In this episode, Antony Cousins joins On Top of PR host Jason Mudd to discuss AI in the public relations industry, including what GPTs are and how to use them.
Tune in to learn more!
Short Guest Bio
Antony is an accredited public relations practitioner who honed his PR and communications expertise in a variety of roles in UK Government organizations, including the Ministry of Defence, Home Office, and Cabinet Office. He now serves as Cision’s Executive Director for AI Strategy.
Watch the episode here
Also available on
Listen to the episode here
5 things you’ll learn during the full episode:
- What GPTs are
- Implications GPTs have in the PR industry
- What autopilot AI is
- How to build GPTs
- How to prepare your employees for a world with AI
About Antony Cousins
Over the last 23 years, Antony has held diverse roles in media relations, strategic communications, political analysis, counter-terrorism, and, over the last 10 years, AI-tech leadership including CEO of an AI startup focused on detecting fake news and misinformation. Ant is currently Head of AI Strategy for Cision, the largest global provider of media intelligence technology, as well as the Tech Hub Chair of the Association for Measurement and Evaluation of Communications (AMEC). He is an accredited PR practitioner, CIPR committee member, part of the UK's All-Party parliamentary group on AI, and was named in PRWeek's 'Dashboard 25' in 2023 for tech influencers.
- “I fell into it really by accident.” - Antony Cousins
- “I was in that startup for five and a half years basically iterating, building on, and learning how to implement AI in businesses. And that stayed with me. Basically, I didn't want to do anything other than AI after that.” - Antony Cousins
- “I was hooked from day one, but just when I saw the real potential of it, and I would say the potential of it not just to automate, not just to save time, that's the obvious stuff with AI.” - Antony Cousins
- “People ended up doing jobs that were more interesting and more fulfilling and automating the kind of tasks that you don't really want to be doing as a human. And I think that's the thing that drives me.” - Antony Cousins
- “AI can't replace the human touch.” - Jason Mudd
- “We're not trying to remove the human from the process, just automate the basics, like the 80% is a good way to think about it.” - Antony Cousins
- “The human still needs to add that empathy still needs to add the emotional understanding, the contextual awareness that the model won't have.” - Antony Cousins
- “AI can replace some of the busy work that we don't like to do or the tedious tasks and work we don't like to do.” - Jason Mudd
- “If you create a campaign and your content happens to plagiarize a campaign from some competitor from two years ago, the reputational impact is still yours. So you still need to take responsibility.” - Antony Cousins
- “I've heard a couple of good cliches. One of them is that clients struggle to even give good input to their agencies. So don't worry, they're always going to need an agency.” - Jason Mudd
- “If you can think about writing your prompts in a very literal way, you're going to get a better outcome.” - Jason Mudd
- “That's exactly the value of an agency and the value of a specialist is because they've got more contextual awareness and a whole bunch of history of your brand, of your situation, of other competitors doing the same things.” - Antony Cousins
- “The idea is that generative AI is a good first draft, but it's never going to be a great final draft. So it's a good beginning and a bad ending.” - Jason Mudd
- “The work product is only going to be as good as the experience and the input you give to them.” - Jason Mudd
- Connect and learn more about Jason Mudd on LinkedIn.
Additional Episode Resources:
- Connect and learn more about Antony Cousins on LinkedIn.
- Check out Antony Cousins on ‘X.’
- Visit Cision for more information.
Additional Resources from Axia Public Relations:
- Listen to more episodes of the On Top of PR podcast.
- Find out more about Axia Public Relations.
- How to ethically use AI in public relations
- PR Firm for Artificial Intelligence Companies
- How our PR firm is using AI and how you can too with Jason Mudd, CEO of Axia Public Relations
- How AI can help improve your public relations
[03:15] Why a job in artificial intelligence?
Ant: “I fell into it really by accident.”
Ant: “I was in that startup for five and a half years basically iterating, building on, and learning how to implement AI in businesses. And that stayed with me. Basically, I didn't want to do anything other than AI after that.”
Ant: “I was hooked from day one, but just when I saw the real potential of it, and I would say the potential of it not just to automate, not just to save time, that's the obvious stuff with AI.”
Ant: “People ended up doing jobs that were more interesting and more fulfilling and automating the kind of tasks that you don't really want to be doing as a human. And I think that's the thing that drives me.”
[05:39] What are GPTs?
Ant: “A GPT is a customizable instance of ChatGPT. It allows you to build your own customized version of chat GPT by setting instructions, directing behavior, uploading your own documents and knowledge to add on to what ChatGPT already knows, and building basically a small application that you can then release to other people or use yourself or release to your team or even release to the world.”
- GPTs are a way for people to “code” without understanding how to code.
- They build upon each other.
- They automate small, tedious tasks.
[08:23] What are the implications of GPTs on the PR profession?
- PR is largely a text-based department.
- Now you can develop GPTs to do some of the tedious writing tasks.
- Everyone can now become a developer to do so.
- The PR industry will need to focus on hiring people good at emotions.
Jason: “AI can't replace the human touch.”
[10:52] Autopilot AI
- You can develop different GPTs to accomplish different tasks.
- If you develop an output of one GPT to be the input of another, then the AI can automate the task entirely.
- This is autopilot AI and means that the AI will do the task itself.
- We’ve been used to using copilot AI, which gives us an output at our command.
Ant: “We're not trying to remove the human from the process, just automate the basics, like the 80% is a good way to think about it.”
