Abandoned Cart Email Generator: Inputs, Logic, and Recovery Copy

An abandoned-cart generator is useful when it turns a real cart event into a safe recovery specification. The generator should know what was in the cart, whether checkout was started, and what the business is allowed to offer.

Required inputs

Output contract

Ask the generator for:

1. subject and preheader options;
2. one primary action;
3. item-level fallback behavior;
4. timing for each step;
5. suppression and exit conditions;
6. measurement events;
7. a list of claims requiring human approval.

Example brief

Goal: recover carts without discounting by default
Audience: known subscribers with marketing consent
Cart age: 2 hours to 3 days
Products: current price and stock required
Tone: helpful, direct, no false scarcity
Exit: purchase, cart cleared, unsubscribe, product unavailable

Recommended sequence

Message 1 — reminder: show the cart and return link. Message 2 — help: answer shipping, fit, or payment questions. Message 3 — final check: only if the cart is still eligible and the customer has not received too many messages.

Do not add a discount just because the model thinks it will improve conversion. Make incentive use an explicit input.

QA and evaluation

Render with an empty name, one item, multiple variants, an unavailable item, a changed price, and a customer who purchased between enqueue and send. A human should approve discount claims, inventory claims, and legal language.

Measure recovered orders, margin, unsubscribes, support contacts, and repeat exposure. A higher recovery rate is not a win if the flow erodes margin or annoys customers.

Turn cart data into decisions

The generator should not treat every cart as the same opportunity. A cart containing a replenishable commodity, a high-consideration product, and a gift have different useful next steps.

This classification can be a deterministic ruleset that the model explains. It should not be inferred from vague copy such as “looks interested.”

A safer generator prompt

Give the model structured facts and require it to mark unknowns rather than fill them in:

Write a three-message cart-recovery draft from the supplied event.
Use the exact item names, prices, currency, and links provided.
Choose a message angle from reminder, help, or availability.
Do not create a discount, deadline, inventory claim, testimonial, or shipping promise.
Return send delay, eligibility, suppression, variables, and unresolved claims.

For production, keep the prompt version and input snapshot with the generated draft. That makes it possible to explain why a message used a particular product or policy after the catalog changes.

Recovery rules worth encoding

• Enqueue only after a cart has been idle for the minimum window; repeated page views should not reset the clock indefinitely.
• Re-check price, inventory, tax, shipping eligibility, and the cart URL immediately before rendering.
• Cancel queued messages on purchase, cart conversion, explicit opt-out, account deletion, or a product becoming non-purchasable.
• If several carts exist, choose the most recent eligible cart or explain the consolidation rule.
• Cap exposure across cart, browse, price-drop, and recommendation programs.
• Keep transactional payment-failure handling separate from promotional recovery.

Evaluate the output, not just the copy

Create fixtures for an empty cart, a single variant, mixed inventory, a changed price, an international address, an anonymous visitor, and a purchase that arrives while the message is queued. Check both the generated text and the workflow metadata.

Review at least these outcomes:

The best generator output is a small, testable recovery plan that a marketer can approve—not an impressive paragraph detached from cart reality.