When Not to Use AI

AI has enormous potential to accelerate our workflows and to accelerate content production, but it can also slow things down. While generative AI tools have enormous potential to speed up your daily marketing operations, relying on them for all of your external communications is creating a massive hidden tax on your productivity. Many marketing teams are trapped in a cycle of relying on automated content, only to find that their messaging has become completely bland, flat, and filled with inaccuracies. In this episode of FiredUp!, we explain how over-reliance on AI can erode your strategic thinking and flag the specific areas where human judgment and taste must remain completely non-negotiable. This week, episode 136 of the FiredUp! podcast is about when not to use AI! 

Download the Multiplier Marketing Megapack today. Exclusive offer for all our listeners — get all our Startup Guides in one go! Over 50 pages of advanced tips and advice that dive deep into content marketing, search advertising, and marketing attribution.

In this episode of the FiredUp! podcast, the Firebrand team shares the two main areas where AI can create challenges in your agency and actionable steps you can take right now to incorporate AI where it counts but keep a human hand in the process. 

Morgan and Chris discuss:

• The Commodity Language Trap: When you use AI to draft your external communications, your messaging automatically reverts to the mathematical average. Relying heavily on automated tools guarantees you will produce commodity language that makes it impossible for your brand to stand out in a crowded market.
• The Doom Spiral of Thought Leadership: Outsourcing your writing to AI doesn't just make the copy worse—it degrades your strategic thinking. The manual process of writing and editing is exactly where you sharpen your unique point of view, discover insights, and find your voice.
• Guard Your External Messaging: While AI is highly effective for internal brainstorming and structure, it introduces major risks to your public-facing assets. Human judgment and editing are mandatory to prevent AI-generated errors and to protect your hard-earned brand authority.
• The AI Brief Paradox: AI-generated briefs often lack context and taste, creating a disconnect that can actually slow down content production. To keep workflows fast and accurate, the strategic foundation of any campaign must come directly from human experience and insight.

We are huge believers in the power of AI, but as marketers, our ultimate value lies in our taste, our judgment, and our creativity. If we delegate our core thinking to the machine, we sacrifice the very traits that make our brands distinctive. 

Thank you for listening! Tune in to all the episodes for practical tips on crushing your startup marketing goals. Don’t forget to follow, rate, and review the podcast, and tell us your key takeaways!

CONNECT WITH FIREBRAND:

Firebrand is a startup marketing agency. We help tech startups secure outsized marketing outcomes on their path to growth.

Youtube

LinkedIn

X (Formerly Twitter)

Work with Firebrand

Apply to Be a Podcast Guest

5.5.2026

Transcript

Morgan McLintic: 0:00

AI has enormous potential to accelerate our workflows and to accelerate content production, but it can also slow things down. Today, we're going to discuss some of the aspects where perhaps we shouldn't be using AI. Hello, everyone. Welcome to Fired Up, the podcast for marketers working in early and late stage startups. Hello, there. Welcome to Fired Up. My name is Morgan McClintic, and today I am joined by my co-host, Chris Albrecht. Chris, how are you

Chris Ulbrich: 0:34

doing? Great. Great to be here.

Morgan McLintic: 0:35

All right, so we're going to talk about where you shouldn't use AI, and I just want to top of the show here, say, look, we are huge users of AI and big believers in it. We have a number of agents that we've developed, we've had some AI-enabled apps that we've developed, we are all in on using it, but we've got to be judicious, because as marketers, we're here for our taste and our judgment and our creativity and our experience, and we need to apply that to this tool of AI, and there are some elements where we are finding it's actually slowing things down, and to do it in a traditional hand-written way, perhaps if it's a piece of written content might be faster, and so we're actually putting together some guidance for us to collaborate with our clients, so that we're all on the same page about how we use AI and how they should use AI in working with us, and we thought it would be useful to sort of unpack the logic behind that, because you guys are probably seeing it yourselves, whether you're working with agencies or you're working with other stakeholders in your business. AI sort of creeps in sometimes undetected, and that can cause all kinds of downstream challenges to know, okay, what was the sort of singular truth in here, and the intent, and picking it downstream can be a challenge. So, Chris, probably there are sort of two main areas that we think AI can create a challenge. Do you want to just frame it up, and then we'll dive into some of the details there.

