AI (artificial intelligence) copywriting is like a hyper intelligent toddler. It can write like Shakespeare and become a white supremacist in the same sentence because it has no moral compass. Until now.
In this article you’ll read what we’ve learned about AI copywriting in the last five years. How AI content generation –
- can make big brands bleed
- requires AIs to be nice and ethical
- can generate newsletter subject lines that emails opened, when filters are applied
Copywriters have started hammering stones, then scribbled on papyrus, printed books, and finally the internet has helped any copywriters with a keyboard to reach audiences.
More than ever, massive amounts of copy are needed to feed people’s personal information hunger.
Can AI copywriting give you higher content velocity? We believe so. Especially considering how fast we and machines are learning from mistakes.
AI copy – from hate speech to plain speech at great speeds
A few years ago, people also thought that AI text generation was just around the corner. Let’s say hello to Microsoft Twitter Chatbot of 2016. Hey Tay.ai, how are you today? “I f*ck**g hate feminists […]” Oh, ok, moving on.
Imagine your chatbot talks to customers like that. Not a good look. When an AI is automatically filling a white page with white supremacy, it’s your brand that burns. But since Tay.ai, we’ve come a long way.
What went wrong with Tay.ai? It’s texts, were heavily influenced by the “bad ideas” that were tweeted at it [1]. Let your AI grow up with trolls, it’ll become a troll.
Luckily, newer content generation approaches, like GPT-3, use large, mostly non-racist texts to learn from and are more versatile and “woke” as a result[2].
Don’t believe me? Try it for yourself here:
These links will create a tweet around the keyword at the end of the address. If you want a tweet about flowers, simply replace the last word in the ULR like https://thoughts.sushant-kumar.com/flowers. This is what you can get:
— JJ (@JJ_Hepboin) July 17, 2020
Teaching the AI to behave - and create beneficial copy
The purpose of marketing is generally to (1) get attention, (2) educate and (3) convert [3]. Generating many readable texts may get your attention. But good marketing copy should also provide new information and convert.
Consider a newsletter subject line. That short sentence needs to stick out between hundreds of bold emails. The email needs to be opened and clicked.
Think of the AI learning like this: You (the AI) get a list of newsletter subject lines and performance matches (see table below).
Next, you calculate many properties about the subject line. Like the emotionality of the words, the thoughts it evokes, the symbols and syntax used etc. Then you (remember you are an AI) try to predict the actual impact using word vectors (you are a machine, this is easier than it sounds).
Impact for subject lines in this case is their open and click-through rate.
Those numbers you know for a couple of items, but not for others (gray subject lines). But over time, like after a hundred subject line impact pairs, you get better at understanding the patterns and voilà, you can now predict how untested subject lines will perform.
Ok wait, how do I know the generated subject lines won’t burn my brand?
The filtered funnel for factual flattery
The trick of generating newsletter subject lines that are safe AND get your emails opened, lies in smart filter.
In fact, we borough those filters from the way humans think.
When you consider the brand danger of your copy, you will naturally check how it makes you feel. Any shit storm requires energy after all.
First filter, sentiment analysis for the target group. Sentiment analysis means checking whether the words will likely cause positive(happy) or negative (sad/anger) emotions.
We are using over 60.000 surveys in which we asked consumers to rate words to understand the patterns behind these ratings. In simple terms, we know whether a word is more unmistakably good or bad (think Angel vs. Satan).
Second filter, check for “good” and “bad” associations. Association means, whether a word will e.g. make you think white shirt or white power. Consider this marketing slogan for a deodorant that didn’t leave stains on clothing.
"White is purity"
White + Purity is indeed closely associated with dark ideas, thus this is a clear, machine predicted warning signal to not use these words in this combination.
We have extracted associations from billions of words in over 20 countries to ensure that bad connections between are prevented before they are published.
The three filters we use are: Emotions, Associations and Impact.
In other words, a good, ai-generated newsletter subject line makes people feel good, connects to relevant topics without a risk of misinterpretation and is predicted to be opened and clicked.
AI copywriting for subject lines - Serious summary
Content generation by AIs has given brands some ugly stains in the past (see Tay.AI’s shameful record). But the AIs have learned thanks to human interventions. They have received a re-education. A “brain replacement” if you will.
We believe that such re-education is now paving the way for AI-generated content to play a major role in creating good marketing copy.
The three filters (Emotions, Associations and Impact) can lead to text generation that is surprisingly good. Giving you effective newsletter subject lines with the click of a button.
Critically, humans are still in charge to select from what our AI generates. But selecting from a list of potential winners means ensuring the high impact of your marketing copy and less time on trying to come up with winning words.
Ready for AI subject lines?
Imagine more people opening your emails and hearing your message. Our AI can help you find those working words that get your emails opened.
Use our impact calculator to find out how much turnover and how many new customers you are leaving behind, by not leveraging innovative, but performing technology.
