I will continue to say this: You can’t use Ai to perform tasks you don’t understand.
If you want to ask Ai to create financial projections, you need to know what information you need to provide to ChatGPT to project. Say, you want to project your next year of revenue based on the last 6 months of trends via client churn/acquisition and average retainer by client, reported monthly. Will you know how to double-check that the GPT output is correct? Will you able to look at the methodology it used as the logic behind the work? Will you know how specific you need to be in the initial prompt itself?
Or, you need to know what Google Ad headlines to choose, modify, or polish that Gemini spits out for the product page you prompted it to use as a reference. Will the client like the headlines? Are the headlines exciting, witty, and accurate? Or are they robotic, irrelevant to the customer’s needs, or repetitive?
If you have a welding torch, but don’t know how to weld or what to weld, it’s not very useful, and you can do a lot of damage with it.
I’m using Ai a lot these days, for everything from financial projections to writing case studies, to writing ads for products.
For all of these, I need to have the theory of what I want to ask, and to analyze the output to confirm if the LLM results are useful and accurate.
What I find the most, is that a lot of these things are a bit flat. I rarely get a good outcome from just one prompt (including the image for this post, which took 8 prompt revisions, plus manual touch-up in Canva).
It’s like working with an intern or junior employee, they know enough to be dangerous, but can be really valuable with the right direction and oversight, but sometimes the time needed for that is not as efficient as just doing it yourself, or in this case, manually.
Here are my quick tips for writing ad copy with Gemini or ChatGPT:
Prompt your robot assistant with the product page URL, and say something like, “Write me ad copy for a Performance Max campaign based on this url: www.website.com/product-page-1.”
Then, I’ll take the URL of the company’s About Page, and prompt with something like, “Incorporate the language and tone of this URL into the ad copy.”
Then, I’ll iterate with a couple of tone prompts. For instance, if I’m selling fresh juice for entertaining for women over 60 from the East Coast, I might write a prompt that says, “Write this in the voice of Martha Stewart.” If I don’t like the output from that, “Write this in the voice of the Barefoot Contessa.”
After a few of these, I might get 3 headlines from each prompt, but will still need some other headlines, possibly around and angle/hook unique to the product, such as: “Write some headlines around how fresh this product is, with no preservatives or pasteurization.”
Out of all these prompts, I will usually get around 60% of the copywriting needed to start polishing things out. From here, I find that what I need to add is some kind of humanity and subtly to the writing. Headlines that hint at a problem or solution that it solves, then allows someone reading the ad to infer what it is that we are selling, or what the magic of it is. It’s like when people see a movie vs reading the book it’s based on - when someone imagines the story for themselves, vs the literal/visual interpretation of the material, they put their own dreams and expectations on top of the material, creating their own memories of the product. People remember clever ads like this that make them feel or imagine things, much more than something super literal like, “Buy this juice, because you need juice.”
This is what I’ve been experiencing the most with Ai these days: it is lacking the subtly that makes good copy writing good. It’s just trying to predict what comes next, not what a consumer is hoping and dreaming and aspiring to be - and how these products can help them have the life they envision for themselves.
Because robots don’t dream, they compute. They see what come next, but don’t feel what comes next. They are accountants, not artists, and nobody asks an accountant what to dream about.