I think I am Sean’s back-up brain -Holly
Here we go with part 4 of our AMA STL conference wrap-up!
Today we are discussing Mike Allton’s talk about AI, and how it can be a help, not a hinderance!
If you would like to be apart of the St. Louis chapter of the AMA please check out: https://amasaintlouis.org/
This month is donate life month. If you are interesting in becoming an organ donor, or you would just like to know more, please visit https://donatelife.net/.
SOURCES Mike Allton – https://theaihat.com/
The Marketing Gateway is a weekly podcast hosted by Sean in St. Louis (Sean J. Jordan, President of https://www.researchplan.com/) and featuring guests from the St. Louis area and beyond.
Every week, Sean shares insights about the world of marketing and speaks to people who are working in various marketing roles – creative agencies, brand managers, MarCom professionals, PR pros, business owners, academics, entrepreneurs, researchers and more!
The goal of The Marketing Gateway is simple – we want to build a connection between all of our marketing mentors in the Midwest and learn from one another! And the best way to learn is to listen.
And the next best way is to share!
For more episodes: https://www.themarketinggateway.com
Copyright 2025, The Research & Planning Group, Inc.
TRANSCRIPT:
Earlier this month, we attended the annual conference for the American Marketing Association St. Louis at Webster University. It was an amazing well-run conference with some great speakers – and today, I’m going to share some of the key takeaways from a presentation by Mike Allton from The AI Hat as well as my thoughts!
Now, I know how weary we all are about talking about AI in 2026. In my conversations with many people, I hear three common attitudes expressed:
- “I use AI every day, Sean! Don’t you? Chat, can you pull up 10 reasons why Sean ought to be using AI so I can email them to him? Sorry for that extra step, Sean – I’d have one of my agents do it, but it’s busy on some other tasks.”
- “I think the technology is neat, but it’s not something I use very often because it makes so many mistakes and I can’t trust it, and the last thing I want to do is babysit a computer program” – This is my attitude, by the way
- “AI? You mean that technology built on the theft of humanity’s intellectual capital that’s creating an environmental disaster and allowing plutocrats to steal jobs?” – That third attitude isn’t unreasonable, by the way. But said in a public space, it’s the sort of thing that tends to make you seem like you’re also looking at pocket calculators suspiciously for fear they’re going to put math teachers out of work.
So, look. Technology is a tool, and tools are meant to help us. Some tools are really useless. Others are so useful they can help us cut a lot of steps out of work that used to be really labor-intensive. 3D printers and CNC routers, for example, allow people who want to make things to be able to do it a lot faster with a higher degree of precision than they every could have by hand.
But there are also tools like hammers and chisels that have been around for millennia and yet we still use them today because they’re still useful for a lot of tasks.
So we have to think about AI as a potential tool to help us be better at what we want to do. And as Mike Allton discussed in his AMA Saint Louis Annual Conference presentation, part of the problem is that most of us haven’t yet learned how this tool actually works, and thus we’re not able to use it properly. But he argued that if we do learn how to use it, we can build a customized electronic brain to help out our business and sharpen our decision-making.
So, let’s see what Mike had to say as I recap his talk and share some of my own ideas!
I’m Sean in St. Louis, and this is the Marketing Gateway.
Before I get into Mike’s talk, let me offer an analogy that’ll be a useful framework for us: Microsoft Excel. When I was in my 20s, Excel absolutely mystified me and I couldn’t even begin to understand how it worked because no one had ever explained a spreadsheet or a database to me before. Microsoft Word and PowerPoint? I had those mastered. But ask me to create even the simplest formula in Excel? I’d spend more time complaining about how hard Excel was to use than it would have taken to learn how to do the task.
And Excel is hard to use, at least at first, because it’s really prone to throw up error messages that don’t make any sense. And look, as a kid, I was pretty skilled at using command-line DOS and writing things in BASIC and opening up hex editors to adjust values in my computer games. As I got older, I got better at using a computer and could do even more with one. It wasn’t my technology skills that kept me from being able to use Excel. It was that no one had taken the time to show me how the program worked and why it can be so useful.
So when I went to business school and finally had to learn how to use Excel as a tool for creating spreadsheets and reports… I actually loved it! Excel’s a tremendously useful program when you understand what it’s actually meant to do. Is it still frustrating and difficult at times? Absolutely. But it’s also really powerful if you use it right.
And I’m going to argue that the same is true with large language model AI tools that utilize a chatbot-style interface – and we’re just going to call that “AI” for short in the remainder of this episode. You know the platforms – Claude, ChatGPT, Gemini, Copilot, and dozens more just like them. They’re no longer new or novel, and as we find use cases for them, they can do some helpful and interesting things… provided we’re willing to learn how to use them.
So Mike Allton’s epiphany about how to do that came when he was on a road trip between St. Louis and California and he began conversing with a chatbot to try to figure out how this technology could be useful for business decision-making. He realized during the course of this conversation that many of the reasons people take a neutral or negative viewpoint towards this technology is because they don’t really understand how it works or how to correct its tendencies towards giving them output that isn’t correct or useful, and he decided to start a consulting business providing the training to fix this problem on the human end of things.
Now, I don’t want to get too deep into the nuts and bolts of this discussion because it made more sense in a 30-minute presentation than it might in a 15-minute podcast. But one of the things Mike talked about is the tendency of people to treat a chatbot as an advisor you go to for occasional needs rather than an assistant that’s working with you on your project. To use a different analogy, it’s like going to a co-worker whenever you need help on something you don’t understand instead of bringing them into your process and allowing them to help you with every step.