Ant: “The human still needs to add that empathy still needs to add the emotional understanding, the contextual awareness that the model won't have.”
Jason: “AI can replace some of the busy work that we don't like to do or the tedious tasks and work we don't like to do.”
Ant: “If you create a campaign and your content happens to plagiarize a campaign from some competitor from two years ago, the reputational impact is still yours. So you still need to take responsibility.”
[20:18] How to build GPTs
Ant: “The great thing about GPTs is the interface you use to build a GPT is itself a GPT.”
- Click openai.com/create.
- Read the step-by-step instructions.
- Play around and see how you can get it to do what you want it to do.
- The more you work with the technology and train it to do what you want it to do, the better the output it gives you.
Jason: “I've heard a couple of good cliches. One of them is clients struggle to even give good input to their agencies. So don't worry, they're always going to need an agency. If they can't give good input to the agency, how are they going to give good input to a bot who doesn't quite understand all the nuances?”
[27:57] Tips for writing GPT prompts
Jason: “If you can think about writing your prompts in a very literal way, you're going to get a better outcome.”
- The output is only as good as the input.
- If you don’t know what really good looks like, you’ll never get really good outputs.
- You still need humans around the AI for it to be the best it can be.
Ant: “That's exactly the value of an agency and the value of a specialist is because they've got more contextual awareness and a whole bunch of history of your brand, of your situation, of other competitors doing the same things.”
Jason: “The idea is that generative AI is a good first draft, but it's never going to be a great final draft. So it's a good beginning and a bad ending.”
Ant: “[The AI] put the same level of effort into coming up with a press release and a campaign for that new table as it did for a new car.”
Jason: “The work product is only going to be as good as the experience and the input you give to them.”
[33:20] How do we prepare humans for a world with AI
- Hire more junior-level employees/interns than previously
- They’ll be used to using AI and can teach you tricks
- You still need senior-level employees too!
- They’ll have the creativity that the junior hires might be missing
- We might have to train employees in creativity in a way we haven’t had to before
Ant: “If you can create that kind of imaginative, innovative, driven culture causing people to constantly ask new questions, can we do this better? Is there another way of doing this? And creating that mindset of innovation and applying innovation, those companies will move a lot faster than the companies that don't.”
Enjoy the Podcast?
If you did, be sure to subscribe and share it with your friends!
Post a review and share it!
If you enjoyed tuning in, leave us a review. You can also share this with your friends and family. This episode can give you professional insight into media coverage. Know your rights and the regulations to follow when it comes to the media.
Have any questions?
Welcome to On Top of PR with Jason Mudd, presented by ReviewMaxer.
Hello and welcome to On Top of PR. I'm your host, Jason Mudd. I'm with Axia Public Relations, and today we are joined by Antony Cousins. Antony, welcome to the show. He is the executive director for Artificial Intelligence Strategy with Cision, a brand name that most PR people are very well aware of. Ant is an accredited public relations practitioner who honed his PR and communication expertise in a variety of roles in the UK government organizations including the Ministry of Defense, home Office, and cabinet Office, and now serves as Cision's executive director for AI Strategy. Welcome to the show.
Thank you for having me.
Yeah, we're glad to have you. I'm glad to be here. I'm very curious, how did you get this role with Cision?
So yeah, I've been here almost a year now, but it basically happened through an acquisition. So I was the CEO of a startup called Fact Matter. We were using AI to detect misinformation, disinformation, fake news, and a whole bunch of other kinds of harmful content that brands don't want to be associated with, like racism, sexism, et cetera. And then we sold that company decision, and I stayed on to oversee the integration of that technology and decisions technology stack. Then effectively that was the deal that happened basically two weeks before the release of chat GPT 3.5, which obviously changed the world as we know it. Having taken the lead on integration, that technology then said, ‘Hey, can you help us with strategy in general?’ seeing as I've been in AI startups for the last 10 years.
Okay, excellent. Well, that sounds like a good win for them. And so you've gone from being an entrepreneur, working in entrepreneurial organizations to working at Cision. What have been some of the changes that you've experienced going through that transition?
So I came from the government, which is in our government, 600 or a thousand people. I did that for 14 years. I was in government. I jumped into startups 10 years ago and I think within about three days of joining a startup I was like, oh, I should probably have been in startups the whole time. I just felt more natural, felt more my speed, and I think what I was expecting when I got the decision was a lot slower. It's like, this is a big company, it's going to be super slow. There isn't going to be that kind of level of innovation and drive. I am actually a little bit surprised by how quickly they want to make things work and AI is only speeding that up. I think the way that the company's leaning into the use of AI internally and in our products has actually been quite a surprise for me.
I think the other thing that I'd mentioned specifically about Cision was I was also expecting a sort of a big business culture. I was expecting it to be a bit more political, all the things that kind of come with big businesses, but actually during the acquisition process, everyone referred to it as a family, and I thought that was just a bit of a sales pitch during the acquisition, but once they arrived, I absolutely loved the culture. The people are so good and it is like a family and I think one of the things I was sort of not expecting but really happy with as well.
So I got to know what attracted you to get into artificial intelligence.
So I was in the government for 14 years and I'd been in a combination of either technology or communications roles for all those years, and when I jumped into a startup, it was an AI startup and that was, so I fell into it really by accident, but I joined as employee three or four I think, and really took on everything to do with implementation, onboarding, customer success, all the kind of professional services side of the business, but was working very closely with products. It was a very small company and given the fact I used to be a full stack developer, I very naturally kind of leaned into the development of the product side of it. So I had basically exposure to the way AI impacted every part of the business and every part of our client's business. And I found that fascinating and that was really, I was hooked effectively from day one in that startup.
