Updated
Since August I have been asking one question, the same wording each time, to as many people as will answer it. Should companies use AI to create content, and why? I asked it face to face on the floor at eTail Boston, and I have been asking it by LinkedIn message since, with the same offer each time that the answer can stay anonymous. Hundreds of people have answered so far, the count is still climbing, and the question is still open on our AI content study page if you want to add yours. What follows is what the pile says so far, in the respondents’ own words wherever a person agreed to be named.
The Largest Group Said Yes, With a Person Holding the Pen
The largest group by a wide margin, in person and by message alike, said yes and then attached a condition that turned out to be the whole answer, a human has to own, edit, or approve what the machine produces. Sandeep Mohan put it in a sentence I have quoted more than any other since he sent it, “AI handling volume while a person still owns the narrative and the facts behind it.” Chris Parsons of Retail Rewired drew the same line on the conference floor, “if you start with a prompt, and you have a perspective, and then you use AI to clean that up, like an assistant or an editor, that’s the right balance.” Reba Hatcher, Chief Commercial Officer at ButcherBox, said companies “should definitely use AI when creating content, but there is a process for figuring out how you fit yourself into it so it doesn’t feel AI generated,” and closed with a line I keep coming back to, “Remember what makes you uniquely you.” Lisa Goldenthal’s version was “AI should accelerate human thinking, not replace it,” and Joe Williamson’s was “if you build tools that you can trust, you should be able to trust the tools you build,” with the caveat that autonomous output is a different matter and a person stays in the loop.
The Flat Yes Camp Argues From Where the Market Already Is
A smaller group answered yes with no condition attached, and their reasoning tended to be about the market rather than the craft. Jeremy Goldman of Rethink Retail said “no is actually a more empirically wrong answer here, because no means you’re entirely avoiding it,” and named the danger as “just AI slopping it and cutting and pasting.” Kyle Standaert of Retention.com framed it as consumer experience, “companies should take any approach possible to create a better experience for their consumers,” while granting “there’s a lot of wrong ways to use AI, so it’s a bit of a loaded question.” Benjamin Noble went furthest, and he knew it, “I absolutely think companies should use AI to create content (and I’m aware that may place me in a minority),” on the grounds that corporate collateral “tends to be pretty robotic, even when it’s written by a human,” and that “ideas outweigh style.” His closing line stayed with me, AI “is almost like a prosthetic for people who can’t find the right words.” He loses me at ideas over style, because style is what wins readership. Yes, you can get ideas from a model, and then you take the ideas and make them yours, polish them, put your spin on them, and the piece that skips that step is the piece nobody finishes.
The No Camp Argues From the Reader, and From the Company Itself
The people who said no were a minority in both halves of the study, and their reasons split two ways. Matt Swan argued from reciprocity, “I write every word myself because I wouldn’t want to read AI written prose, so I don’t expect my audience would either.” I do not hold myself to his standard, I ask the model for an outline or a draft and then I do the writing, and I respect that he draws the line further back than I do. William Jowett of Bidnamic argued from detection, “I can really tell when it’s been procedurally generated by a computer,” and from what readers want, “people are latching on to the fact that they don’t want to just be fed this slop.” Christa Marie Clark made the argument nobody else made, that the case for human writing is about the company and not only the audience, “for companies that want to preserve their soul, which is created by the people that work there, human-created content is our past and future.” Mathew Adams, Director of R&D at Adeia, said no because a model “is just regurgitating and repackaging what it’s been trained on,” then added in the same breath, “we use it all the time.”
A Fourth Group Refuses the Question and Answers a Better One
A steady share of people declined to give a single answer, and the refusals taught me more than the clean answers did. Anna Meredith of Lume Deodorant said “you’d be hamstringing yourself not to use AI in some part of the creative process,” then drew the boundary, “whenever you’re trying to connect in an authentic way, that’s not the time for AI.” Tom Puchalski split it by audience, the final piece “should be written, rewritten and proofread by a human because the content is being created for humans,” while content written for machines, the pages that exist to be read by a search or answer engine, can be written by one. Matt Dunphy tied the answer to what a person values, noting that business people “love to use AI for art, and criticize casual use of AI for things like contracts, proposals, and budgets,” while creative types do the reverse, so “views on AI give a window into what a particular person values.” Two respondents rejected the word “should” outright. George Lawton compared it to asking whether people should use traffic to get to work, something that “just arises and sometimes you have to suffer through it,” and Chris Sheridan put the safeguard in a different place, his position being that AI content is fine so long as it is “labeled AI-generated so that the consumer knows it was produced by a non-human entity.”
