How to Overcome Limitations of Prompt Templates

Prompting helps but it’s not enough to keep GenAI on track. If you’re building roleplay sims or coaching scenarios, you need guardrails that live beyond the prompt. In my first roleplay sim, I didn’t just ask Gemini to follow a structure. I designed the system to make sure it did. That’s how I prevented: ❌ Topic drift ❌ Repeating the same question ❌ The AI “taking over” the conversation Instead of trusting the AI to follow directions, I used code to manage the flow. Example: if conversation_step == 1: conversation_step += 1 return "How do we mitigate this risk to ensure..." Even though the AI got the learner’s input, I didn’t use its reply. I used a hardcoded one to stay on track. That’s the difference: The prompt helped with tone and context The code enforced sequence and structure The design decided when GenAI should contribute (and when it shouldn’t) If you’re using GenAI for simulated conversations, prompting alone isn't guaranteed to prevent chaos. Your system has to. How are you building in real guardrails in GenAI powered learning experiences? #InstructionalDesign #LearningDesign #eLearning #WorkingOutLoud #EdTech #DigitalLearning #AIInLearning

The longer your ChatGPT conversations go, the better your outcome will be. 11 months ago, I thought prompt templates would be a panacea for shortcutting long processes into small repeatable copy and pastes but this approach yielded lots of basic, half-baked output. Template limitations become evident with complex tasks, such as long-form video script generation, where it lacks depth and nuance. Better use of ChatGPT involves engaging in iterative dialogues and chaining thoughts together in a sequence, treating it as a collaborative partner to progressively refine and expand content. So, steps: Research - seed the chat with context and info. If you don't have it, you can ask ChatGPT to source it. Organization - have it outline or sequence your content in whatever logic you need. Format - ask it to spit out the content in the format and style you need it in. Expand / Refine - ask it questions to build, expand, or trim specific aspects. This process allows for more context accumulation, like rolling a snowball downhill, enabling more insightful and tailored responses and ultimately better deliverables. Link to example in the comments. 👇

Even if your AI prompts are performing well, you should still keep iterating and improving on them as new prompting research emerges and you gain more prompting expertise. I got a great reminder of this after a live training session a couple of weeks ago. Scott Brown emailed me the next day to share an experiment he did. He applied two prompting best practices he learned in the live training to MY long-standing voice training prompt I shared with the group. This prompt analyzes your writing to understand its unique voice, style, and tone and then writes a prompt snippet to help AI emulate this style when it writes the same kind of copy for you. I started iterating on this prompt about a year and a half ago. It's been improved several times, but I have only thought to improve it when it stops performing due to what's called prompt drift -- when the AI stops responding as you expect due to model updates. (In my mind, I think I was thinking "if it's not broke, you don't need to fix it). Scott's improvements (based on what he learned in our training) included adding: 👉 Instructing the AI to take its time and think step by step 👉 Embedding an emotional motivation to stress the importance of the task These additions significantly improved the prompt's performance, even though the original prompt had served me (and my clients!) well for a long time. Scott's initiative highlighted a blind spot I had around remembering to update my own prompts even when they work! I did a bunch of A-B testing adding my own prompt updates, and oh WOW, it made a gigantic difference. The version I use has definitely now been updated with these two elements! (If you're enrolled in my Foundations of Generative AI for B2B Marketing course, the prompt document in the "Teaching the AI to Write Like You" chapter has been updated, too!) Scott caught my own blind spot on ALWAYS going back and updating old prompts that work with new prompt understanding. This experience definitely underscores the importance of revisiting and improving even your good prompts regularly. I love when the student becomes the teacher. I love that he learned these strategies from me and then applied them immediately! If you have a copy of the voice training prompt and are not taking my course (where you can just download the updated prompt), I highly recommend you consider adding these two prompt elements to significantly enhance it's performance: ✔️Instruct the AI to take a deep breath and think step by step. ✔️Provide emotional motivation to do a great job on the task. Thanks again for the reminder Scott! ---- Want to learn more about how to get the most out of generative AI for Marketing? Definitely hit that + Follow button because I post a ton of content like this on here! And if you want to go really deep, there's a 🔗 to my Foundations of Generative AI for B2B Marketing Course in my bio.

Prompting tips from someone that spends probably $13k+ per month on OpenAI API calls. I'll break the tips into chatGPT user interface tips as well as API tips. My bias is of course going to be about outbound sales and cold email because this is where we spend from and 100% of this spend is on 4o mini API calls. Chat GPT Prompting Tips 1. Use transcription as much as possible. Straight in the UI or use whisprflow(dot)ai (can't tag them for some reason). I personally get frustrated with a prompt when I'm typing it out vs. talking and can add so much more detail. 2. Got this one from Yash Tekriwal 🤔 - When you're working on something complex like a deep research request or something you want o3 to run or analyzing a lot of data, ask chatgpt to give you any follow up questions it might have before it runs fully. Helps you increase your prompt accuracy like crazy. 3. I've found that o3 is pretty good at building simple automations in make as well so we will ask it to output what we want in a format that we can input into make and often we can build automations just by explaining what we need and then plugging in our logins in Make. API prompting tips 1. Throwing back to the Chat GPT UI, but we will often create our complex prompts in the user interface first and then bring it into the API via Clay asking ChatGPT along the way on how to improve the prompt and help us think of edge cases. This can take any team member to a prompting pro immediately. 2. Examples are your best friend. Giving examples of what you would want the output to be is how we can get our outputs to be the same format and not put "synergies" in every email we are sending. I tell the team, minimum 2 examples for single line outputs. 4 examples for anything more complex than that. 6 examples for industry tagging because that gets so odd. Save on costs by putting some real examples in your system prompt. 3. Request the output in JSON. It keeps everything more uniform in the format you need. 4. Speaking of JSON, ask the API to prove to you why it thinks what it thinks and then output the answer. Especially for company category tagging, I find this works really well. I see this greatly increase the accuracy of our results for 2 reasons. I think if AI has to take the extra second to prove to you why a company is an ecommerce brand, the results are demonstrably better. This is just a guess, but I also think that because LLMs basically work by guessing what the next best word is, if you have it tell you why it thinks something is a certain industry and then it gives the output, I think it's much more likely to be correct. Anything else you've found?

