In this post, I’ll take you step-by-step through an example of how to use the best prompts—the specific instructions you give an artificial intelligence (AI) chatbot—to yield a framework you can actually use to help organize and compose a blog post. You could create this framework without leaning on AI, but significantly, the bot can help you do it in record time.
First, a quick rewind: In my last post about artificial intelligence, I looked at the compelling reasons AI in the guise of a chatbot—for example, Open AI’s ChatGPT or Google’s Bard—is destined to fail as a wholesale replacement for the human copywriter, because it lacks the emotional intelligence to speak to human users. But I conceded the chatbot will prove its mettle as a tool to make the writer’s job more efficient—in short, it’ll act in the capacity of assistant to the copywriter to help us hand our clients the same high-quality deliverables we always do, only quicker.
Before we strap on our goggles and dive in, a few reminders. While we can agree an AI chatbot is useful at
- brainstorming content ideas
- analyzing keywords
- creating outlines
- generating FAQs
- and doing a bit of rudimentary research
—and all of this at warp speed—we also must understand the inherent risks of using AI-generated results, including
- generating duplicate copy potentially marred by inaccuracies and misinformation
- failing to write with the nuance that makes good copy so meaningful and engaging for a human to read
- to say nothing of treading into muddy legal areas
so we must proceed with due caution.
In short, we’ll lean on the bot to enhance what we do, not replace it.
Now let’s roll up our proverbial sleeves and try out one of these new-fangled machines.
The Art and Craft of Using Chatbot Prompts
If you’ve never used a chatbot, set up an account or two and have a little fun before you start dropping in real-world prompts, to see how the tool works. Ask the bot to explain in a short paragraph, for example, how to bake a chicken, but in the voice of Yoda. Ask it to define ‘algebra’ for you like you’re five, or copy and paste some existing content into the prompt field and ask the bot to rewrite it in the style of a famous author or celebrity. These silly little ice breakers will help introduce you to your new writing assistant and show you what it is (and is not) capable of.
After you’ve gotten more comfortable, approach the bot like you’re having an actual conversation. As you go, build upon your prompts to generate more refined answers. What the chatbot lacks in three dimensions, it more than makes up for in its powers of retention, so it’ll remember what came before in your thread.
The Chatbot Experiment: Create the Framework for a Blog Post
1. Make a List of Resources
First I’ll ask Open AI’s ChatGPT and Google’s Bard to create a list of “ten authoritative online resources” I can use during the research phase of writing my blog post, which will be all about ‘tips for taking a family vacation to the beach’ (and those will be my target keywords). I’m including the term authoritative in my request in the hopes this will filter out amateurish travel blogs and other websites of questionable provenance and reliability.
First, the results from ChatGPT:
And here is what Bard gave me:
There is only a little overlap in these two lists, and all in all they're good results; the bots served up some dependable resource materials I can chew on, lickety split; note that ChatGPT also reminded me to cross-reference the information for accuracy. Whenever I’m writing a post for a client, the research benchmark I typically use for accuracy is finding the same or similar claims from at least three authoritative online resources, so those lists gave me plenty to get started.
2. Build Upon the Resources List To Generate Ten Tips
Now I’ll ask the bots to generate 10 travel tips for planning a family beach vacation, based on what I’d find if I looked at the resources they generated.
From ChatGPT:
And from Bard:
Again, there is overlap, but I can blend these two lists with minimal effort, capturing the best points in each of them to create the strongest list for my blog post; note that Bard was kind enough to generate some bonus tips for me.
3. Ask the Bots To Organize the Tips into Optimized Sections
Now I’ll ask my little bot friends to get down to business and organize this topic more tightly. Here was my specific prompt: “Can you please organize these ten tips into a few meaningful sections, and then supply an SEO header for each section?”
From ChatGPT:
From Bard:
For this more demanding task, ChatGPT clearly outperformed Bard, although notably absent is the modifier ‘beach,’ easy enough for me to work in. So I mentioned to Bard what it generated struck me as “pretty thin”—as if we were having a conversation—and asked for a few more details in each section. This time I got a much better result:
I’d jettison some of these bullet points and reassign others. But between what each of these chatbots generated, now I feel like I have plenty to organize into an outline to suit me.