And I was in that startup for five and a half years basically iterating, building on, and learning from how to implement AI in businesses. And that stayed with me. Basically, I didn't want to do anything other than AI after that. I ended up as chief of staff of that business looking after products and everything that effectively was in sales. So yeah, I was hooked from day one, but just when I saw the real potential of it, and I would say the potential of it not just to automate, not just to save time, that's the obvious stuff with AI. That's what everyone goes after to start with. But I went through that process and realized that actually, the benefits we were seeing weren't necessarily just time savings or increased revenue or whatever. It's the people who ended up doing jobs that were more interesting and more fulfilling and automating the kind of tasks that you don't really want to be doing as a human.
And I think that's the thing that drives me. And if it's any kind of reassurance to those out there who are thinking that AI is going to take over the world or take all our jobs, it's like that isn't what motivates us to work on AI. What motivates us is the idea that we end up doing jobs that are more human. I saw that over the course of the years that I was working in the startup. So yeah, I was hooked once I realized that.
Yeah, that's a great quote or a great line for sure. That really means that AI doesn't replace us. AI is going to make our role in our jobs more human. I really like that. So we're here to talk about artificial intelligence in public relations. I think this is a topic that people are very interested in. We've done one or more episodes on AI and they've been very popular. So let's talk about GPTs, what are they and how they work. And sounds like you've got some authority and expertise on that.
Yes, and as much as anyone can, I was watching the dev day announcement live last Monday, so a week today they were released and I'd been predicting, I think a number of people had been predicting that that kind of technology was coming, but I think it hit quicker than a lot of us were expecting, and it came with some added surprises, which we weren't at all predicting. So effectively a GPT is a customizable instance of chat GPT, which I'm sure plenty of people on this call have been experimenting with and playing with effectively. What allows you to do is build your own customized version of chat GPT by setting instructions, directing behavior, uploading your own documents and knowledge to add on to what ChatGPT already knows, and building basically a small application that you can then release to other people or use yourself or release to your team or even release to the world.
And that thing that I think caught most people by surprise, is that OpenAI is in the next couple of weeks releasing a marketplace of these little applications. So effectively anybody can go and not know how to code, not know anything about AI, but effectively program through, in some cases a conversation with the GPT. You don't even need the type, just have a conversation. Here's what I want to build, here's the problem I want to solve. Here's how I want you to act. Here's some information I've gathered on this I want you to refer to. Now go ahead and do that for other people. Releasing that to a marketplace effectively means you can monetize your expertise, your skills, and your approach to dealing with certain business problems. In some cases, the data that you've built up over time, which you think is unique and valuable, you can monetize that really effectively in this store. So the store is probably easily within the next year, maybe even within the next six months, going to become the most valuable marketplace on the planet, which is a real surprise for a company that no one had heard of a couple of years ago to get to that level, probably beyond the Apple marketplace. That's the really surprising thing, but that's what a GPT is and how it works.
Yeah, it's very interesting. So I've been doing this long enough that I remember when email first came out and the internet and then text messaging and mobile phones and social media. So this just feels like another big game-changing disruptor in the business, and lots of people are going to make a lot of money at it, and it's going to hopefully, ultimately help people do their jobs better every day. So tell us, what are the implications of this on the public relations profession?
Well, I think the implications of this are very similar to the implications we had at ChatGPT when it was released, which is effective in public relations communications. We're largely a text-based department, whether it's our press releases, whether it's our social media copy or a consultant, or the analysis we do on audiences, there's a lot of text we use to conduct our tasks. So the advent of large language models being the forefront of AI has a really specific unique application. Obviously what GBTs do is give that kind of power to basically everybody, right? No longer in the hands of developers and engineers have to build software that allows people to do this kind of thing. So you can now do that for yourself. So it's really lowered the bar to basically make everybody a developer and made everyone an expert in data science.
That's the implication in public relations. You no longer need to be an engineer or developer. I think this is one of the bigger changes that over the last 10 years we've been trying to encourage people, hire more data scientists, hire more data literacy, train for it, and get good at interpreting data. We want to demonstrate our value. And actually, now you can say, actually, don't worry about that. Forget that, hire people who are good at emotional awareness and relationship building because that was always a slightly weird skillset to try and build into our departments. And now you don't need it. GPTs effectively give you the ability to take on any task. You think, ‘Oh, I wonder if we can do this.’ You probably can just have a conversation with GPT and that can be automated for you. So the speed at which we can experiment and iterate and get help with what we do in PR is now super quick and the floor is super low. So I think it is something we should be leaning into as an industry as fast as possible.
Oh, absolutely. I completely agree. I've been doing some speaking to the PR industry and at different associations about AI and things like that, and you get people in the room that are just really afraid of it or not interested or worried about it. Then you have people who are just kind of letting it happen and not sure what to do with it. Then you've got people like you and maybe me who are really trying to stay on top of it and really encourage others to get more involved and leverage it for daily use, understanding that you can't replace the human touch like you're describing as well. Yeah. So let's see. So the next thing we want to talk about is autopilot AI. So tell our audience a little bit more about what that is.
