AI copywriting has already arrived. In certain contexts, it already outperforms humans.
Given the right conditions, AI copywriting software beats humans basically all the time.
For instance, I previously covered Baidu’s AI Copywriter, which can:
- Write better product copy than humans
- Pass the Turing test
- Spit out 20,000 lines of copy per second
As mentioned, it’s already being used extensively in China.
Just as with any other cognitive “game,” once AI understands the rules, it always wins.
However, today’s AI copywriting software has serious limitations:
- It requires massive amounts of data
- It’s expensive
- It can’t talk to you or do certain types of research
Only very specific types of short-form copy can be automated
In this article, we’ll take a look at how these limitations impact the automation of copywriting tasks.
English AI Copywriting Software
Let’s start with the current state of the industry.
When it comes to AI copywriting in English, Persado is a leader in the space.
It’s an AI-powered platform that can:
- Create personalized, emotional marketing copy
- Learn your brand’s voice
- Generate AI-driven creative in 25 languages
- Analyze and maximize marketing performance
It’s used by big-name brands like Microsoft, Dell, Turbotax, American Express, and many others.
Persado Analytics is its umbrella analytics platform that:
- Breaks apart marketing creatives into separate variables, such as calls-to-action and descriptors
- Uses AI-driven analysis to determine how these variables impact marketing metrics such as engagement and conversion rates
- Can understand, optimize for, and predict the best creatives for audiences, channels, and campaigns
- Offers “critical insight into how variables like emotion and formatting affect success”
- Predict and create the perfect marketing message for a given campaign
Two days before I wrote this, Persado released a new feature for this analytics suite, Descriptive Insights.
It improves upon its existing toolbox, “enhancing marketers’ holistic understanding of the impact of marketing language across audiences, channels and industries throughout the customer journey.”
So they are clearly continuing to evolve their product.
But Persado and Alibaba aren’t the only players in the market.
There’s also Phrasee, which supposedly outperforms humans 98% of the time.
How Worried Should You Be?
Hurley Write covered this same topic on their website.
They said that despite massive investments, Persado primarily does online ad campaigns and “is a long way from killing off any form of business writing.” After all, “writing ad copy for Facebook is a far cry from writing an in-depth white paper or technical report.”
Now, at this point, I can’t help but remember Gartner’s prediction that, by 2018, 20% of B2B content would be written by machines.
Among the types of writing they predicted for demise by automation…?
White papers.
There several major problems with this type of claim:
- The statement was obviously way off base (along with other predictions they made)
- We have no idea what type of “research” or data went into this statement
- It is misleading and potentially damaging to people who might use such “information” to make business decisions
- It demonstrates a lack of even a basic understanding of AI and white paper writing
You don’t need technical qualifications to see why their claim is wrong (I have some, but they aren’t necessary).
All you need is some common sense.
As we’ll see below, it’s likely that white papers will be among the last things thing to be written by AI.
If ever.
Why?
First of all…
AI Isn’t Intelligent
AI does knowledge work.
That’s not the same as intelligence.
AI methods can currently automate very narrow cognitive tasks, such as playing a game. But it can’t think — that’s the goal of artificial general intelligence (AGI).
From this perspective, we could even say that “artificial intelligence” is a misnomer.
So-called artificial “intelligence” uses techniques such as machine learning and deep learning.
These techniques:
- Have a narrow goal
- Generate models from massive quantities of data
- Use those models to perform a task, such as make predictions
For instance:
- Images of cats train AI to recognize cats
- Images of faces train facial recognition software
- Spoken language trains voice recognition
In each of these examples, specific types of data help AI generate very narrow models that perform a single task.
If you tell an AI to find images of faces on the web, it can do it.
It can also recognize an individual’s face.
And it can even create fake faces of people who don’t exist.
So it’s conceivable — even probable — that AI will be used to generate fake photos for websites and marketing campaigns … photos that are, of course, optimized to maximize engagement and response rates for target audience segments.
However, because AI isn’t actually “intelligent” it can’t strategize.
Without human assistance, AI couldn’t create compositions that tell stories.
If AI Can’t Talk or Think, It Can’t Write a White Paper
AI may be able to “stitch together” narrow tasks that form parts of a larger task, but it can’t create strategic compositions. Especially those that need human collaboration.