Chris Ulbrich: 2:18

So, as you say, there are two categories where we've noticed that the use of AI is actually gumming up our work rather than accelerating it. The first big category is in communications between our client and our teams, where when AI is used to generate those communications and or instructions to us, it is often adding words, it's adding meaning that the client doesn't necessarily intend, isn't wedded to, but we have no way of telling the difference, and we suspect that such and such a line is an AI creation, but we, because we don't have the original prompt, we don't know really what was intended, and whether the AI was following what the client asked for, or they're just made it up to fill space, as AI sometimes does, and the other major category closely related is external messaging, where clients are creating, say, Q and A's off of materials like product sheets and other core materials, and just having the AI create external facing messaging just automatically, and often again the AI will use turns of phrase that are strange and that a human just wouldn't, or they'll, they'll make arguments that are like false distinctions, or they're just arguments that don't need to be made. And we receive this messaging, especially if we haven't been involved in constructing it. We try to be, but it's not always the case. And if we're not involved in putting the messaging together, and we're going to take it basically as gospel. This should be usable word for word by a reporter, and a lot of times we're finding that it is not in fact ready for external use, and it hasn't been scrubbed. So, in both cases, I think what we're seeing is there is a tension, a cultural tension going on, but where a culture has emerged in tech where the use of AI is virtuous in and of itself, and so our contacts are often under a lot of pressure to use it, but insofar as it's often used in marketing as a way to get to a first draft or a way to get a direction or to ideate, just come up with something rough, but a lot of times that's where it ends. The rough cut that the AI comes up with is passed to us like the final cut, and so the question is, Are you actually saving any time by prompting the AI to elaborate? On a document that we are going to need to unpick what was your intention and what was the AI's overall, that's the problem we're grappling with in the agency world right now.

Morgan McLintic: 5:11

There's the sort of communications with an agency or between different members of the marketing team, and then there's the sort of impact on messaging, is this directionally right or is this gospel, and knowing how much those things can edit. So, let's just start then with briefs, like a lot of marketing efforts. If you're creating ads, if you're writing a press release, or a byline, you're going to start with a brief. What are we trying to achieve? What is our messaging? What's the audience? What's the outcome we're trying to come up with, and that the intention of developing the brief is that it then saves time for the written outcome, whatever it is, so that it's correct, and you get fewer revisions. Now you've got almost a garbage in, garbage out issue here. If the brief is written by AI, right? So let's just start there.

Chris Ulbrich: 6:06

Yeah, that's exactly it. Garbage in, garbage out. Although it's originally, I think the definition of garbage in, garbage out assumed there was something going in. We've now come up with a new kind of garbage, in a way, which is absence, the absence of information, and I think one of the big challenges of using AI effectively in marketing is that I think we all intuitively, if you've worked with AI at all, you understand the more guidance you give it, the more structure you give it, the more information you give it, the more satisfying the end result tends to be, but on the other hand, the more time you spend assembling that information and structuring that information, the less time you're saving with the AI. There's a real incentive to skimp on the structure and information, and you hand over to the AI to fill out, say, a press release brief, a pretty sketchy set of materials, and then AI will do what AI does, which is fill in the blanks, right? And so all of a sudden now you have AI doing its best to just generate your messaging for you, but it's not really based on anything specific to your company, it's just taking a general average of how a company, maybe like yours, or maybe just any tech company, would approach this problem, and the kind of messaging, or the kind of wording they would come up with, and so you get this material in the brief that's either flattened out or irrelevant or wrong, just sometimes flat out wrong, because it is doing its best to come up with answers that it doesn't really have, like it's not just structuring information, which is what theoretically you're using it for, but it's dreaming up information, and as we know, AI's don't tell you what you know, the usually when it's making things up and when it's structuring information it already has, and that means that we receiving the brief back then have to spend a lot of time thinking that doesn't look right. I've never heard you say that before. Wait a second, isn't that wording? Isn't that from messaging that you changed five years ago? Because that will often happen. The AI will just go out on the web because you didn't give it the latest messaging, and it will find messaging from a press release you did a while ago, which was three product revs before ago, and it will insert it into the press release brief, not realizing that you discarded that a long time ago, and we, if we just started working with you, say six months ago, might not realize that's out of date, which is just all to say that you can save some time producing the words by getting the AI to write this brief for you, but if you don't painstakingly go through it and validate every word, then you're really just pushing that uncertainty over to us, and we're actually less able than you are to detect where the AI has gone wrong, so that means that we actually spend more cycles than you would on on teasing all those mistakes out, and just increasingly, and again, I think this is cultural, we are seeing clients just take the brief, throw it into Claude or Chat GPT, say here's the product doc with the new and with the new product for the announcement, answer all these questions in the brief, and that the end, the results are not optimal,