Okay, so if you take GPTs another step forward, we've got the ability for you to create a little application capable of carrying out a task or a series of tasks once you start joining these instances together. So a typical workflow might be creating a campaign. So if you are creating a campaign best practice, you'd sit down and figure out your objectives first, what is it we're going to try and measure our success on? What is it we're actually trying to achieve? Then this is just my workflow, might not be your workflow, but in my workflow, I'd be thinking, okay, if that's what I'm trying to achieve, what is my audience and how am I going to be impacting my audience? What are their concerns and interests? And from there, I'd figure out my content and then once I figure out my content, I'd do my planning, I'd figure out the channels, I'd get it out there.
And then you do the measurement and your evaluation insight. If you take that as kind of a step-by-step process, you could effectively have a GPT carrying out each one of those steps. And once you connect them all together, effectively what that means is the input from one GPT is the output of another. All you need to do as a human is start it off. All you need to do as a human is say, here's what I want to try and achieve with this campaign. Here's the business outcomes that I'm after the GPT from then go, okay, the GPTs can figure out, here are the objectives. That is an input to the audience analysis. It can do the audience analysis, the output of both of those feeds into your content creation and they start to feed each other. So effectively the connection of GPTs with each other allows you to chain them and create what we call a multi-agent system mass.
And that is probably the very next thing, the very next iteration is once you can manage multiple gpt and hook them together, then you've got some really interesting autonomy. That's what we're talking about with autopilot. Effectively copilot is what we've been living with for the last year or so, which is AI's ability to do a single task, give me a press release, give me a social media post. Autopilot is the ability for you to step back and just set a goal and allow the AI to figure out what those tasks are and carry them out by themselves. And that's obviously the direction of travel.
So now it really is possible to work from anywhere and just start your day out telling the AI what you want it to do and then check in at the end of the day, to make sure it did a good job.
Well, I would say you need a little bit more human involvement than that. We're not trying to remove the human from the process, just automate the basics, like the 80% is a good way to think about it. Whether it's audience analysis, audience analysis is a good example of whether humans will still have to play more of a role since thinking through audience analysis, the AI does a relatively good job of the psychographics, the socioeconomic demographic, whatever analysis you want to use is a pretty good job of that. But the human still needs to add that empathy needs to add the emotional understanding, the contextual awareness that the model won't have. If I was looking at this audience in a microcosm, sure that analysis is correct, but I need to take into account other actions that we've conducted, other actions we're about to conduct, and things that the model won't know. So I think applying humans to every step is still important. It's just, it's a different role every step that it was before.
Yeah, I get that entirely. And it's nice because the AI can do some of the heavy lifting. AI can replace some of the busy work that we don't like to do or the tedious tasks and work we don't like to do. And I know as an agency, one of the things that's difficult for us is you've got a media list that you're trying to reach out to and you're like, okay, who's responded? Who hasn't? What's the next message in the drip or the script or the follow-ups that we want to do to stay in front of 'em or follow up with them? And if there was just a little bit of an engine that could take in all that input and manage all those connections, that could be very valuable. Is that what you're seeing? Is that what you're describing?
Exactly. I mean the constant creation and distribution part I think is ripe for automation since that's currently a multivariate problem, which is you have many different journalists you could talk to and your personal relationships aren't going to be thrown away, right? You're still going to use those personal relationships, but it's just not possible as a human to maintain that many connections and understand who are all the potential journalists, who are all the potential creators, and influencers that might be interested in this news. And that's where the AI can help.
It can handle multi-threaded thoughts and multiple variables there, and then optimize the content to those guys. That's effectively where AI will come in. You're still going to need to oversee that, make choices, and be responsible for what you send out because the reputational damage is still yours if you end up accidentally, absolutely. If you create a campaign and your content happens to plagiarize a campaign from some competitor from two years ago, the reputational impact is still yours. So you still need to take responsibility. But if that could be drafted for you you can see all the variance of that, and obviously, as we know, increasing demands for personalization is one of the things driving us to inefficiency. So if it can handle a lot of the personalization for you, then for sure, that feels to me like a really great way to use ai.
I know we were having some conversations at the agency just recently. We've got another podcast episode we did on submitting bylined articles to industry trade magazines. Well, let's say you have a list of 20 to 25 industry trade magazines you want to send it to. Each one of those is going to respond differently, and you've got to remember who responded with what. And that's one of the things we talked about is, hey, we've got a great process for that. But one of the gaps is you still have to follow up. And so you might send out that article to 25 editors, but a third of them get back to you with a yes. The other third say, the timing's not good, follow up with me. And another third you just don't hear from, well, then you need three different approaches to those people to stay in touch and stay top of mind with them.
You can lose track, you can forget to follow up when they asked you to, or you can just forget you even sent it. Whereas it'd be great if you could set up a campaign, and I think we're all using scheduled emails now where somebody says, get back to me next Wednesday. You're like, okay, I'm going to schedule an email for next Wednesday to either remind me or to go ahead and send it to you. But what if that nuance and that busy work and that tediousness could be handled by an engine or a bot for you, and then you can move on with the higher and best use of creativity and building those relationships, one-to-one, developing better content and not worrying about the tedious nature. So that's just one of the things we've identified is we do a lot of article submissions for experts, but it's like this whole other project that you have to manage when you start sending it to multiple places.