4. Generate an Optimized Meta Description
Before I start my composition,I’ll ask each bot to write for me an optimized meta description that will also serve as the first sentence of my blog post (because I know there’s a good chance Google will pull the description from there to begin with). I used this language as a prompt:
“Can you please generate a sentence of 155 or fewer characters that will serve as both the opening sentence of my blog post and as an optimized meta description?”
ChatGPT answered with this:
“Embark on a memorable family adventure with these expert tips, ensuring a stress-free and enjoyable vacation for all ages."
And Bard with this:
“Plan your perfect family beach vacation with these 10 expert tips, from choosing the right destination to packing for success.”
I’m not in love with either of these, but Bard came closer to what I had in mind.
I went back to ChatGPT and refined my prompt:
“Can you please try again, but this time use more imaginative and engaging language?”
And it gave me this:
"Unleash your inner explorer with our secrets to crafting unforgettable family vacations – where every moment becomes a cherished memory."
That one has a bit too many whorls and flourishes for my tastes, but you can see how adjusting the prompts can get you a bit closer to the mark. In each version ChatGPT generated for me, the modifier keyword ‘beach’ is still missing; Bard emerges as the winner again.
5. Wrap Up With Optimized Web Page and Post Titles
Finally, I’ll ask for a web page title of 50 or fewer characters, and a search engine-optimized title for the blog post.
From ChatGPT:
Page title, “Family Vacation Guide,” and post title, “Ultimate Family Vacation Planning Guide: Memorable Adventures for All”
From Bard:
Page title, “Family Beach Vacation Tips: 10 Expert Tips for a Fun and Safe Trip,” and post title, “Top 10 Family Beach Vacation Tips for 2023”
Bard supplied character counts for each (although I didn’t stipulate a count for the post title), of 47 and 49 characters respectively, but got the first one dead wrong—the actual character count is 66. Despite this misstep, I prefer Bard’s results to ChatGPT’s for this piece of the exercise. I’d probably just go with “Top 10 Family Beach Vacation Tips for 2023” to stand in for both titles, and leave it at that—the parity between the web page title and the post title is nice, and the SEO is solid.
Now I’m ready to crack my knuckles and dive into my blog post.
Chatbot Prompting Takeaways
- In my early experience using a chatbot to generate SEO queries or outlines, a recurring theme seems to be repetition. In the exercise above, I see several instances of repetitive or redundant language I’d leave out of one or more sections.
- It’s helpful to have a “conversation,” as the chatbot typically remembers everything in the thread. (ChatGPT inexplicably sidestepped my quest for ‘beachy’ answers after the first results). You can also paste blocks of copy into the thread and ask it for an answer to a question based on what it finds in the block, but be advised ChatGPT can’t analyze a ‘token’ count—one character, or one word—greater than 4096 in the current free version. Bard insists its token count is unlimited, although users have challenged this assertion.
- I’ve also found that, in spite of referencing a specific character count in a prompt, the algorithm still sometimes gets it wrong, as we saw in Bard’s page title generation above.
You Can’t Always Get What You Want
But if you try sometimes, you might get close to what you need. Keep in mind, chatbots use inputs from you and me to learn, so when you ‘feed’ them new requests, they’re continually modifying how they respond based on what you asked them, albeit not in real time. So the answers these bots generated for this exercise today, for example, ostensibly won’t be precisely the same as what they’d give me, say, a month from now. As pioneers in this brave new chatbot world, we’re likely to see better and better results from the algorithms the more sophisticated they become and the more we refine our prompts. Even at this relatively early stage, we can still harness the power of AI to streamline our work.
Eight Oh Two is a proven leader in search engine and performance marketing focused on driving superior results by leveraging best-in-class technology with the very best in human intelligence. Contact us today to learn how we use the tools at our disposal to compose engaging, top-shelf content that drives traffic to your website.