Industries / Consumer goods / Greeting cards
Design and writing
Design and writing is the creative work on a card: the picture or lettering on the front, the verse or joke inside, and the choice of which occasions and relationships to make cards for.
5
Milestones achieved, from 1880 to 2026.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1880
A prize for card designs
In 1880 Louis Prang, a Boston printer, offered prizes totaling $2,000 for the four best Christmas card designs and drew more than 800 entries. A contest in February 1881 drew nearly 1,500. The Hudson River Museum says the contests ran from 1880 to 1884, with prizes from $200 for fourth place to $1,000 for first, and that the winners included the artists Rosina Emmet and Dora Wheeler. For the last contest, at the end of 1884, Prang limited entry to 22 chosen artists, and then he stopped. The Henry Ford says the flood of amateur work had held little value for him, "artistically or monetarily."
LessonAn open call brings in volume, and sorting the entries can cost more than the good ones are worth.
Credit
Louis Prang & Co
2015
Cards for serious illness
In May 2015 Emily McDowell, a Los Angeles graphic designer, released what she called Empathy Cards, for people who do not know what to say to someone with a serious illness. She was diagnosed with Hodgkin lymphoma at 24 and was a 15-year survivor. She told NPR that "the most difficult thing about my illness was the fact that it was so lonely."
Credit
Emily McDowell
2023
AI drafts the message
In November 2023 Moonpig added a tool called Smart Text to its card editor. The customer types a short prompt about the recipient, the occasion and even inside jokes, and an AI model drafts the message.
Credit
Moonpig
2025
Prompts alone are not authorship
In January 2025 the US Copyright Office concluded that, with the technology then generally available, "prompts alone do not provide sufficient human control" to make the user of an AI system the author of what it produces. It added that work where a person controls the expressive elements can still be protected. So in the US, a card design generated from a prompt alone is unlikely to qualify for copyright.
Credit
US Copyright Office
2026
Ghostwriting brings guilt
In five experiments with hundreds of participants, the marketing researchers Julian Givi, Colleen P. Kirk and Danielle Hass found that people who used generative AI to write heartfelt messages, from birthday cards to love letters, felt guilty about it. People who bought cards with a preprinted message felt no guilt at all. The authors explain that "greeting cards carry no deception."
LessonA printed verse is accepted because nobody is misled about who wrote it.
Credit
Julian Givi and Danielle Hass (West Virginia University) and Colleen P. Kirk (New York Institute of Technology)
Watch for
AI-written Christmas cards
Royal Mail research reported that 11% of people in Britain planned to use AI to help write their Christmas cards, a figure its headline put at nearly 8 million people. The page does not give the sample size.
Achieved whenan independent survey with a published sample and method measures how many cards carry AI-written messages.
Claimed by
Royal Mail, December 2025
Open
01
Verses read blind
A blind reading test of card verses. For each of 40 occasions, readers see three unsigned verses: one by a professional card writer, one by an AI model, and one by an AI model then edited by a writer. They pick the one they would buy and guess who wrote each.
The briefCredit · unclaimed02
Test the AI-art detectors
A labeled test set of card designs, half made by artists who agree to take part and half made by AI image generators, with commercial and open detectors scored on it. The headline number is how often each detector wrongly flags an artist's design as AI-made.
The briefCredit · unclaimed
Open 01 · the brief
Verses read blind
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A blind reading test of card verses. For each of 40 occasions, readers see three unsigned verses: one by a professional card writer, one by an AI model, and one by an AI model then edited by a writer. They pick the one they would buy and guess who wrote each.
Why it is still open
The tests found so far are informal. In an Instagram poll in February 2023, the data company Appen found that 73 of 89 respondents picked out the Valentine's card ChatGPT wrote; 27 people said they would buy the AI card and 21 the copywriter's. The 2026 experiments by Givi, Kirk and Hass measured the sender's guilt, not the quality of the verse.
Built with what exists
current models such as Claude, Gemini and Mistral Large 4, whose weights are free to download (see the Note "A trillion open parameters"); verses commissioned from freelance card writers; a paid survey panel such as Prolific; a short analysis script.
First step
Commission 20 verses from two freelance writers, have an AI model write 20 to the same briefs, and pilot the test with 50 raters.
Done looks like
120 verses and at least 500 raters, each rating 20 sets. Publish each kind of author's share of first choices with a 95% confidence interval, and how often raters guessed the author correctly.
Open 02 · the brief
Test the AI-art detectors
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A labeled test set of card designs, half made by artists who agree to take part and half made by AI image generators, with commercial and open detectors scored on it. The headline number is how often each detector wrongly flags an artist's design as AI-made.
Why it is still open
The closest study, by Anna Yoo Jeong Ha and six co-authors (2024), set human art in 7 styles against images from 5 generators. It found that the detector Hive and expert artists both did well but made different mistakes, with the experts wrongly flagging more human work. It did not test card designs such as hand lettering, flat illustration or photo collage. The US Copyright Office's January 2025 report means that whether a person made a design affects whether it can be protected.
Built with what exists
Hive's AI-generated image detection and other commercial and open-source detectors; c2patool, the command-line reader for Content Credentials, the provenance data some generators attach to images; current image generators; a spreadsheet of labels.
First step
Collect 100 card designs from artists who agree to take part, and generate 100 to the same briefs.
Done looks like
1,000 designs, 500 by people and 500 by AI, with each detector's catch rate and false-accusation rate published, overall and by design style.