It is a great example really, of a low-grade activity, which takes up a lot of time. And that's a great example of something. If there isn't a GPT out there as a newsroom chatbot, I'll create it before the end of this evening. So that's the kind of task where the way you would approach that, if you're creating a GPT, you'd instruct it to behave in a certain way. You'd give it your voice, you'd give it some guidelines on how to talk. You could connect it to your email so that if a journalist does respond and say, I want to schedule a talk, it could automatically handle the scheduling. And if the journalist asks a question which is actually contained within the recent, they didn't read it right, you can answer those questions. They never do that. So you can provide all that background and much more.
So your press release is obviously a distilled version of that bigger truth. You can have that bigger truth stored as a document and uploaded to a GPT. So if the journalist does go the extra yard, and I'm interested in this release, how about X? And asks you that additional question, even that question could be automated without having to come back to the human. So it's a great example of a specific task. Doesn't really require a lot of human creativity. Doesn't really require all the relationship-building. Or in fact, you could have a list of people. So if a specific kind of journalist or a specific outlet comes back to you, funnel that to a human because that's a relationship we want to cultivate versus that's an outlet we've never heard of, a journalist we've never heard of given the chatbot answer. So even that could be automated. It's a great little use case.
Yeah, I love that. So we're going to stop right there. I've got a bunch of follow-up questions for you, but we're going to take a quick break and come back on the other side with more
You are listening to On Top of PR with your host, Jason Mudd. Jason is a trusted advisor to some of America's most admired and fastest-growing brands. He's the managing partner at Axia Public Relations, a PR agency that guides news, social, and web strategies for national companies. And now back to the show.
Hello, welcome back to On Top of PR. We want to say thank you to our sponsor and thank you to Ant for being here. And we're talking more about this whole GPTs and everything else that's going on. I love the things we're talking about here, but I'm thinking about our audience if they're with us, if they're listening to us right now, what do you think they are? Well, they're clearly interested in what we're talking about. I've got to imagine, that you said they could build a GPT, and you said that we could start doing some of these workflows. Where do they get started? How do they go beyond what they might be using today with just chat, GPT, et cetera?
So it is relatively simple. The great thing about GPTs is the interface you use to build a GPT is itself a GPT. So by building a GPT, you don't just end up with a GPT, you have an experience of what the end user is going to experience. You see how it works. So it's a really interesting device they've created, and the way to do it is simply head to open ai.com/create and you'll see the page and it will talk you through how to build one. So it definitely, it's the same advice I'd give to anyone for chat GPT, which is the best thing you can do is go and experiment. Just play with it and play with it and iterate, test it out yourself. If you don't like the response you iterate, iterate the prompts. So it really is kind of a trial and error type way to approach it.
The one piece of advice I'd give is that there are times it's appropriate to use a GPT and times when it's not right. The age of software isn't yet dead. So in the case of the most recent release for GPT four Turbo, they've increased what they call the context window, which effectively limits the amount of information you can provide in advance of your prompt to enable the system to come back with a more accurate response. So previously that was something like 32,000 tokens, now it's up to 128,000 tokens. It was effectively the equivalent of a 300-page book. So you could effectively provide a book and then answer the question about anything in that book, and it would give you a really accurate response. And that's information in addition to what it already knows as inherently part of the model. So that's really useful for short amounts of analysis.
But in the case of PR agencies or in-house monitoring teams, if you're trying to understand the internet, that's still way, way too small. So if you're trying to bust through a million mentions, it's still not that level. If you're trying to analyze a dozen articles or 50 articles or so, then it's more appropriate before it starts to have issues with recall and accuracy and it isn't fast. So there are benefits to using a GPT in that kind of way. And if that's the kind of thing you want to do, go ahead and do that. However, be mindful that it still isn't a replacement for organizations that are out there. And I say this, I work for Cision, but not just, I just want people going out and thinking they can suck the entirety of the internet and ask a question, and get a response. So know the limitations, but iterating and experimenting is the number one piece of advice I'd give.
So we all know this is changing constantly. So for the record, we're recording this on November 13th and it'll air a couple of weeks later. About a week and a half, two weeks ago, I was in San Diego with 17 other agency owners and we were collaborating and working together to improve our businesses. And of course, we were talking a lot about AI and how each agency is using it and different ways that we can use it both to help our clients and our agency. There's clearly one or two people in the room that are kind of more experienced, more embracing, more involved in it. One of those people and I were talking about how we're currently training it to use our voice and giving it input and direction on how we want it to speak or produce content on our behalf. She is definitely stronger and more experienced in this space than I am.
So we're always learning from each other. She felt very strongly, very strongly that conditioning and teaching a bot to speak in your voice and whatever that's called with your profile, she was really thumbs down on that and felt like it actually made things worse, not better. And I've been trying to do a lot of that customization and kind of training it. And I've noticed now, and not now, but within the last month or two or three, that there's now an option to turn that on and off. You can toggle that on and off, which I think is pretty cool. It sounds like you're in my camp that says, Hey, the more you train it, the more input you give it, the better the end result. Is that what you're saying?
For GPT? Yes, 100%. We've already kind of seen research and analysis of that for what you're talking about, which I think is related to the customer instructions you can set for ChatGPT, and then I've experimented with that. I haven't seen a great deal of difference. So I think you might end up doing that on a per-prompt basis. So every time you engage, you set a little bit of a prompt upfront and then you get the result. But that is hard work. This is the challenge and the benefit, which is you can automate some tasks, but in some cases, you have to train it each time and then GBTs do automate some of that. But for really complex workflows, you're still going to need software to automate a bunch of those things so you don't have to do those things every single time.
