Skip to main content

How to write better prompts: A practical guide for B2C marketers using Klaviyo’s ChatGPT app

Profile photo of author Kelsey Capps
Kelsey Capps
9 min read
Artificial intelligence
January 28, 2026

If you’re like most B2C marketers, your days are already packed.

From launching campaigns, to troubleshooting attribution, to keeping up with shrinking budgets and rising expectations, time, data access, and bandwidth are constant obstacles to doing your best work. You’re asked to move faster, with fewer resources, on tighter timelines without sacrificing personalization or performance.

That’s exactly why we built the Klaviyo app for ChatGPT: to make your data easier to access, your insights faster to uncover, and your workflow more conversational so you can get more done without opening another tab or waiting on another report refresh.

Today, you can use the Klaviyo app in ChatGPT to see how your campaigns and flows are performing, get recommendations, and jump straight into next steps. The app returns real-time performance tables, deep-dive analytics, and plain-language insights, all from a simple question.

To get the best results and avoid hallucinations, though, it helps to know what to ask.

This guide will not only explain how to prompt ChatGPT for data analysis, campaign optimization, ideation, and strategic recommendations so you can do your best work faster, no prompt engineering background required, but also what the Klaviyo app and MCP API can and cannot return accurately.

Why prompts matter (hint: they don’t have to be complicated)

Many marketers think “prompt engineering” means writing long, complex requests. The truth is that great prompts mirror the way you already think about your work.

You regularly ask yourself things like:

  • “Why did the open rate for my flow named [Flow Name or ID] drop this week?”
  • “What should I optimize in my account before BFCM, given my current campaign and flow performance?”

Now, you can ask ChatGPT those same questions and instantly get back the data you need without dashboards, spreadsheets, or digging.

The goal isn’t to learn a new language for a new tool. It's to ask better versions of the questions you're already asking.

However, because ChatGPT may attempt to answer questions using its general knowledge when the app can’t return a request, it’s also important to understand how to write prompts in such a way that the app will reliably call the correct Klaviyo API instead of hallucinating data.

Three principles for better prompts

1. Be direct about what you want to know

ChatGPT works best when you provide a clear intention. Simple is best:

  • “How are my campaigns performing this week?”
  • “Show me the last 90 days of flow performance.”
  • “Compare my Abandoned Cart Reminder flow’s performance this month to its performance last month.”

These natural-language questions map directly to live Klaviyo performance data and return interactive tables you can drill into.

2. Add context when you want deeper insight

If you want ChatGPT to analyze or interpret your data, first have it pull data so that it has that data in its context window. Then give it more context, like business goals, strategy, or other things to align to, and direction:

  • “Pull performance for my last 10 campaigns, then analyze which are under performing.”
  • “Pull my “Welcome” flow performance for the last 30 days, then tell me what happened and why.”
  • “Analyze my summer sale campaign performance and give me ideas for a better strategy retention strategy”

The app will provide metric comparisons, micro-insights, and next-step suggestions like changes to subject lines, for instance.

3. When in doubt, click Analyze

The fastest way to turn insights into action is built right into the UI. Any time ChatGPT returns a clickable table––whether you’re exploring segments, campaign performance, or customer cohorts—you’ll see an Analyze button. One click automatically generates a deeper interpretation of the data, along with clear recommendations on what to do next.

Think of it as your instant “now what?” moment: Which cohorts are most primed for a follow-up? Which messages need tightening? Should you resend, resegment, or refine your offer? The Analyze view lays it all out for you.

Foundational prompts every marketer should start with

Here are some of the best starting prompts for speed, clarity, and impact.

Performance reporting: Fast visibility without dashboards

These prompts show you campaign and flow data in real time. They are ideal when you're on the go or juggling competing priorities:

  • “How are my campaigns performing this week?”
  • “What are my top and bottom performing campaigns this month?”
  • “Show me the last 90 days of campaign performance.”
  • “Show me flow performance from the last seven days.”

You’ll get rich, mobile-friendly tables with revenue, open rates, click rates, conversions, and more. Then, tap Analyze for deep-dive insights.