Kelsey Knight’s Answer Moved the Question Off Taste and Onto Law
One answer arrived with a reason for changing, which no other answer did. Kelsey Knight wrote that “the new EU laws announced yesterday have evolved my thinking on this a bit,” and then gave the position, “Yes for research, fact-checking, and speed but never for origination. AI is derivative by construction, meaning it can only recombine what already exists, so anything that starts in a model starts as someone else’s work with the attribution stripped out.” Her expanded note argued that with the EU’s transparency rules in effect, “origin is becoming a permanent, auditable property of the file rather than a matter of taste,” and that the market would sort itself between brands that used AI to move faster and brands that used it to replace having a point of view. Whether or not you agree with her reading of the rules, she is the one respondent who located the answer in something outside her own preference, and I think that is where the question is heading. My own position is simple, companies need to start labeling it, whether or not a regulator is making them.
Where the Two Halves of the Study Differ
The face-to-face answers and the message answers came from different rooms and read in different registers, so I report them as two sets and do not blend them into one number. On the floor at eTail Boston, people answered on the spot with a recorder running, so the replies were short, confident, and heavy on the word “slop,” which came up unprompted again and again. When someone said slop to my face they were picturing bland AI content that lacks emotion, the piece that reads fine and leaves nothing behind, and nobody I talked to needed the word defined. By message, people had time, and the replies ran longer and split into finer positions, the disclosure answer, the inevitability answer, the reciprocity answer, none of which anyone gave me in person. The order of the answers held in both frames, yes with a human first by a wide margin, then the flat yes and the flat no as smaller camps of about the same size, then the people who would not pick. I am not going to put a percentage on any of that, because this is a convenience sample of people who were willing to answer a stranger, and the value of the study is in the reasons people gave and in running the same question again each quarter to see whether the reasons move.
What I Take From Hundreds of Answers
What surprised me was that people at AI companies said no to AI. The person with AI in the job title, the AI architect, the AI strategy lead, the R&D director at a company that builds the technology, answered no more often than I expected, and they answered it from the inside, knowing what the models do. People recognize AI writing, they say so unprompted, and they resent being handed it, which matches what our earlier piece on readers telling when a machine wrote it argued from the other side of the desk. At the same time few people, including the people who said no, work without the tools now, and the disagreement is over where the human sits in the process, since the machine is already in the building. The arrangement the largest group described is the workflow Content Cucumber runs, a person with a perspective at the start and a person with their name on it at the end, and it is why the question of what a blog post costs comes down to how much of that human time a company is buying. The study stays open, the wording stays the same, and answers that arrive after publication go into the next quarterly update instead of back into the post you are reading.
If you want writing that starts with a person and ends with a person, see how Content Cucumber prices human-written content, and if you have an answer to the question, add it to the study.
Explore our writing services
Content Services →Keep Reading
Frequently Asked Questions
One question, in the same wording every time, should companies use AI to create content, and why. It was asked in person at eTail Boston and by LinkedIn message, with the offer that any answer could stay anonymous.
The largest group by a wide margin said yes with a condition, a human has to own, edit, or approve what the machine produces. That held in both the in-person and the message halves of the study.
Yes, a minority in both halves said no, arguing from the reader, that people can tell and resent it, and from the company itself, that human-created content preserves what the people who work there bring.
The two halves came from different rooms, short recorded answers on a conference floor versus longer written replies, and the study is a convenience sample, so the two sets are reported as two sets and no percentage is put on either.
Yes, the study stays open at contentcucumber.com/ai-content-study with the same wording, and answers that arrive after publication go into the next quarterly update.