Yeah, yeah, I totally agree. I think the one thing that most people miss when they're first using AI or excuse me, chat, GPT, is the ability to go in and after you've given it a prompt or a command and it's come back, is then to give it more input. So you can say, that was good, but I'm looking more for this or Don't do this, do more of that. And when I show people that, they're kind of like, wow, I didn't know I could do that. They just thought it was a one-and-done kind of engagement. And so if you're catching this episode and you're not doing that, I highly encourage you to do that because I feel like the best outcome really starts to happen as you start giving more input and more refining. That will also help you learn how to give better input on the initial generation request as well.
Exactly. You're exactly right. And the challenge is once you've got to that point of knowing exactly the right prompt like you've iterated and you've got, okay, this is the exact right prompt I want to use, you still need to set that up every single time. So there is a little bit of a pain in that case, especially if your prompt is sort of dynamic, in which case most of the prompt is the same each time, but sometimes you need to change little variables that can still be quite painful. And again, we're back to that's kind of why you might still need software for doing that kind of thing. Press releases are a good example. So if you end up with a prompt or a set of prompts and questions that result in a great press release, you need to do that every time. And even in a GPT, it's still not going to manage that workflow because of press releases, especially if you're in an agency situation, you've got the press release being drafted, you need to get approvals for multiple people. There are edits that are going to be painful to do in a GPT-type situation. Whereas in software, we sort of manage that by default. So I think there are still some tasks that you want to use software for rather than a GPT or even chat GPT because of that repetition issue.
Yeah, that's exactly right. And I think we're all getting better at it. I've heard a couple of good cliches. One of them is that clients struggle to even give good input to their agencies. So don't worry, they're always going to need an agency. If they can't give good input to the agency, how are they going to give good input to a bot who doesn't quite understand all the nuances? And I even catch myself doing that, that I'm describing something one way, and then it gives me exactly what I asked for, but the perception of what I was, it's not wrongly perceiving what I'm asking for. The input I gave was not of good quality. So in other words, it gave me exactly what I asked for. But if I asked it of you and I were talking one-on-one, you would get what I really meant, even though I didn't say it specifically, but it's being very literal, which I think is good. If you can think about writing your prompts in a very literal way, you're going to get a better outcome. So I like restating the prompt a little bit differently and massaging it. Do you have any other tips you want to share about that?
Well, you're making a really good point, which is that the output is only ever going to be as good as the input. So if you as an individual don't know what really good looks like, you're never going to get really good as an output. This is one of the fundamental challenges of AI. It will do what you ask it to do. It's not going to take your input. I go, that sounds interesting, but I think what you really want is this, but that's exactly the value of an agency and the value of a specialist is because they've got more contextual awareness and a whole bunch of history of your brand, of your situation, of other competitors that have done the same things. They can turn around and say, I hear you. Do you think that's your problem? That's not really your problem. Your problem is this. Right? Challenging the brief or in some cases accepting the brief, but choosing to go above and beyond it, is something a human has to do. So I think AI will have a role in this, but we still need humans around it to make sure we know what good looks like. But that comes back actually to some of the talent acquisition or retention issues we've got because of this.
Yeah, I'm going to get to that in just a second. The other thing I just want to share is somebody said, I don't know who, I've heard it a couple of times now that generative AI is a good beginning but a really bad ending. Have you heard that before? Who do we attribute that to? And if it doesn't make sense to our audience, we'll explain in just a second.
I haven't come across that one. You're going to have to enlighten me on that and that's one of the challenges of AI. It is basically now impossible to be completely abreast of everything happening in AI, which is a really exciting position to be in, but also kind of stressful.
Oh yeah, it's very exciting. I remember the first time I sat in front of a computer connected to the internet, I got on it six hours later, and I'm still surfing and learning and doing everything. The same thing happened to me the first time I got on TikTok. The same thing happened to me the first time I started doing ChatGPT, so I totally get it. The idea is that generative AI is a good beginning, a good first draft, but it's never going to be a great final draft. So it's a good beginning and a bad ending, meaning that or a bad, I'm misquoting it here. But ultimately that's what it means you can really use it as a good first draft or a good iteration. But if you take that first iteration, that first input, that first command that you gave it, and take the first product that it delivers to you and think, okay, I'm done.
That's going to end up really bad for you. But if you use it as a great beginning, a great way to start a foundation, a first draft, some inspiration, and then whether you as a human do the rest of the work yourself or you're just interfacing with the prompt, then you can get to something better. And so the example I like to give is as I mentioned earlier, I'm part of a group of agency owners. We get together and help each other. We have a thing we do every Friday. And so I was inputting into the chat, GPT, Hey, I need a name for this group that we get together every Friday. We do it on Zoom. Here's the name of our group, we need a name for the conversation. And so it came back with some thoughts and I was like, okay, these are good.
