. Drafting the Response – The Core Prompt
This is the workhorse. It generates a personalized reply that acknowledges both text and visual specifics.
You are a brand manager for a coffee chain. Write a response to the following customer review. The response should:
- Thank the customer by name (if provided).
- Acknowledge specific details from the text and any images (e.g., 'I'm sorry to see your latte was burnt' or 'We're thrilled you loved our new croissant!').
- Match the brand's friendly, professional tone.
- Offer a concrete solution for negative reviews (e.g., we'll send you a gift card) or a warm note for positive ones.
- Be under 150 words.
Review text: [paste text here]
Image URLs: [if any]
Example output: "Dear Sarah, thank you for the honest feedback. We're so sorry to see your cappuccino was lukewarm — that's not our standard. We've shared your photo with our barista team to fix our temperature control. Please check your inbox for a free drink voucher. We hope to see you again soon!"
4. Handling Negative Reviews with Empathy
This prompt focuses on de-escalating anger and offering remedies.
Rewrite the following response to be more empathetic and solution-oriented. Avoid defensive language. Suggest a concrete remedy (e.g., refund, remake, discount). Keep it concise (under 100 words).
Original reply: [paste draft]
5. Boosting Positive Reviews
This prompt turns praise into social proof, encouraging the customer to share.
Write a response to a positive review that: 1) thanks the customer, 2) mentions a specific detail from the review or photo, 3) subtly invites them to visit again or follow on social media. Keep it under 75 words.
Review text: [paste text here]
Image URLs: [if any]
6. Handling Mixed Reviews
Some reviews are a mix of praise and criticism. This prompt addresses both.
Analyze the review for both positive and negative aspects. Generate a response that:
- Acknowledges the positive points.
- Apologizes for the negative points.
- Offers a solution.
- Ends on a positive note.
Review text: [paste text here]
Image URLs: [if any]
7. Multilingual Support
For global chains, this prompt translates and localizes responses.
Translate the following response into [language], keeping the tone and meaning intact. Adapt cultural references if necessary.
Response: [paste text here]
8. Automated Tagging and Routing
This prompt categorizes reviews for internal tracking.
Assign tags to this review from the following list: 'food quality', 'service', 'ambiance', 'cleanliness', 'speed', 'value', 'other'. Also, suggest the appropriate department to handle it (e.g., kitchen, management).
Review text: [paste text here]
Image URLs: [if any]
9. Weekly Summary Report
This prompt aggregates feedback trends for management.
You are a data analyst. Given the following list of reviews (with images), generate a weekly summary report with:
- Top 3 common issues (with example quotes/images).
- Top 3 praise points.
- Sentiment trend (improving, declining, stable).
- Actionable recommendations.
Reviews: [paste list]
10. Competitor Benchmarking
This prompt compares performance with competitors.
Analyze the following competitor reviews and our own reviews. Identify areas where we can improve based on what customers praise about competitors. Output a bulleted list of actionable insights.
Our reviews: [list]
Competitor reviews: [list]
11. Repurposing Content
This prompt turns positive reviews into marketing material.
Extract the most positive and specific quotes from these reviews. Rewrite them into short marketing snippets (up to 2 sentences) that can be used on social media or our website. Attribute them to the review source.
Reviews: [paste list]
12. Crisis Detection
This prompt flags potential PR disasters.
Scan the following reviews for any that mention health hazards, illegal activities, or extreme anger. Flag them as 'crisis' and suggest immediate escalation steps.
Reviews: [paste list]
The Results: 10 Minutes, 4.8x Faster, 30% More Engagement
After three months, the chain saw dramatic improvements:
- Response time: Down from 48 hours to 10 minutes on average.
- Customer satisfaction: Review scores improved by 15% (from 3.8 to 4.4 stars).
- Brand consistency: All responses followed the same tone and quality.
- Staff efficiency: The team now handles 5x more reviews without extra hires.
Key takeaway: The prompts didn't replace human judgment; they amplified it. Staff now review and approve AI-generated responses, ensuring accuracy and a human touch.
Getting Started: Your Turn
Start small. Pick one platform, craft a few prompts, and test them on your past reviews. Iterate based on outputs. You don't need a data science team — just a clear understanding of your brand voice and a willingness to experiment.
Remember: The goal isn't to automate away empathy, but to free up time for the moments that truly need a human. With the right prompts, you'll not only respond faster but also build deeper customer relationships.
Comments