Establish the brand baseline before writing a single word
Most people open ChatGPT, type in a vague request for a winback email, and wonder why the output reads like generic marketing fluff. If you do not feed the model your brand context first, it will invent details that sound plausible but fall completely flat.
The first step in my workflow is feeding the AI our store URL and asking it to analyse the target audience, brand tone, and value propositions. You want the model to understand who you are speaking to before it writes a single headline. Doing this upfront has saved me hundreds of hours of drafting time across our client accounts.
Feed your top sellers, testimonials, and active offers
Once ChatGPT understands the brand voice, you need to give it concrete commercial assets. Do not rely on the model to guess which products drive your revenue. I prefer to manually drop in the exact URLs for our top three best-selling products so there is zero ambiguity.
Along with product links, instruct the model to pull in customer testimonials, current promotions, and recent blog posts. This gives the AI the raw material it needs to craft compelling body copy, overcome objections, and present relevant cross-sells to customers who have gone quiet.
Verify the facts to catch hallucinations early
Large language models will occasionally make up product specs, ingredients, or return policies out of thin air. It is infuriating when an email goes to review with fabricated claims. That is why you must explicitly prompt the model to confirm where it sourced its information before generating copy.
Ask the AI to confirm its findings and list the reference points it gathered from your site. If anything looks off, correct it immediately. Even with this guardrail in place, I always have a copywriter review the final output before anything goes live. AI builds the foundation, but a human needs to verify the facts.
Standardise your email formatting rules
Before scripting individual emails, establish strict structural boundaries. If you skip this, you will get walls of unbroken text that look terrible inside an inbox and fail to convert mobile readers.
- Subject lines: Set a strict limit of 50 characters to prevent truncation on mobile screens.
- Preview text: Require concise, complementary secondary copy that supports the hook.
- Creative prompts: Ask the model to suggest header image concepts and badge placements to guide your design team.
Once the model confirms these formatting rules, it will output structured drafts that your team can drop straight into Figma or Klaviyo templates without major restructuring.
Scripting email 1: The cross-sell and value nudge
The first email in a winback sequence should never rely on heavy discounts. Handing out margin cuts to early churn risks is leaving money on the table. Instead, the primary focus of this initial message is to re-engage the customer by introducing best sellers and complementary collections they have not tried yet.
Prompt the model to research the specific collection URLs you want to highlight, extract their unique selling points, and build a persuasive case for returning. The output should provide clear product recommendations, customer review snippets, and a strong call to action that reminds the customer why they bought from you in the first place.
Scripting email 2: The delayed discount offer
If a customer does not bite on the first email, it is time to escalate. In our typical flow architecture, this second email fires after a calculated delay (often 45 days after the initial lapse plus a 7-day buffer, meaning roughly 52 days since their last order). At this point, offering a financial incentive makes commercial sense.
Feed the model your specific offer (such as a 10% discount code) along with instructions to re-feature top collections. The prompt should position the discount clearly in the copy and provide clear urgency. One thing I like to do is ensure the discount code is placed directly in the suggested header banner so mobile scrollers catch it immediately.
Final Thoughts
Using ChatGPT to script your Klaviyo flows is not about replacing human strategy. It is about removing the friction of the blank page and accelerating production across your automated sequences. Feed it structured data, enforce strict formatting, and let your copywriter polish the final draft.
Automate your revenue with high-performing flows
If you want automated lifecycle flows built, tested, and optimised for your store, take a look at our email automation services to see how we help ecommerce brands maximise customer lifetime value.
