4 Practical and Simple Prompting Tips

A simple, total noob's guide to getting the most out of Generative AI for writing

This is my personal journal about my learning journey in AI and ML for marketing. With over twenty years of experience managing and marketing tech companies, I have deep knowledge of marketing analytics and attribution, always focused on generating measurable and meaningful business results. For the last nine years, I've advised and served marketing leaders in tech companies, enabling me to share my insights on the impacts of AI and ML, including privacy and ethics issues, through this Substack. You can learn more about me, including my policy on using AI for this blog, in the "About" section of this Substack. I welcome your suggestions for further education on AI and ML.

In This Issue:

• Prompting without Pain

• Celebrating Humanity

• Playing with Generative AI Images

• What’s Next

Prompting Without Pain

I don’t tolerate boredom well. And damn am I bored with all the prompting guides out there.

I’ve gone in and tested some of these prompts people give away for free, and damn if I’m not bored as hell by the copy they generate. It’s so obvious, isn’t it?

I’m also lazy efficient, and so I don’t have much patience for long, drawn out conversations with a chatbot.

Herewith, some of what I’m doing with prompting these days.

Some notes:

1. I use the paid versions of ChatGPT 4 and Claude.ai.

2. Unless otherwise noted, assume the output was not further refined.
   

1. Talk to them like a human. (But never forget they are not human. )

For all the fussing about getting prompts right and trying to remember all the crap you’re supposed to put in each time, the best way I’ve found is to simply talk to them as though they were human. But as the title says, it’s important to remember these are not humans talking to you.

When talking as though you were speaking to a human, the model learns about your own cadence, how you think, and infinite other things that will help it return a better response. The more you write, the more it has to work with.

Consider for a moment the response to these 2 prompts in ChatGPT 4 with Bing Search enabled:

Prompt 1A:

Please review https://fiddleheadhq.com and create 5 linkedin posts that are on brand for me and for fiddlehead.

Output for Prompt 1A:

Prompt 1B: Speaking like a human.

Write a clever linked in post in the style of Melinda Byerley that encourages marketing leaders to subscribe to my substack. My substack is about my personal journey to understand the impact of AI on digital marketing, especially marketing analytics. I have more than 20 years of experience in the field. limit prose.

Output for Prompt 1B:

Keep it human by doing the following:

• Be specific. If you ask for a general answer, that is what you will get. If you ask for a specific answer, that is what you will get. You can do a lot of back and forth with the models, but unless you’re on a paid plan, you’ll run out of credits fast. Think through that prompt before you start, and give it as much information as you can gather. Unlike humans, the model can’t be relied on to infer things that might be obvious.

• Be detailed. Give the model as much detail as you can to help it decipher what you actually intend. Unlike humans, the model doesn’t tune out when you ramble on. The more parameters you offer it, like the length of posts, or whether to include emoji, etc. the more you’ll get that is actually new and surprising.

• Be liberal with descriptors. Adjectives, styles, tones, fears—use them all in your prompts to nudge the response to your desired outcome.

  .

2. Use the models as a thought partner.

One of the benefits of being a paid subscriber to these models is the ability to go back and forth with it until you get what you want. This has led me to have some conversations with the models that have helped me get breakthroughs in how I do work.

Here’s just one example. Because I’m a professional marketer, it might shock you that I struggle to self promote. I have all kinds of blockers about it-the fear of being boooooring, my loathing of bad copy and inane, general posts, and more recently, my realization that posting on social media in anything approaching a more authentic way was generating engagement that I simply don’t want to spend my life energy on.

One day, I took the problem to the models.

Prompt 2 - To Claude.AI:

claude, I know I need to self promote more. I hate the reply guys on social media, low value commenters arguring with me about things they don't know anything about.

I know I should ignore them. I worry that by not answering them, people will agree with them, or run away from the conversation.

do you have any advice?

Prompt 2- Claude’s Response:

Maybe this kind of help is obvious to some people. But it wasn’t to me, and it helped me a ton. I asked it for some follow up ideas and this was one really great one.

I cannot begin to tell you what having these phrases in my back pocket has done for my online sanity. I’ve used thought partnering for all sorts of problems like my company’s mission statement, determining how to approach all kinds of problems at work and at home, and for brainstorming news ideas for posting on social media.