Morgan McLintic: 9:51

and I can see that happening, even though you add, here's a new product announcement, here's the stuff in product marketing, I'm doing this within my Claw project with our markdown. Files and all the latest messaging, maybe that gets outdated, and so it's synthesizing all the documents, so it makes perfect sense, right? Okay, take all of our messaging and fill in this stuff, but it's coming up with new words, it's coming up with different claims. We're often going to be trying to use those exact words in a press release, or at least some of the key sort of phrasing, and on a different note, I see this with with RFPs, right? Appointing an agency is an infrequent purchase, let's hope, and so often you'll want to come up with an RFP. What goes into an RFP? Oh, I know, Claude knows all about that, and come up with 10 questions of the kinds of things that I, you think I should ask. Yeah, these all sound perfectly reasonable, right? But it won't come up with 10, it'll come up with 25 questions. So these RFPs are enormous, and they're very specific, and you're like, okay, no one's ever asked me this, and but you end up like, okay, do you really need this information? Are you actually going to use this to determine which? Why are you asking me this, and when I've pushed back on the brief, which of course you're going to always push back on the brief, but oh no, I don't actually want that information. Then they're not a competitor. Then this company isn't a competitor, or yeah, actually that's not one of the verticals that we're going after, and I don't care about that. It happens right at the beginning of an engagement here, but day to day often happens in the briefs, and one of the things that we're saying here is a short human written brief is better than a long AI one, right? Just get put the time in humanly at the front, and that's going to save time on the back end too. Okay, the next area that we see AI coming in, Chris, is with quotes for this is a real PR one, but quotes for the media, right? They just to be clear, the media cannot use AI written quotes for in many publications, they are trying to screen out reporters, are trying to screen out AI quotes, because they are not looking for a quote from Chat GPT, they are looking for a quote from a specific spokesperson, and feeding them in is actually damaging to your company's reputation and the agencies and all their other clients, we're poisoning the well here, if we're, if the agency is not catching this stuff, and I'm seeing this as a, as an increasing problem.

Chris Ulbrich: 12:28

Yeah, it is. I, and I will want to acknowledge up front that it's in practice, it's nuanced, right? It would be easy to make some sort of sweeping claim that you can never use AI to generate quotes. What I think you are correct that in a lot of cases reporters or publications that explicitly prohibit that. At the same time, I do see what are plainly AI-generated quotes finding their way into the media. I see entire AI-generated articles finding their way into the media. Right, I think the right way to approach this is to assume that unless you know otherwise, you need to write something original. Yeah, or at minimum, say Claude generated this. Here's the prompt that I gave it, that's key, because we need to know what your original intent was, and then could you rewrite this in a human way, so that we can check that it's totally on message, and that it sounds like a human. Now, this is something I think we're going to talk about more than once in the course of this episode, and we've talked about it before, but just aside from the ethical question that publications might be concerned with about running quotations that weren't actually generated by a human, there's just a practical problem that the characteristic turns of phrase that AI comes up with are just, they're getting people are getting more and more aware of them, and I think that is contributing in some way to this growing sense of unease, and the statistics you read about Gen Z, like increasingly uncomfortable with AI, and wishing that it wasn't as prevalent in their lives as it is, and that they're talking about rejecting AI. I don't know if that's a phenomenon that's gonna be true over the long term. I don't know what direction things are gonna go ultimately, but we are in a moment where a lot of the tricks that AI can pull that seemed so novel and amazing, a couple years ago, people are starting to see through them now. They're starting to see whenever they see a word like quietly or shifts, or the benefit compounds, those turns of phrases that, like, AI just adores, they're seeing that it's AI, they're treating it as slop, it. Increasingly, and they're turning off, so part of what you're trying to do when you're providing a reporter with a point of view is you're trying to be distinctive, you're trying to say something that others aren't saying, and AI, as is well documented, tends to give you the lowest common competent response. Yeah, it's not going to give you something that sounds terrible. In fact, it'll often sound better than it means. If that makes sense, the semantics are kind of messed up a lot of the time, but it's good at generating smooth sounding pros, but it's not going to generate anything that is different, much different from the spokesperson at your competitor, who's also using AI to come up with their input to this same opportunity, and so the goal is really to set yourself apart with something punchy, with something easily digestible, and something distinctive, and AI is usually not going to give you that,

Morgan McLintic: 16:03

and in fact, for a lot of these written opportunities, they're going to be coming for in a call for sources that's competitive. There can be up to 60 others who are submitting quotes to here. So speed is important in, we often call it rapid response, and it needs to be rapid. Therefore, using AI feels apt here, but the reporter is going to receive 50 quotes, all of which sound fairly the same, because you're all prompting, hey, what should I say about this, and it's going to, it's going to come across, so having something distinctive there, I think, is really important, because just being quick with something that's bland isn't going to get you picked out of the pot here, and so you're just basically wasting the opportunity.

Chris Ulbrich: 16:51

Yeah, AI is going to give you a commodity opinion, and look, I can't say it's never going to work. Sometimes speed matters more than originality, but again, as a rule, I think you need to structure your rapid response program around the original creation of material, which means having pre-approved messaging that your agency worked on with you on various topics, so that they have a starting point for comment that is human-generated, where they understand the sort of the ideas and the point of the messaging that underlies the comment, and then they can be quickly productive when an opportunity comes along. It's more work, but you're probably going to get better results in most cases.

Morgan McLintic: 17:40

Agreed, our next piece of guidance is around editing, so we've got the brief. We've talked a bit about production. Now a piece of material is going to go through, let's say it's a byline, it's going to go through approval, and if the article has been written by hand, what we don't want at this point, Chris, is to edit it with AI, but why not?

Chris Ulbrich: 18:07

Well, this is another example of the tangled web problem we described earlier, where the AI's contributions are impossible for us without any context to pick out from the changes that matter to the client. Increasingly, we'll turn in a piece of writing, and we'll get back something that has been edited top to bottom by AI, not necessarily because there was anything really deeply wrong with the original version, but because if you give the AI a simple prompt like, make this tighter or put more in here about such and such, or I want more emphasis on it. It will often, unless you tell it otherwise, just restructure the entire thing, and so we'll get back a press release, which is often full of errors that the AI has inserted, or extraneous arguments, which we're not trying to make, and because we don't know, we don't usually get the prompt, and this is sort of a best practice we're leading up to here. Give us the prompt, because we don't have the prompt, we don't actually know what changes the AI made that the client is wedded to, and what changes the AI just added superfluously, and again, because I keep coming back to this cultural problem, because I think that culturally our clients are under a lot of pressure, not just to use AI, but to use it to spend as little time as possible on the tasks that you're using AI for, they'll just outsource the review to Claude, and without a lot of guidance, and then Claude will generate something, and they'll just send it right back, and they'll say task done. Yeah, and in theory it should work that way, but in practice it doesn't in our experience. Maybe someday AI will be more reliable. And will be so in touch with your, as the client, your individual preferences, and it will have such a deep understanding of the context of the company, of the business, and the technology that it would basically do in a couple seconds the same review you would have done in 20 minutes, however, that's not the case now. What we get back often is just a horrible mishmash of material. There might be some good ideas in there, and then we don't know if it's yours or if it's the AI's, or.. and you might say, well, it doesn't matter, but it does if you're changing direction, right? Because even a good idea that is not to the point that doesn't have a place in the document, so we'd understand, are we changing direction here with this document, or did the AI just come up with a different take, and it's not actually applicable, so again, moral of the story, if we really would prefer you not do the edits with AI, but if you do, please just give us the prompt that you gave it, because then we will have some idea what you were trying to achieve, and we can come back and say, okay, you gave it this prompt, this is what you care about, the AI didn't really achieve that here, and here it had a good idea. Here we can keep that right, but it's us having to guess what is AI and what is you, but what is your intention, and what is just probabilistic sentence completion that actually uses up a ton of cycles on our side, and that's your budget, it feels like you are speeding things up, because it only took you five minutes to do the review, or 10 minutes, or whatever, but it actually, I can tell you, you're kicking back an hour or two of work to us, so it is a balancing act, but increasingly we're seeing quality take a back seat to speed, and it is - it's gumming up the workflow when we produce content when we're trying to get a press release for a major announcement together,