Optimization & strategy: Quick wins for busy teams

These prompts help you turn insights into action. They are critical for marketers who don't have time to manually sift through data or build new tests from scratch:

  • “Pull performance for my flow called [Flow Name] and highlight the top opportunities to improve conversion.”
  • “Review my recent campaign and flow metrics and tell me which optimizations will have the biggest KAV impact before BFCM.”
  • “Scan my account for underperforming campaigns or flows and recommend where I should focus to lift conversion this week.”

These are perfect for marketers facing resource limitations, and who need good ideas quickly for improving performance.

Creative ideation: Get unstuck fast

When bandwidth is thin and pressure is high, ChatGPT can help generate ideas:

  • “Create an urgency-driven email template for my sale.”
  • “Draft a campaign promoting a new product.”
  • “Create a multi-language version of this SMS in French and German.”

Because these prompts sit alongside live performance data, you can align creative with what actually works, not just what sounds good.

Prompt templates you can copy/paste today

  1. Diagnose performance: “Analyze this campaign’s performance and tell me what’s driving the click rate change week over week.”
  2. Spot problems early: “Pull my recent campaigns, then analyze which have declining metrics in the last 3 days.”
  3. Optimize a flow: “Analyze the campaign named [Exact Campaign Name] and give me three recommendations to increase conversion.”
  4. Build content quickly: “Draft a product announcement email using a friendly, high-converting tone that fits a beauty brand.”

How to troubleshoot prompts when something feels off

Even simple prompts can sometimes return results that aren’t what you expected. Here’s how to quickly course-correct:

  • If ChatGPT fails to return data or returns data you cannot verify in Klaviyo: Ask the LLM which API tool it called and help it course correct with something like “Pull my Klaviyo campaign performance reporting,” then ask your question again.
  • If the output is too generic: Add specifics like a timeframe, a flow name, or a goal. Example: “Analyze my welcome flow performance over the last 30 days and identify one optimization.”
  • If you need a strategic lens: Tell ChatGPT your objective. Example: “Assume my goal is to drive repeat purchases. What should I optimize next?”
  • If you want the fastest path to action: Follow every table with: “What should I do next?” This alone saves hours each week.
  • If ChatGPT says a flow or segment doesn’t exist: Always reference flows or segments by their exact name or ID. Example: “Analyze my flow named ‘Abandoned Cart Reminder.’”
  • If performance comparisons behave inconsistently: Ask ChatGPT to pull performance data first, then request the comparison.

To avoid hallucinations, refrain from asking for reporting that isn’t currently supported by Klaviyo’s MCP.

  • “How are my flows performing by channel?” ✔️
  • “Show me flow performance by segment.” ❌ (Not supported)

Great prompts help ChatGPT correctly call Klaviyo’s MCP tools (i.e. our public APIs). When a prompt requests unsupported use cases, ChatGPT may guess. Using the techniques in this guide will reduce hallucinations and ensure you get accurate, actionable insights. You can learn more about Klaviyo’s API tools here.

The real goal: Make your work easier

B2C marketers don’t need another tool to learn or another dashboard to dig through. They need clarity without friction, insights without delay, and recommendations without guesswork

The Klaviyo app for ChatGPT is designed to remove blockers and help marketers do their best work, no matter where they work. And great prompts are the key to unlocking speed.

Kelsey Capps
Kelsey Capps
Kelsey Capps is a product lead at Klaviyo overseeing the company’s reporting and analytics strategy to help brands transform data into actionable insights. With 16 years of experience in CDPs and MarTech, she’s worked with entrepreneurs and global enterprises alike to turn data into strategy that drives results. Her focus is on making analytics approachable and impactful—empowering marketers to personalize at scale, build stronger customer relationships, and accelerate growth. Prior to Klaviyo, Kelsey was driving the strategic roadmap for Square’s Marketing platform for sellers.

Related content

How brands are adapting to agentic shopping
Agentic shopping explains how brands put AI agents and tools to work across discovery, service, and conversion in AI-driven commerce.
How AI is Rebuilding the Walled Garden in Commerce
AI is transforming how consumers discover brands—but LLMs may be rebuilding the same walled gardens brands fought to escape. Why owning your customer data matters more than ever.
Practical ways to use Klaviyo AI
Learn how K:AI (Klaviyo AI) accelerates strategy, optimization, and customer engagement with unified data and autonomous AI built inside Klaviyo.