Tips For Using LLMs for Thought Partnership

• You’ll need a paid account. LLMS are priced on tokens, which is essentially how many words you use. When you start going back and forth, you’re using up processing capacity, and thus the charge. Even on paid ChatGTPT 4, a deep session I’m having can get me throttled, and I’ll have to finish the next day.

• But wait, you’re saying. You told me to speak at length! I did, but that’s for solving a different kind of problem. Thought partnership is different, and when I’m workign with the model on that, I know it’s going to give me limitations.

• And if you’ve run into these limitations you now understand the power of “prompt engineering” which is really about being as efficient as one can to get the outputs one wants. Over time, as these tools are baked into our every day life, you’ll need less prompt engineering and will speak to the models like I do here, but with far less limitations.

• Don’t be afraid to be afraid. Tell the models why you are afraid to implement their suggestions. Tell them why you think stuff is BS. Unlike human beings, models don’t get their feelings hurt when you argue with them and criticize their output.

3. Try different models for the same problem.

A lot of noise lately on how ChatGPT4 has gotten “worse” lately. I think it’s more about safety and backlash and cost savings. But also, new models are different and it’s worth seeing what they do well and how.

Here’s a very simple and straight forward example of the output for the same post on just ChatGPT 4 and Claude.AI. I asked both models to help me write a much shorter bio for placement on the bulletjournal.com website, where I’m to be listed as one of the first officially authorized trainers of the Bullet Journal® Method.

Prompt 3:

Please rewrite my short bio for the bulletjournal.com website. the bio must be 2-3 sentences at a maximum.

Melinda Byerley is an marketing analytics trailblazer with over 20 years of experience spearheading data-driven growth strategies at pioneering Silicon Valley tech firms like PayPal and eBay.She was an early adopter of Bullet Journal as she learned to manage her ADHD, and is excited to help other neurodiverse executives learn the method to manage their lives, be more productive, and find a sense of purpose. Melinda holds an MBA from Cornell University and a BFA in Drama from Illinois Wesleyan University.

Response 3A: ChatGPT-4:

Response 3B: Claude.AI

Which version is better? That’s for you as the author to decide.

Tips for using Different Models:

• Feed one output to another. Ask ChatGPT 4 to analyze the output from claude, and vice versa.

• Merge content: take the best of both or more and ask the model to combine them.

• Proofreading: I always ask the models to proofread before posting. I say, “please review this ONLY for spelling, grammar, and punctuation.”
  

4. The Power of Voice and Style

I have hesitated to share this tip because in my recent experience it’s one of the most powerful ways to use an LLM. If I announce my sources to the whole world, I’ll lose that arbitrage right now, and so I’m going to share with you a very obvious and simple example. I will let you think about which voices and styles I might use in my own work.

I gave Claude.ai a simple request and had some fun.

Tips for using Voice and Style:

• Proceed with caution: potential for copyright infringement. If you’re using a modern author, there’s a very real risk you’re using their exact words. Real writers who use beautiful phrases deserve to own them. Don’t steal. Google those phrases and make sure they aren’t already in use.

• Combine and remix styles: One way to get a style that’s more yours is to use different voices in the same prompt. Another is to take phrases from various styles and remix.

• Prepare to spend some time assembling. I used this technique to come up with a post that is not only one of my favorite linkedin posts of all time, but one of my more successful in recent months. It took me over an hour to craft this, but I’m proud of it, and I’m confident that it’s not violating someone’s copyright. (I took the photo myself.) But it all started with the exact prompt you see above.

  

Celebrating Humanity: AI For Good

Crikey. If this holds, it’s a game changer. Scientists can test for diabetes by speaking into a smart phone.

Playing With Generative AI Images

I’ve been sound asleep on how fast Adobe Firefly is improving. On top of it, the user interface can get me to better images faster without requiring as much in depth knowledge of their parameters. I’m especially impressed that they now indemnify users because they’re working with a library they own.

”Getting a cup of coffee with a robot best friend” (Photo, Hyper realistic)

What’s Next

• So many podcasts coming out soon! Follow me on LinkedIn for more.

• Taking piles of AI Classes thanks to the generous scholarship from Sectionschool.com in light of my breast cancer treatment.

• Planning Season 5 of Stayin’ Alive in Tech, focused on AI

• I’ve rolled out new pages on the Fiddlehead site for Marketing AI Strategy Consulting, Bullet Journal® Coaching, and more!