Can AI create content people actually want to read?

“And, O, Tyger! What shouts (with thee?)
What glo­ry (was with us?)
Was ’mid those forests (thee?)
When our first ances­tor (and thee?)
Being (a tiger?) slew.
I can imag­ine you
Proud­ly roar and say,
‘I am the tiger’”

The above text is part of an AI-gen­er­at­ed rewrit­ing of William Blake’s clas­sic poem The Tyger, as pro­grammed by Amer­i­can writer Gwern Bran­wen using OpenAI’s lat­est GPT‑3 tech­nol­o­gy. Apart from the Pythonesque rep­e­ti­tion of “thee”, it unerr­ing­ly evokes the spir­it of mys­tic roman­ti­cism – and, cru­cial­ly, it feels poet­ic. 

Fright­en­ing, right? Maybe not. It should come as no sur­prise that machines are get­ting quite good at gen­er­at­ing inter­est­ing strings of words. Fed on an ever-expand­ing diet of online con­tent, their algo­rithms can con­tin­u­al­ly learn how to improve their out­put. 

Numer­ous soft­ware ser­vices, such as Nichesss, Copy­AI and Writeson­ic, are giv­ing mar­keters access to these sys­tems for copy­writ­ing pur­pos­es. They can cre­ate social media posts, blog out­lines, prod­uct strate­gies and cor­po­rate slo­gans. But is any of it actu­al­ly worth read­ing? 

Impressive, but not fresh

Kim Dar­ragon, a mar­ket­ing expert and founder of the Kim Does Mar­ket­ing con­sul­tan­cy, has test­ed out some of the soft­ware. She thinks that “AI-gen­er­at­ed con­tent can be pret­ty impres­sive. With a hand­ful of key­words and in just a few sec­onds, you can have some sol­id copy for prod­uct descrip­tions, Insta­gram cap­tions, Google meta-descrip­tions and LinkedIn ads.” 

But the impres­sive­ness goes only so far. Because AI’s relies on exist­ing data, “it’s not com­ing up with fresh ideas”, Dar­ragon says. “While the con­tent often looks good at first glance, it’s often shal­low and still can’t repli­cate the depth of human-gen­er­at­ed copy.” 

Bernard Huang agrees that such writ­ing is often lim­it­ed to fac­tu­al para­phras­ing. He’s the co-founder of Clearscope, which uses an “AI-pow­ered plat­form” to help mar­keters pro­duce con­tent that bet­ter meets the rank­ing cri­te­ria of var­i­ous online search engines. Despite being immersed in this sec­tor, Huang doesn’t “explic­it­ly use any AI tools” to enhance his own writ­ing. He also doesn’t expect robot scribes to replace many human copy­writ­ers. 

“My hunch is that the things that will get auto­mat­ed have a fair­ly low impact,” Huang says. “There’s already too much con­tent on the inter­net. Mak­ing it eas­i­er to pro­duce doesn’t help to keep that inven­to­ry down.” 

At some point, quan­ti­ty can can­cel out qual­i­ty. As Dar­ragon puts it: “If every brand is sud­den­ly fan­tas­tic at cre­at­ing con­tent, brand copy may begin to sound the same – and every­thing will start to be a bit bor­ing.”

Android paranoia

AI’s mim­ic­ry of exist­ing pat­terns of writ­ing – and, by def­i­n­i­tion, thought – pos­es oth­er risks. AI-gen­er­at­ed con­tent “can also include human bias­es, faux pas and oth­er nasty things. After all, it’s just an algo­rithm,” Dar­ragon says. “At a time when brands need to be super-care­ful with lan­guage, this is some­thing to be wary of.”

Mar­ket­ing pro­fes­sion­als are indeed becom­ing increas­ing­ly con­scious of the dan­gers of algo­rith­mic bias. Christo­pher Ken­na is co-founder and CEO of Brand Advance, a media net­work that helps com­pa­nies to find a more diverse audi­ence for their mar­ket­ing cam­paigns. He says: “Despite the sig­nif­i­cant ben­e­fits of AI, there are chal­lenges, because the data sets that it learns from are steeped in the his­tor­i­cal social inequities of the real world.”

Such con­cerns are espe­cial­ly impor­tant in the case of sen­si­tive projects such as the NHS’s recent dri­ve to encour­age BAME peo­ple to seek vac­ci­na­tions. Brand Advance has been help­ing to fine-tune some of its cam­paigns using IBM Watson’s lan­guage sys­tem, which osten­si­bly reads what is being writ­ten about them online. 

“We can clas­si­fy an article’s sen­ti­ment to gauge whether the reac­tion is pos­i­tive or neg­a­tive,” Ken­na says. “We com­bine this with real-time data about arti­cle engage­ment to help inform our audi­ence strat­e­gy.” 

A tool, not a rival

AI will prob­a­bly have the most impact on such data-heavy and ana­lyt­ics-dri­ven mar­ket­ing tasks in the com­ing years, accord­ing to Dar­ragon, who adds: “You can’t beat a machine for that.”

But mar­ket­ing cov­ers so many more dis­ci­plines, of course. The pro­fes­sion exist­ed long before these AI-pow­ered tools and tech­niques came along – and it will prob­a­bly out­live them. Indeed, mar­ket­ing has often been pre­sent­ed as one of the most automa­tion-proof careers, since it strad­dles cre­ativ­i­ty and human empa­thy. With this in mind, a wide­ly held view in the pro­fes­sion seems to be that, regard­less of how much the dai­ly work of a mar­keter changes, AI should present more of an oppor­tu­ni­ty than a threat over the com­ing years. 

No mat­ter how sophis­ti­cat­ed the AI toolk­it becomes, a thought­ful and imag­i­na­tive strate­gist should still be need­ed to make sense of the whole, accord­ing to Dar­ragon.

“The roman­tic in me (and the wish­ful thinker) hopes that mar­ket­ing will con­tin­ue to include a crit­i­cal human func­tion,” she says. “It lies in steer­ing a brand’s mar­ket­ing ship in the right direc­tion, mak­ing the big strate­gic calls and keep­ing the beat­ing heart of a brand alive – that je ne sais quoi that, I hope, no machine can replace.” 

Huang is sim­i­lar­ly san­guine. “The val­ue of cre­ativ­i­ty will shine in a world of automa­tion,” he says.

So, no, your new favourite author is unlike­ly to be a machine, but the next prod­uct descrip­tion you fall for might just have been helped along by one.  

The MarkeTuring test

While com­pil­ing this arti­cle, I had a lit­tle play with some of the lat­est copy­writ­ing AI soft­ware by typ­ing in short prompts relat­ing to the top­ic of AI in copy­writ­ing (which is about as meta as it gets). Here’s what came up.  

Some­times, what it came up with was plain­ly ridicu­lous: “What hap­pens when you do spend the time and mon­ey mar­ket­ing your busi­ness? Nobody does!” (Copy­AI.)

On oth­er occa­sions, it was incred­i­bly pro­found: “Humans are emo­tion­al crea­tures. We make deci­sions on gut instinct and our every­day actions are dri­ven by emo­tion. AI can help mar­keters to under­stand their cus­tomers’ needs and wants based on their past pur­chas­es, but mar­keters must first under­stand where the human brain is head­ed.” (Nichesss.)

And, in a few cas­es, it pro­duced dis­arm­ing­ly ver­bose, yet some­how chill­ing, robo-threats: “As a mar­keter, do you know what it means to have your brand align itself with the val­ues of AI? You’d bet­ter be able to answer this ques­tion in the affir­ma­tive if you want to stay ahead of the curve and make sense of how it will affect your busi­ness.” (Copy­smiths.)