AI has made it dramatically easier for B2B SaaS teams to produce content. But it has also made one problem painfully obvious: more content does not equal better messaging.
In our latest episode of Performance Delivered, host Steffen Horst sat down with Chris Silvestri, Founder of Conversion Alchemy, to unpack how AI personas, synthetic research, and AI-assisted workflows can help SaaS teams improve messaging — or completely derail it if used incorrectly.
The takeaway was clear: AI does not fix messaging problems. Systems do.
Below, we break down the core insights from the conversation and what they mean for modern SaaS marketing teams.
The Real SaaS Messaging Problem Isn’t Copy. It’s Clarity.
Most SaaS teams believe they have a copywriting problem. In reality, they have a clarity problem.
AI tools like ChatGPT can generate landing pages, emails, and product copy in seconds. But when the underlying positioning, audience understanding, and narrative strategy are weak or inconsistent, AI simply scales the confusion faster.
As Christopher explained, teams often finish a messaging project and then dump all their materials into ChatGPT, asking it to “write a homepage” or “rewrite a pricing page.” The result is usually generic, inconsistent, and misaligned with how the product is actually sold.
The issue isn’t the tool. It’s the lack of a system behind it.
Why Prompting Isn’t Enough (and Context Engineering Matters More)
Many teams obsess over prompts. But prompting only works when the AI already understands your business.
What actually drives strong AI output is context engineering:
- Real customer interviews
- Internal stakeholder perspectives
- Competitive and market research
- Clear positioning and messaging frameworks
AI should be treated as an input machine first, not an output machine. When teams use AI to analyze research, synthesize patterns, and pressure-test assumptions, the resulting copy becomes far more accurate and useful.
Without that context, even the best prompt will produce shallow results.
AI Personas Only Work When They’re Grounded in Real Data
AI personas and synthetic research are often marketed as shortcuts to customer insight. Used incorrectly, they are exactly that — shortcuts that lead nowhere.
Christopher emphasized that AI personas should never be created in isolation. Asking an AI to “act like a CFO” without feeding it real interview transcripts, objections, language patterns, and decision criteria leads to vague, internet-average answers. When AI personas are grounded in:
• Customer interview transcripts
• Sales call notes
• Competitive positioning
• Internal product knowledge
they can deliver directionally accurate insights that help teams test messaging, uncover objections, and iterate faster — without replacing real human research.
What a Real AI-Assisted Messaging System Looks Like
High-performing SaaS teams don’t use AI as a single chatbot. They use structured systems.
A practical AI-assisted messaging setup typically includes:
1. A general marketing assistant trained on raw research and internal knowledge
2. A strategy-focused workspace for positioning and messaging frameworks
3. Dedicated copywriting environments for different channels
4. An explicit human editing and validation step
The goal is consistency, not automation for automation’s sake. AI accelerates thinking. Humans remain responsible for judgment.
AI Doesn’t Replace Marketers. It Raises the Bar.
One concern that often surfaces is whether AI agents replace marketing roles. The answer is no — but they do change what good marketing looks like.
AI can:
- Speed up synthesis
- Generate first drafts
- Test variations quickly
But it cannot:
- Conduct empathetic customer interviews
- Make strategic tradeoffs
- Own brand judgment and nuance
The most effective teams use AI to compound human expertise, not substitute it.
Messaging Frameworks Are the Foundation for Scale
Whether working with internal teams, agencies, or contractors, consistency comes back to one thing: documented messaging frameworks.
Clear positioning documents, narrative strategies, value propositions, and brand voice guidelines act as guardrails — for both humans and AI.
When those frameworks exist, multiple people (or AI agents) can produce on-brand, aligned messaging without reinventing the wheel each time.
Without them, every new asset becomes a guessing game.
The Future: Continuous Research, Not One-Off Projects
One of the most compelling ideas discussed was the shift toward continuous research.
Synthetic research platforms and AI-driven testing are making it possible to validate messaging continuously, rather than treating research as a one-time initiative. This opens the door to:
- Faster iteration
- Earlier validation
- Better alignment between marketing, sales, and product
The teams that win will be the ones who know where to collect data, how to feed it into their systems, and how to close the loop with real human insight.
Final Thought
AI is not a shortcut to better messaging. It’s a multiplier.When SaaS teams invest in clarity, research, and systems first, AI becomes a powerful ally. When they don’t, it simply accelerates noise. If you’re exploring how to use AI personas, synthetic research, or AI-assisted workflows inside your SaaS organization, start with the fundamentals — and let AI amplify what’s already clear.
Episode Transcript
INTRO (00:01.89)
Welcome back to Performance Delivered, insider secrets to marketing success, the podcast where we uncover strategies, frameworks, and real-world lessons driving results in today's fast-moving marketing landscape. I'm your host, Steffen Horst, and today we're diving into a topic that every B2B SaaS marketing leader is thinking about, even if they're afraid to admit it. How do you bring AI into your messaging workflow without losing your brand voice, your strategic edge, or your understanding of the customer?
AI has made it easier than ever to produce words, but harder than ever to produce meaning. And as companies jump into synthetic research, AI personas, and automated copy generation, one thing has become crystal clear. Tools don't fix messaging problems. Systems do. Joining me today is Christopher Silvestri, founder of Conversion Alchemy.
where he has helped B2B SaaS companies turn fuzzy messaging into scalable systems that align leadership, marketing, and sales, and convert across web sales and email. Today, we're exploring what AI is getting right, what teams are getting wrong, and what an AI-assisted messaging system actually looks like inside a high-performing SaaS organization. Now, Christopher, before we get into the techniques and frameworks, I'd love to start with the why.
