We knew A.I. could write the copy. Could we replace the template text without compromising the design?
I led an A.I. personalization system that customizes 40K templates, from classic to kitschy, modern to ironic, retaining the design’s style, tone, and balanced composition.
The system evaluates a template, identifies what a person would need to change on it, selects short questions to gather input, and rewrites the copy to fit. It retains the nuance. It greatly simplified graphics editing on mobile.
The problem
Paperless Post stands above other event platforms with first-rate design. Its custom editor makes this possible, similar to Squarespace and Canva in one product: it configures event pages and it edits dynamic graphics. It’s used by both customers and the fantastic in-house designers.
On desktop, personalizing a design is a challenge, but manageable. On mobile, that same act is an ordeal. Users needed to tap and edit a small date line, adjust font-size and element shape to fit a space, and remember how to spell out a wedding date in longform script. Hundreds of thousands of people would start an invite on mobile; a small percentage would complete the task there.
We knew AI could write the copy. Could we replace the template text without compromising the design?
It gets complicated
A template is a designed composition. On a formal invite, replacing “Twentieth of May in the year two thousand twenty-six” with “June 30th” doesn’t just change a date — it leaves a hole, visual imbalance. Each word is data and an interconnected design decision.
And the range is enormous. The same pipeline runs on an ironic thirtieth birthday, a memorial service, a baby announcement, a quinceañera, and a business product launch. Each has its own tone, and each has words that should and should not be edited, sometimes hidden in puns.
And a date is not simply a date. A TGIF happy hour, a graduation year, a child’s birth announcement, a grandparent’s date of passing and memorial; an Xmas party on the 23rd, and a New Year’s party in ’26 celebrating ’27.
A name? Let’s not talk about Karen hosting a gender reveal party for Baby Johns to celebrate Stacey and Joe. How do you guide AI to reliably ask for a name in this scenario and dozens of others?
How it works
The stages:
- Analyze the template to identify the text that needs personalization
- Choose the relevant and most useful questions
- Receive user input; generate and replace text
- Evaluate the results
More info and insights coming soon.