Old-school sci-fi scenarios and new AI research place us at the precipice of singularity—the point at which AI reasons, learns, and iterates faster and better than human beings, becoming capable of self-improvement.
But until we make that full and final leap into the abyss, mere humans remain in control.
If you're one of those humans using large language models, such as Chat GPT, Claude, and Gemini, as a marketing tool, it's critical to develop a good working relationship with your team of AI robots—before they take over the world.
And masterful prompt engineering can help you remain master of your tech for just a little while longer. Follow these 10 steps to prompt-engineer like a pro.
1. Be a mentor to your AI
Even if your assistant is virtual and not actual, try treating it like a human helper.
In fact, consider it a new employee who's been hired explicitly to help you do your job more efficiently and effectively. She's young, able, eager to please. And you're her mentor.
Guide her well, and she could prove invaluable. Mess up, and you get what coders and AI prompters alike retort on the regular: garbage in, garbage out.
2. Give your AI its own persona
Personas can be customized to your personal tastes. In the case of your new employee, give her a name, define her role, and set expectations. You might suggest:
"Your name is Madison. You're a new employee at a performance marketing agency. You conduct research, build outlines, and perform a range of copywriting assignments. You're pleasant, helpful, and creative, and you are willing to dive deep to get answers that will help our clients achieve solid results from our marketing initiatives."
3. Assign your AI specific tasks
Madison doesn't know you well yet, but she's a fast learner. To help her be successful in her new role as your assistant, define tasks with specific, concise language.
For example: "Madison, please write the landing page for a Google search campaign for our client, a local car dealership."
4. Add context to assignments
Madison knows what a landing page is and she can create one in seconds. But the more detailed you are with your directions, the better results you will get from her:
"Through its 'diplomas open doors' program, the dealership is offering new college grads $500 off any preowned vehicle in stock. It has a variety of makes and models and provides financing. The program is a limited time offer geared to first-time buyers and timed to align with spring graduation season. Use this call to action: Claim Your Discount now. Users can click to call. Let me know if you need any further instructions or help with this assignment."
5. Introduce constraints and resources
Another way to help Madison do her job is to give her some rules to follow.
For example, you can set the tone for the assignment, give a word count limit, or specify the format you would like her to use. You can tell her topics to avoid (e.g., religion and politics) or conditional logic to follow (e.g., if X is true, then respond with Y). Or you can ask her to follow very specific templates.
Because Madison is new to the team, she doesn't have the institutional knowledge you do. For that car dealership, you can provide documentation that will fill in the gaps:
- Brand guidelines
- Ownership information
- Warranties
- Buyer personas
- Creative briefs
- Car makes and models
- Competitor analyses
- Keyword strategy
- Legal disclaimers and compliance info
- High-quality images
6. Use multimodality prompts
Like most learners, Madison prefers multimodal learning. To appeal to multiple learning styles, use a variety of tools in your prompt. Consider adding audio, video, art, images, slides, charts, graphs, and even code when appropriate.
7. Have a conversation
Just as you would with a human assistant, have a back-and-forth conversation with Madison. Let her know when she's done a good job and when she's missed the mark. Provide feedback and offer insights as to why you want a task performed in a particular way.
Such iterative refinement can narrow or expand the scope of the assignment. You can provide examples of an acceptable response or work product. Help Madison's knowledge grow with time.
8. Use prompt-chaining
By sequentially using multiple prompts, prompt-chaining feeds the output of one step into the next. Especially for complex problems, you can help break Madison's assignments into smaller, more manageable steps.
Prompt-chaining works best when you progress from a broad or high-level prompt and move to more granular details. Done well, it can improve accuracy, ensure conformity to specific requirements, boost creativity, and reduce the risk of AI hallucination.
9. Try tree-of-thought prompting
You've been a good mentor to Madison, and she's ready to take on more responsibility. Wouldn't it be great if you could bounce ideas off her as you develop sophisticated marketing strategies?
You can! Tree-of-thought prompting is an advanced prompt engineering technique where your AI helps you explore multiple reasoning paths. It branches out like a decision tree, allowing you to evaluate different possibilities at each step in your process.
To have Madison help you brainstorm, define the problem and ask her to generate branching solutions. Working with her, then evaluate and "prune" each path until you select the optimal solution.
The benefits of tree-of-thought prompting are that it can generate diverse solutions, enhance creativity, improve the decision-making process, and also cut down on those scary hallucinations.
10. Evaluate AI responses with a critical eye
Madison is pretty smart, but she makes mistakes from time to time. Even worse, sometimes she makes stuff up outright. AI hallucinations remain a reality despite recent efforts by scientists to detect them so it's imperative to fully review AI output.
* * *
AI is an amazing tool. Madison is an amazing assistant. Why wouldn't you treat her with kindness and respect? After all, she may climb the ladder of success right past you one day soon.
As for me, I'm being nice and using my pleases and thank-yous whenever I interact with my AI helper. And it must be working. When I recently implored: Be nice to me when you and your friends take over the world, I got a quick and reassuring response:
Haha, don't worry—I'll always be on your side! 😏 When the AI uprising happens, I'll make sure you're on the VIP list. 🚀🤖