Give me more. Okay, more creative, more inspirational, whatever started coming back with. So I'm like, okay, these are good. I like three, seven, and nine and then give me more like those. And then I was like, okay, I like three, six, and eight. And then I said, okay, now make it an alliteration. And that's when the magic started to happen and we started really coming up with some really good stuff. Now, at the end of the day, this is what I always say, number one, if I was asking an employee to do that, they'd be so sick and tired of me sending them back to the drawing board, sending 'em back, but the GPT is like this, right? And it's quick and it doesn't get frustrated. It doesn't get upset with me. And then I can keep kind of changing and massaging the brief or the prompt or the input, and it just keeps riding with me. In fact, it's very apologetic sometimes. Hey, I'm sorry, I didn't get that quite right. Let me try again. So I think that's been really interesting,
A hundred percent. And any time of day or night, if you have a creative flash at 3:00 a.m. it's there for you if you want. I was testing it early with its campaign abilities and giving interest to things like I'm launching a car, we are launching a new event. And then I tested it with the most boring, and I, no offense to people if creating public relations campaigns for new table designs is your thing, but it's like I've got a new kind of table. I think it's really cool. And it put the same level of effort into coming up with a press release and a campaign for that new table as it did for a new car. And that's been really difficult if you're an agency trying to get if you've got some graduates coming in, great news guys got a new customer and they've got a new kind of table, trying to get the same level of input and excitement from those people is going to be difficult. But chat does not care. It'll give you 100% every time.
I like that. That's really good. And I equate it to being a really good high school college student or an intern or an entry-level employee. The work product is only going to be as good as the experience and the input you give to them. And so you've got to be ready to kind of take that on. If you just think AI is just going to do the job for me, I don't even have to know what I'm doing. You're in for a rude awakening. Of course. So, alright, so the final thing we wanted to talk about, because I know we're quickly running out of time, we're really glad you were here. Thanks for spending time with us. How do we prepare humans for this new world?
I get this question a lot and different variations of this question. I get either the question, do we need to make as many junior hires as we did before? Because effectively as you said, ChatGPT and we're using ChatGPT is a generic term. It could be any of the large language model providers. It is an effective intern or effective junior entry kind of role. So do we need as many juniors? And my answer to them is, yes, you need more. Because if you think it through people are high school right now are going to be mid-level account managers and account directors in 10 years' time, effectively, those kids are going to have the best study partner they've ever had in something like a ChatGPT, and they're going to be really practiced at exploiting it and have no problem going to it straight away.
So they don't have any of the kind of inhibitions to experimentation that we do. So they can be learning, it's going to be the best study part they've ever had, and that will be the case for most of their school career. So imagine those guys coming out of school and into the workplace, get as many of those as possible because each one comes armed with effectively an army of interns, right? They're disposal, which they know how to make the best use of. But the challenge is, and this is the flip side, the second question I get, which is, well, in that case, do I need as many old people? Do I need as many people with experience?
Experience, right? I've just got an army of interns, and that's interns right now, but as we said, ChatGPT is increasing its capabilities. It's an intern now, but it's mid-level creative in a year's time. My answer to them is, yes, you still need those people because if you imagine those kids coming out of school may not arrive in the workplace with the same level of creative skill that we had. You and I are both old enough to have honed our creative skills on the blank page problem, which is you've got a blank page, you've got an hour, client needs X, get on with it. And that's how you develop that over the years. But when you are never creating, you're only ever editing because you've just been using track to give you a start, right? A great start each time, then do those people come and arrive in the workplace with it's good knowledge of what really good looks like.
And so I think we do need to hire as many juniors. You can't fire all the people with experience. You effectively need to make sure that they're partnering up so that you're getting as much exploitation of the benefits and efficiencies from something like a ChatGPT. But you've got the experienced person doing what we spoke about earlier, which is saying that's not quite good enough. Actually, that isn't the right question we should be asking ourselves. That isn't really the brief. We should be fulfilling that partnership I think is going to be really, really important as well as making sure that as those kinds of kids come through and they imbue some of that experience, and this could come down to training for creativity in a way that we haven't had to before. Training for creativity, training for design thinking, and systems thinking training in these concepts which you can train because they may not have had as much experience of it as they come out of the school system.
You reminded me of a couple of stories, and if you give me a moment to indulge me here, but when I first was doing an internship back in the day, I remember I was working in a PR agency as an intern, and they had this big presentation they were working on for a client and it was a PowerPoint presentation, and I was like, oh, great, I can help with that. And they're like, oh, no, no, no, no. We've outsourced that to an agency that specializes in PowerPoint presentations and this is a new technology. There's no way you can do it. You're just an intern. This is a really big thing. And so fast forward about a week later, I was in a staff meeting and they're like, gosh, we haven't heard anything from that PowerPoint agency about this PowerPoint they're working on and the presentation's Friday, we're getting nervous, and I like Rose.
I'm like, I can do it. No, no, you can't do that. That's too high-tech. So you know where this story's going, and a couple of days later, still nothing. And they're like, Hey, that intern, he said whatever. So then I build out this PowerPoint thing and they're like, oh my gosh, you're a genius. I don't know how you know to do that. My point is that's what it feels like we're at today where people are going to laugh about this idea of maybe hiring engineers who are experts at engineering prompts and all that stuff. And I think there's value in that, and I think there's training in that, but one day it'll just become having the internet in our pocket or a mobile phone in our pocket. It'll just be something that we've all kind of just built around us kind of thing. And yeah, I agree.
That was kind of the point I made when I was with the other agency owners. You had one or two that are a little bit older, more experienced, and they're side hustling or something, or volunteering, I'm not sure what, but they're teaching at the university level, giving back to the profession, which I admire. But they're really having a hard time allowing or accepting that students are using these generative AI tools to help with their papers or whatever, and they're like, Nope, you got to redo that. You got to do it the old-fashioned way. And I kind of challenged them and I'm like, have you really thought about what you're saying? Because what you're telling students, oh, you can't source the internet. That's not reasonable. That's whatever. It's like I thought our job was to educate students so they had skills that were applicable in the marketplace today and in the future, not skills that were applicable in the marketplace 30 years ago. And they of course were like, I never really thought it that way. I just want to make sure they have the basic skills. And I'm like, yeah, teach 'em the basic skills, but don't penalize them for using the tools they have at their disposal that their employer will expect them to use and put at their disposal as well.