Morgan McLintic: 22:12

and I can understand why that happens, right? Of course, because you get a piece of material and you're using Flawed for a lot - I use it a lot, and it's got the project's got all the documents, it knows your writing style, it, and it's produced some great, good stuff. And then here comes this other piece of material, like it's one of 15 other things I gotta do today. Okay, just this article, it's missing this message, can you just add it in? And then you turn it around, send it back, and I've seen it. Oh, no, I had Claude tighten this up a little bit, and then here it is, but as you say, it's reworked the whole thing and added its clawed smoothness to the words, and so you end up with something that is going to flag all the AI-generated content detectors, and sometimes if it's an editorial piece, that means it can't be used right when you've taken something that is handwritten, spent 10 minutes just getting clawed to refract, but the output is now unusable. And okay, then we've got to rewrite the whole thing.

Chris Ulbrich: 23:16

You're talking about, like, a contributed article.

Morgan McLintic: 23:18

I'm thinking about a byline, yeah, contributed article. Let's just say that, because again, the small changes it's made has just smoothed over and added its clawsms into it and made it an unusable output. Again, it really kind of hasn't saved time, and I'm aware that this could come across as you just don't like edits on your material or whatever, but actually we love edits coming through, and the sort of dial, honestly, if you work in an agency, you've got no ego about the words or other people putting in comments on your doctors, just it's all collaborative, and so actually it's better to, as Chris has said, prompt us, not Claude, on some of that stuff, because I think we're going to get to an outcome that is faster,

Chris Ulbrich: 24:02

could we just dial in on that, because I think that's a really important principle, and it's a principle that I think it would be one of the big takeaways from this conversation, that in a lot of cases the prompt that you've given Claude is really the most important communication you could have given us, and if you step back for a second, what did you gain by prompting Claude? If you just, if you gave us literally the words you gave Claude, we too can write a press release. Granted, we're not going to generate the words as fast as Claude will, but then once we've generated them, we can explain exactly why we picked every word that we did, and I think in a lot of cases we are finding that the prompt you're giving Claude is what matters in the communication with the agency, and that anything Claude is producing beyond the prompt, beyond simply restructure. Structuring information that's already there, like your product information, anything that it spins up is superfluous to the process, right? Again, using the AI to structure information that already exists, that makes some sense. Using the AI to generate ideas and whatnot, only if you are prepared to check it word by word and say I agree with every word that this AI has come up with, because this is going to be external facing material. This is material that ought to be safe for quotation. This ought to be material we're proud to put in front of a journalist, and it's good enough is not good enough for these kinds of use cases,

Morgan McLintic: 25:42

I find when I'm ideating with it, it will push you to say things that, like definitively, it's this and not that, it loves that, that you don't want to make that argument, it's because often arguments are far more nuanced than that, so it will often make claims that are wrong, and when you ask it, oh, it's like, oh, I just thought this is the kind of thing that you would say based on your experience. I'm like,'Well, actually that's not true, and I don't want to say that. And so, yeah, you absolutely need to be careful about the positions that it's inserting,