The problem with large language model AI, though, is that it is built on a foundation of information known as “training data” and then it’s further informed by instructions provided by the platform owner that tell it how to respond to user prompts. Each AI program has its own tone and approach towards output because of these instructions. But if you want the AI program to produce output that goes beyond the basics, you need to understand how AI platforms’ memory works.
See, each platform has a memory that’s measured in what are called tokens, and that means that the AI platform can only remember so far back into a conversation before it runs out of tokens and then can’t reference its older output anymore. All it can reference is what it has in memory or what it has in its training data or instructions.
So if you have a very long conversation with ChatGPT, it will start forgetting what it said in the beginning of the conversation and either begin repeating itself or coming up with new suggestions that don’t match its original output. Likewise, if you attempt to overload the chatbot with a bunch of knowledge using your initial prompt and then have a lengthy conversation with it, the chatbot will eventually stop responding to the initial prompt and start simply basing the conversation off what it’s been saying.
So, Mike Allton provided a list of the different kinds of memory a chatbot uses.
- Chat memory, which is what I’ve been describing up to this point – the chat itself, measured in tokens
- System memory, which is the training data and platform-level programming
- Custom instructions, which come either from the platform or from the custom instructions set by the user before a prompt begins
- Deep research, which compiles reference materials, examines them thoroughly and then integrates them into a report you can further interact with
- Retrieval-augmented generation or RAG, which is a function that allows chatbots to integrate external sources into their responses, essentially allowing them to search the internet or a source document to reduce training data hallucinations
Many users really stick to the first two types of memory, and thus they’re unable to really see the power than an AI assistant has at its disposal with custom instructions, deep research and RAG.
For example, Mike recommended using custom instructions to provide your AI assistant with an understanding of who you are, what your business is and what your objectives are so that it’s able to respond more specifically, and not generally, when you give it a prompt. And you can also use these custom instructions to create a personality for the AI assistant. Mike said he’s named his CLU – a reference to the AI program that becomes the bad guy in the movie Tron: Legacy – and has given it the role as his AI Chief of Staff so that it conducts itself in that persona.
With an AI assistant more knowledgeable about your business, you can generate better output, but Mike pointed out that because of the limitations of chat memory, you sometimes have to remind the AI assistant what you’ve been talking about to keep the main scope of your discussion from veering off course. The practical tip he offered is to occasionally ask the chatbot to summarize what’s been covered up to this point. This allows the chatbot to reference the summary in a more efficient way than to have to read the evolving discussion and also ensures that the most salient points don’t get lost in the discussion.
Beyond that, Mike recommended using something he called the PRICE framework. I don’t know if this is proprietary to him or comes from another source – I didn’t find it in common use when I searched online – but it is useful. It boils down to five components that make a good prompt for opening up a discussion with a chatbot.
They are:
Persona, Resources, Instructions, Constraints, Examples
Persona instructs the chatbot how to respond to you. Even if you have already set a general persona, you ask the AI chatbot to role-play. “You are the CMO of a fortune 500 company” or “You are a consumer in the target market for this product” or “You are the Incredible Hulk and confused about my brand” or anything else you’d like the chatbot to be.
Resources are the information you provide to the chatbot as reference materials. These can be text-based, but many chatbots are becoming multi-modal and can also review raw data output, images, charts, graphics, web links, sound or even video. Generally, the more specific these resources are to your prompt, the better. Most tools allow you to attach files as resources.
Instructions are what you want the actual chatbot to do, and these need to be crafted as specifically as possible to give you maximum control over the output. Remember that chatbots tend towards broad styles and terse output, so you can tell them to give you output that’s different from the norm or which has more depth and detail. Just remember that chatbots sometimes take instructions quite literally, so be precise with your words and avoid idioms or figurative language whenever possible!
Constraints are your guardrails for how you want the output to come out. I personally have found it’s very useful to require a chatbot to both cite its sources and also to check them to ensure that they are valid if I’m using a deep research tool or RAG-enabled tool. There’s nothing like manufactured quotes or sources to make you feel like the output is entirely worthless! But telling the chatbot to limit itself and to produce output that is genuine does help to limit hallucinations, though you still should check over the output.
Finally, examples are useful if you want the output to come out in a certain way or if you have a particular style or format you want the chatbot to stick to. This is very useful for proposals, form letters and other boilerplate-style output. Just remember that you always should check over the output to ensure that it doesn’t simply look like the example, but actually is factually and substantially correct. Far too many official documents have already been revealed as being written by AI due to obvious issues, and examples can only go so far towards keeping the chatbot from hallucinating!
Mike’s ultimate point was that if you can learn to use AI as a back-up brain that supports your business and helps you make decisions, you can use it as a force multiplier for a lot of tasks. This is more or less what those daily users of AI platforms tend to articulate to me, and while I am myself an optimistic skeptic of the technology, I think there’s some merit to the idea of using the technology to boost your productivity and help you get through the day.
And Mike Allton does something that I think is a must for any AI user, and that’s credit his platform for its output when he uses it. He’s not afraid to admit CLU authors articles or documents for him, and he’s even proud of what he’s able to do with the help of his AI assistant! That’s exactly what I recommend – always disclose when you utilize AI for any purpose, especially since it may make mistakes and you’ll be better off if you acknowledged that from the start!
I hope you found this recap useful. And as with all my podcast scripts, it was 100% human generated!
I’m Sean in St. Louis, and this has been The Marketing Gateway. See ya next time!
Leave a Reply