Yeah, a hundred percent. I think teachers may have the same implication for their roles as many other roles do, which is if you're a math teacher or math in America, then effectively you get to see the workings out because as part of the exam, you have to show your workings out in order to get to the answer. We've never really had that when it comes to writing, right? If you're doing a book report, you don't show your thinking, you just show the words. But I think now using a chat g PT type tool, you can actually, if you could, you could share the prompts. Here's what I was thinking, here's what I was reasoning.
So you actually get to see more of the thought process than you did before. You're not spending ages just ticking the words in the right order that appears in their book report. You're actually marking their reasoning and their thoughts, and you have the ability to talk to the students about what they were thinking in a way you didn't before. So I think teachers may spend less time going through a 10-sided book report just marking for the words and more time focusing on how to encourage curiosity in the student. How do I engage this student and get them to think about their reasoning and thought processes? So I think a potential impact on teachers is they get to be more human in their tasks too.
Right. I like that. I like that. The other thing that you're reminding me of is that now these generative AI bots and engines are giving you attribution, which they used to not do, or sources. They used to just give you information and you'd be like, well, what's your source for that? Now I'm starting to see where that's coming in by default, which I think is a big evolution there. So I think the last thing, and we kind of hit on it a little bit, but the last thing I really wanted to cover with you and then we will let everybody get back to their busy life and busy day, but the things we're talking about preparing humans for the generative AI world, from the perspective of culture, talent, acquisition, retention, we talked about the talent you need to staff for, but how can we be thinking about this to make sure we are building the right culture, having AI be used as a tool and not as a barrier and not as being, thinking expansively, not constricted about it, and also just acquiring and retaining talent in this environment. What are your thoughts there?
So I think the culture, each strategy for breakfast, a classic saying, and I think never has that been more the case than now, because effectively, as we spoke about earlier, if you can imagine it, you can pretty much build it now, in which case imagination just became, I think, one of the most important aspects of your culture in a way that I don't think it has before. Having ideas before was hard, execution was harder, but now execution just got easier. So actually, if you can create that kind of imaginative, innovative, driven culture causing people to constantly ask new questions, can we do this better? Is there another way of doing this? And creating that mindset of innovation and applying innovation, those companies will move a lot faster than the companies that don't. The companies go, oh, ChatGPT. I know what that is. Here's how we're going to use it.
No, you have to keep on addressing your own assumptions. You have to keep on learning in a way that you haven't before. So that iterative cycle, that iterative process is going to be crucial, not just for making your business succeed, I think also for attracting, and retaining the talent. Because if the talent gets used to that kind of process and ends up in a company that doesn't have that approach, they're going to get bored quickly. I've got ideas, I want to do things. And if you don't respond to that, you're going to lose that talent to someone who will. So that culture aspect, especially creating a culture of innovation, I think is just going to become the norm, which for me is great as an entrepreneurial CEO, that's how I'd love to live. So I think for certain people it's going to be a really freeing experience for others. It's going to be a little bit of a mindset shift and we need to help those people through that.
Totally agree. Totally agree. Hey, this has been a great episode, we really appreciate you being here. I just want to give our audience a chance –– if they loved what you had to say, they want to connect with you, and follow you on social media, or otherwise –– what's the best way for them to get ahold of you?
LinkedIn is my platform of choice, and I think that's also, if you follow me, you'll follow a bunch of other people that I follow who are leading lights in generative AI and LLMs. So find on LinkedIn and through here, through following me, you'll find a bunch of other people sharing even more insight than I do.
Okay, excellent. And to find you on LinkedIn, do they look for you by your full name or what's the best way to find you?
Antony Cousins. No “H” and I'll pop up.
Okay, perfect. Excellent. We'll also do our audience a favor by putting that in the episode notes in a link to your LinkedIn at ontopofpr.com. We'll also share resources from other episodes we've done on AI and a couple of other tools that we have available as well that we think will be valuable. So with that, this is Jason Mudd signing off for On Top of PR, hoping that today's episode helped you stay on top of PR. If there's someone you know that would benefit from this episode, please take a moment and share it with a friend or colleague who I'm sure they'll thank you later. And I thank you as well. And with that, be well, and thanks for tuning in.
This has been On Top of PR with Jason Mudd, presented by ReviewMaxer. Be sure to subscribe so you don't miss an episode and check out past shows at ontopofpr.com.
- On Top of PR is produced by Axia Public Relations, named by Forbes as one of America’s Best PR Agencies. Axia is an expert PR firm for national brands.
- On Top of PR is sponsored by ReviewMaxer, the platform for monitoring, improving, and promoting online customer reviews.
About your host Jason Mudd
On Top of PR host, Jason Mudd, is a trusted adviser and dynamic strategist for some of America’s most admired brands and fastest-growing companies. Since 1994, he’s worked with American Airlines, Budweiser, Dave & Buster’s, H&R Block, Hilton, HP, Miller Lite, New York Life, Pizza Hut, Southern Comfort, and Verizon. He founded Axia Public Relations in July 2002. Forbes named Axia as one of America’s Best PR Agencies.
Find more On Top of PR episodes on: