Al on A.I.
A.I. (ARTIFICIAL INTELLIGENCE) continues to be a hot topic in the world of marketing as well as in the world at large. In this installment of the Mint blog, Co-Founder Al Navarro shares how he uses this ever-evolving technology.
It’s no secret that I was not a big fan of A.I. until a short time ago.
Here’s why: As recently as a year or so back, it wasn’t very good. In fact, based on 2 of 5 my experience with it, the technology wrote at about the level of a junior copywriter.
This might sound great to you, but put yourself in my shoes. In my nearly 35 years of experience working at ad agencies of all sizes, junior copywriters get it right the first time about 50% of the time. Creative directors and departments bake this factor into schedules, staffing, etc. And it’s all part of staff development. The juniors share their work with more experienced writers and creative directors and get tips on what’s good/bad/how to improve.
But at a boutique agency like Mint, a 50% hit rate doesn’t really help. It means I have to invest as much time in revising the copy as I would if I had written the piece from scratch.
Additionally, I could nearly always tell when a piece was written by artificial intelligence…there were just specific formulaic structures and phrases that it would use. (Believe it or not, I once caught a freelance writer submitting work that had been A.I.-generated…I found out because it wasn’t very good and when I used A.I. to draft a replacement, IT GENERATED A NEAR-IDENTICAL PIECE!)
I’ve also seen A.I. make factual errors and flat-out make stuff up. It’s probably why ChatGPT has the following footer: “ChatGPT can make mistakes. Check important info.”
However, I’d say that ChatGPT has gotten a lot better over the past few months. A lot closer to the mark the first time out. And that’s been a real relief.
How I Use A.I.
First, I want to say that I don’t use ChatGPT to write headlines or taglines. Not even as thought-starters. Call it personal/professional pride. I also feel like, “that’s what clients are paying us for”…if they could write a great headline via A.I., what value are we bringing? We even have a client in the tech space that prohibits using A.I. for their creative work.
The A.I. tool I’ve used most consistently and for the longest time is the Grammarly writing assistant. Essentially, it’s a proofreader/grammar checker. Grammarly has become a staple of written communication, with more than 40 million users and 96% of the Fortune 500. I was an early adopter of the paid version of the program.
But even Grammarly gets it wrong sometimes. For example, for one client, I often use a closing phrase that goes something like this: “If you have questions, reach out to your [CLIENT NAME] contact.” And Grammarly always insists on changing “reach out to” to “contact”. Which would result in the phrase reading “contact your [CLIENT NAME] contact.” Which is, of course, bad writing. So, you can’t always just accept what Grammarly recommends.
When I do use ChatGPT to write, it’s for a very specific application. We have a client for whom we draft two blog posts monthly, mainly for SEO purposes. To help with SEO, the posts should be at least 750 words long. Now, I don’t know about other writers, but writing 1,500 words each month on the same topic month after month can burn me out fairly quickly.
With the client’s approval, we started using ChatGPT to write these blogs. As I’ve already mentioned, it was very hit-and-miss in the beginning. This was frustrating because I had to go through each piece with a fine-toothed comb, often taking more time to edit and revise than if I had written the piece myself.
But I’ve noticed something in recent months. And that’s that ChatGPT has gotten better. I’m ready to ditch the “junior” from its title. Things get closer to where they need to be the first time. I’ve also noticed that it takes direction better.
Al Navarro: Prompt Engineer
To ensure that my extended ramble is useful to you, I will share the prompt I use for the blog posts I mentioned above. Hopefully, these tips will help you the next time you use A.I.
Okay, so here is the prompt I used to A.I.-generate the latest blog (some info has been redacted to maintain client confidentiality):
“You are an expert copywriter. Write a 750-word blog post entitled [USE A TITLE THAT CONVEYS TOPIC] for a general audience. SEO optimize for words like [INSERT KEYWORDS]. Try not to use too many lists/bullet points.”
Let me break this down a little so you understand why the above prompt works better than a simple “Write a blog post on [TOPIC].”
- “You are an expert copywriter” I got this tip from a webinar on A.I., and I think it helps. A.I. can write to various levels of sophistication, so you should specify what level you want.
- “750-word [blog post]” One big difference between marketing writing and other writing is that, generally speaking, the copy in marketing writing has to be a specific length to fit the design/media. For example, a 20-word headline is not considered a “best practice” for a billboard. Again, for my needs, the content has to be at least 750 words. I will say that, until recently, the copy would often fall short even with the count included in the prompt.
- “For a general audience” Specifying the audience helps the A.I. engine tailor its writing. For example, if the audience was more expert (versus a general audience), then you’d expect the content to be more in-depth and sophisticated.
- “Try not to use too many list/bullet points” For the client and topics in question, “listicles” (lists/articles) can make sense. But I noticed — especially in earlier generations of the ChatGPT tool — that it would frequently default to lists/bullet points. And while lists and bullets can help with readability, too many make copy just as challenging to read. So add direction in your prompts as needed.
Anyway, that’s a lot for you to digest. I promise that this wasn’t written by A.I. If you have any “Working with A.I.” questions or tips of your own to share, feel free to send them to me at al@mintadv.com.