On Friday we had our first fully hands-on class for Day 5 of 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐯𝐞 𝐀𝐈 𝐟𝐨𝐫 𝐌𝐚𝐧𝐚𝐠𝐞𝐫𝐬: 𝘏𝘢𝘯𝘥𝘴-𝘖𝘯 𝘗𝘳𝘰𝘮𝘱𝘵 𝘌𝘯𝘨𝘪𝘯𝘦𝘦𝘳𝘪𝘯𝘨 During the class students got their feet wet over the course of 3 modules: 1. 𝐏𝐫𝐨𝐦𝐩𝐭 𝐢𝐧𝐣𝐞𝐜𝐭𝐢𝐨𝐧𝐬: how your commercial chatbot will get fooled 2. 𝐏𝐫𝐨𝐦𝐩𝐭 𝐭𝐨𝐨𝐥𝐬: how to analyze financial statements in under a minute 3. 𝐏𝐫𝐨𝐦𝐩𝐭 𝐭𝐮𝐭𝐨𝐫𝐢𝐧𝐠: how to use prompting to learn new skills quickly In Module 2, many students had trouble getting ChatGPT to analyze Apple's 10-K by calculating iPhone growth and overall financial health of the company. The AI would often just say "I can't complete this request". This happened whenever the student asked for everything in one go. 🔨 One fix applied for this was to break the ask down into baby steps for the AI: • First just summarize the 10-K to ensure ChatGPT processed it correctly and has the information saved into the context window (or RAG pipeline if used) • Then ask for a simple calculation of iPhone CAGR • Finally end with a full F-Score analysis to assess Apple's financial position (but even this we broke the 9 components of F-Scores into 3 batches of 3). 💡 As you get more sophisticated in prompting (and more realistically when the LLMs continue to improve), this analysis will become more streamlined, but when starting out, it's often helpful to treat the AI as you would a new hire: by giving piecemeal tasks gradually instead of overloading them with a mountain of responsibilities. Just like a new hire, LLMs that are overloaded with tasks often give up or fail (see attached video) ⤵ Any prompt experts out there: what are your favorite prompts? Comment below!

Building with AI = Failing A Lot Take writing custom GPT instructions. I've probably written 50+ at this point and for the most part they have all followed a similar markdown pattern I learned over a year ago from Rachel Woods. But models evolve and what worked yesterday might not work as well today. When GPT-4o came out a few weeks ago, I definitely noticed some of my custom GPTs acting weird. Some got better. Some got worse. Some straight up broke. So I went back to the lab again and started to rebuild. Lots of failures later, I now have a new custom GPT prompt structure that still has the bones of my older markdown method, but incorporates a lot of OpenAI's recent guidelines. And now I have GPTs performing better than ever. You can check out the full article, but here are some good guidelines for any prompt writing: ✔ Simplify Complex Instructions -break larger steps down ✔Structure for clarity - use delimiters and examples ✔Promote Attention to Detail - encourage the model to focus on certain areas of the prompt ✔Avoid Negative Instructions - frame instructions positively ✔Granular Steps - break down steps as granularly as possible ✔Consistency and Clarity - be explicit with terms and be sure to define what you want with examples.

⚠️ Every untested prompt is a potential bottleneck in your AI's performance! Behind every prompt lies a system of expectations that must be tested, measured, and refined like any other engineering artifact. I've been watching teams discover that the same prompt can trigger spam filters for enterprise clients while working perfectly for startups. I have 57 other examples (couple of them captured below)! The difference between a prompt that scales and one that fails isn't creativity - it's systematic evaluation. The most successful AI implementations I'm seeing now treat prompt engineering like software development: test, measure, version control, deploy and thats what we have streamlined and automated Future AGI by integrating our evals right into the prompt workbench. Here is how it is helping teams of all sizes-: 1. Parallel Testing at Scale- While we're all running A/B tests one at a time, this runs thousands of prompt variations simultaneously. Think: 38% accuracy jumping to 75% - but now you know exactly which prompt got you there. 2. Beyond Gut Feelings- Custom metrics that actually matter. Not just "does it work?" but "how does it perform against YOUR specific success criteria?" 3. Side-by-Side Reality Check- Every variation laid out visually. No more spreadsheet hell or manual tracking. The winning patterns become obvious. 4. Production-Ready Deployment- Version control built-in. Test → Validate → Ship. One commit at a time. As quoted by GPT- you can’t build a skyscraper with LEGO instructions and so is the case with your AI, you cant build it on fragile prompts. What are your biggest prompt engineering challenges? #AIEngineering #PromptEngineering #ProductDevelopment #GenAI #MachineLearning