White papers, for instance, present a number of insurmountable obstacles for AI:
- They require a combination of storytelling, strategic thinking, and judgement calls
- The products or services being promoted are new — so there are no data sets to feed learning models
- A combination of offline and online research is required
Writers must collaborate with stakeholders, SMEs, etc.
Today, AI is only being used to write short-form copy.
While it is conceivable that AI could write long-form copy (maybe some AI lab experiments already do), certain limitations prevent that, as we’ll see below.
Fake Text Generators, the Marketer’s Ultimate Weapon
As I was writing this article, the Guardian released an article about OpenAI’s new fake text generator.
OpenAI claims that this program is “too dangerous” to be released.
It works like this:
- You feed it some text, even a small amount
- It creates well-written, fake text from that input, which is similar in tone, style, and content
For example, when it was fed a few paragraphs of a Guardian story about Brexit, it generated:
Asked to clarify the reports, a spokesman for May said: ‘The PM has made it absolutely clear her intention is to leave the EU as quickly as is possible and that will be under her negotiating mandate as confirmed in the Queen’s speech last week.’
The 14-word opening line of 1984 generated:
I was in my car on my way to a new job in Seattle. I put the gas in, put the key in, and then I let it run. I just imagined what the day would be like. A hundred years from now. In 2045, I was a teacher in some school in a poor part of rural China. I started with Chinese history and history of science.
The creators worry about people who would use this technology for malicious purposes, so they’re keeping the technology under wraps for now.
AI Has Limitations … But Text Generation Isn’t One of Them
To recap quickly:
- AI can currently only perform narrow tasks, such as recognize images, optimize recommendations, or create fake text
- It is not intelligent, it cannot reason, and it cannot feel
- It requires massive data — Persado and GPT2 both require mountains of data to work
Until the OpenAI program came along, AI couldn’t:
- Read a small text sample
- Understand its content
- Develop similar, relevant, well-written copy
It still can’t do interviews, collaborate with humans, or perform multi-channel research.
Outside of specific contexts, it’s impossible for commercial AI software to keep up with 95-99% of human copywriters.
As of early 2019, AI copywriting software only does short-form copy.
Next decade, who knows?
OpenAI’s project shows that things are changing faster and faster every year.
Combine analytics with software like GPT2 and you have a program that can ingest a small text sample, then automatically generate top-converting copy.
Certain types of copywriting, like white papers, may be forever out of its reach.
Or maybe copywriters will be replaced with Copy Supervisors…
…copywriters who collaborate with humans, feed data to AI, and enjoy long smoke breaks while robots do the work.
Conclusions: The Question Isn’t “Whether”
It’s “how much,” “when,” and “what” copywriting will be automated.
The way I see it, there are two major possibilities:
1. The data barrier keeps the automation rate low.
Augmentation is inevitable, but AI’s data needs may never be overcome.
In this case, most copywriting won’t be machine-generated.
I certainly hope this is the case.
If this is the case, we don’t really need to worry too much.
However, my philosophy is that only the paranoid survive.
2. AI, IA, and RPA continue to converge, creating automated copywriting solutions that leave humans in the dust.
Expect new AI copywriting software to keep getting more affordable and sophisticated.
For businesses, this means profit.
For copywriters, this means something else…
Yes, AI that writes as well as humans seems like science fiction.
The data barrier could still prevent SMBs from exploiting platforms like Persado.
But using GPT2 simply to automate copywriting, they could still win.
Imagine…
- You feed GPT3 an ad
- The app spits out a couple dozen ad variants for a buck a piece
- You split test all of these variants against each other
- Choose the winner
- Make money
- Rinse and repeat
Why even use a copywriter in this case?
Simply by combining a fake text generator with split testing, marketers could avoid copywriter fees entirely. That savings could then be rolled into ad tests, building up a data pile that could inform future marketing campaigns.
Such possibilities should be sobering to every copywriter who’s thinking ahead.
If copywriting software automates 20%, 50%, or 80% of copywriting work in the next 10-20 years, who will survive?
Basement-dwelling lit majors who need some extra cash…?
Overseas competitors whose native language isn’t English…?
Low-skilled staff lounging around the front desk at Foot Locker complaining about their low wages…?
Obviously not.
In my opinion, the biggest competition for copywriters (excluding the AI software itself) will be “augmented marketers” who are tech-savvy, skilled, and hard-working.
If automation is as great a threat as Kai-Fu Lee says in AI Superpowers, then upskilling is a matter of survival.