Chris Ulbrich: 26:15

and we've touched on this, but again, given that this is external facing material, it is not nothing that it's not just that Claude produces smooth prose. It has that sort of uncanny valley effect familiar from like the animation world, where an animated figure is just human enough to seem not like a traditional like Disney animation, like a Bambi or something, but it's not convincing enough to be fully human, and so it actually creates this kind of negative reaction, and Claude Chat GPT, the words they choose a lot of times, it's like they, they've taken a word and they've picked the synonym that's about three degrees off what any native English speaker would have used, but it's almost like the choice was made just for variety's sake, just to mix it up a little, and so Claude will often just come up with word choices that are bizarre, but because the structure of a sentence is so forceful and punchy, you just, your eye just glides by it on first reading, and then you actually read the sentence, and you say that just doesn't make any sense, right? And that matters in documents that are going to be parsed closely as messaging is

Morgan McLintic: 27:35

right. Do you think that the use of AI daily by all of us is changing the language that we expect and we accept, because now handwritten LinkedIn articles, blog posts, they're sort of lumpy, and they can sometimes seem almost less well drafted than the smoothness we're so used to, the AI smoothness and the sort of structure that it has, it feels very polished, and sometimes when you read someone's, like, their thoughts vary, and they go back, the structure is a bit odd, and you have to work hard to actually understand a little bit. Do you think it's actually changing our accepted language?

Chris Ulbrich: 28:16

That's a really good question. I have heard on LinkedIn and elsewhere, people saying that they are intentionally roughing up

Morgan McLintic: 28:26

or putting on typos,

Chris Ulbrich: 28:28

putting in typos. I heard of some AI, some founder that intentionally uses a double dash rather than an em dash to communicate this was written by a human, right? And so I feel like both things are true. If you live on LinkedIn, you would conclude that most people are writing with AI. LinkedIn is an absolute fire hose of AI slop, and some of my favorite posts, cynically, are posts criticizing AI slop that were themselves in PR writing, like PR people posting about AI slop and the need to have a human voice, and their post was clearly generated by AI

Morgan McLintic: 29:11

all the time,

Chris Ulbrich: 29:12

right, and it is so demoralizing, honestly, to watch that stream of AI-generated pros go by, so on the one hand, if you looked at just that, you would say the momentum in society is toward using AI to communicate, and that it's hard to say whether is that your communication anymore. When you say, like, is it changing the way we communicate, is it changing the way we write? If you're not writing at all, then it's a philosophical question, but I do think it is generating a reaction, but like hipsters spinning vinyl, I don't know if it's ever, if it's going to be more of a niche phenomenon. The mass phenomenon does seem to be the move toward using AI. Yeah, do. To communicate, I don't love it, but it, but objectively, that does seem to be the direction.

Morgan McLintic: 30:05

You could see people get out of the habit of writing, and when you're writing, you often form your thoughts, and it makes you come up with the ideas and put in the sort of mental tension to do it when you have to do it by hand, and it's far easier to just outsource that to AI, and so that not only does the writing get worse, but the thinking can get worse too, and so that's the sort of, I guess, the concern, or the sort of doom spiral of thought leadership there, and I guess overall education that we obviously need to avoid, putting in the time to write is often really very valuable, because you sharpen up your thinking and you're not outsourcing it to the edge,

Chris Ulbrich: 30:51

and again, to come back to messaging, I think we talked about points of view needing to be distinctive, but your external messaging around your product, that's exactly where you want to be at your most distinctive, and if you use AI, you will be at your most average, and this kind of goes back to a principle we've been talking about a lot as an agency, the importance of brand awareness in a sea of brands fighting for attention using more or less the same language, you can't hope to stand out by using commodity language yourself, because that's what you're going to get from AI. Commodity language

Morgan McLintic: 31:30

that feels like a great place to wrap up. Chris, thanks for your time. This is an important and evolving discussion, I think, and we love using AI, but there are certainly some areas when it comes to the brief, and it comes to editing material and communicating externally, where we got to just flag some notes of caution. Thank you all for listening. If you like that, we'd love it if you could follow us or drop us a comment, and we will see you all again

Unknown: 32:00

next.