Steffen Horst (01:24.13)
What was the moment you realized SAS teams weren't struggling with copy, they were struggling with clarity? And how did that insight lead to building conversion alchemy?
Chris (01:35.056)
Yeah, so thank you so much for having me, Stefan. So excited to dive into these topics, which I think a lot of people are kind of like a buzz behind a of people's minds, but no one really dives super deep into them. And a lot of companies keep their AI secrets and AI usage kind of behind closed doors. But yeah, like one of the first signs that kind of
told me, okay, like we need to actually approach this a different way was when a lot of my clients, what they used to do was we finished our messaging work or the research, the copywriting that we produce at the end. What a lot of them did was taking all of what we produced and dumping into ChatGPT and then literally ask it, okay, write me one page or write me a product page or a pricing page, right?
And that literally doesn't work for a lot of different reasons that we can dive into it. But that was kind of the first sign. like people know that they can use AI to scale things, but there's no process, no system behind it.
Steffen Horst (02:45.998)
A lot of SaaS teams are excited about tools, but get disappointed with the output, because it all depends on how you prompt. In your experience, what do most teams get wrong? And when they bring AI into their research and messaging workflows, what makes the difference?
Chris (03:01.628)
Yeah, so you had it right. Prompting is super important, but it all starts, think, with context. So context engineering now, it's probably where I would focus the most time on at first. And that's because I think one of the biggest mistakes that companies make is they treat AI like an output machine, right? Like you give it the stuff and then you want the output.
What I actually like to approach AI with is more as an input machine first, and then you get the output that you want, which simply means first you create this context, you educate your AI, but you first use it to actually inform you, to educate you, to help you empathize with your customers, and to actually understand your landscape, your market, your competition better so that you can actually...
differentiate and stand out from the rest. Once you have that clear understanding, then you can actually structure your prompts and have AI actually produce work that's at least a pretty good first draft that then you can obviously revise, edit. But I would say that's probably one of the biggest mistakes, like treating it just like a plain chatbot that you have a conversation and you expect a perfect output right away.
Steffen Horst (04:14.317)
Yeah.
Steffen Horst (04:24.652)
Yeah, when you create this input, does that mean you set up specific AI agents for clients? Or when you just create a general prompt, you give it some more color, basically?
Chris (04:39.088)
Yeah, yeah. So first, obviously, everything starts from real research. So first, I conduct all the research, which means internal research, I understand my client's team, their point of view, the product, and then the external research, so prospect and customer, and then market research, so the competitors. Once I have a full picture, and all of our materials, transcripts, reports, then I literally feed that.
to the AI. Typically for me, I try different agents, but I think the simplest thing that you can do, start even in Chaggpt, even better if it's something like Gemini because of the bigger context window. And so it can actually make sense of the information much better. Or in alternative, I use another app, which is great for marketing and for collaborating in marketing teams, which is called TeamGpt. They actually...
Steffen Horst (05:19.47)
Mm-hmm.
Chris (05:33.308)
They're actually going to change their name tomorrow, think, because ChiaGBT sued them. It's really good app, especially to manage your context properly, to divide it into different projects and to have your own brand voice guidelines and to be able to collaborate on the same chats with teammates. I'm not an affiliate or anything, so it's just a purely recommendation because I use it for clients with my team.
So that's a very good one to get started. So yeah, start with the real data context. Then from there, what I've seen works best is rather than always share all the research, all the raw data, what I'm seeing work well is distill your research. So at the start, you start with the raw data.
As when you move to the strategy, you distill it into simpler documents, more condensed information. When it comes to the actual writing, you basically literally have a couple of your core documents, which are the positioning canvas, messaging framework, value proposition canvas. With those documents, the AI then can actually write really good copy because it's able to make sense of your context way better than if it was a mountain update. So it doesn't go...
Steffen Horst (06:51.256)
Yeah.
Chris (06:54.406)
hallucinate or hallucinate a bit less. And so yeah, you're able to write copy that's very close to what you actually need and you just need to the final polish.
Steffen Horst (07:03.99)
Yeah, so it's interesting that you said kind of finding a way to create almost like one basis when a team works across one client, right? Because everyone might use different information.
Chris (07:13.349)
Hmm.
Steffen Horst (07:17.55)
to build out this information input part. Some people might forget some things or not deem it as important. And then where they all start from is completely different, which means the output is going to be different too. From your perspective, if people don't use that app that you mentioned, how would you recommend a team and an organization to ensure that there's one basis, one foundation from which they start?
Chris (07:31.58)
Mmm.
Chris (07:47.524)
Yeah, it's definitely the foundation of the approach that any company should actually normally have when it comes to positioning, messaging and copy, which is to have your centralized documents and systems around positioning. So clearly defining what you do, who you do it for, how you do it better, differently, uniquely. Then how do you translate that into a messaging framework that goes over your narrative strategy, your sales pitch, your key messaging pillars.
And then how do you expand all of that and make it more specific for different ICPs into value propositions? And then from those, you can actually write a copy. So once a company has those simple documents, everyone in a team is actually able to even create their own project in ChudgePT and the output is going to be very similar, even though it's not a shared project or chat.
Steffen Horst (08:38.646)
Yeah, they just use the same documents, basically. Yeah, makes sense. Now, AI personas in synthetic research are becoming big passwords. How can teams use them to understand customers faster without turning messaging or messaging generic, messaging or replacing real customer insights?
Chris (08:40.132)
Yeah, exactly.
Chris (08:59.74)
So one mistake that I'm seeing a lot of marketing teams do is that they know that AI can actually help them simulate customer behavior because maybe they do quick tests. But what they do is typically they expect the AI, chat GPT or cloud based on a simple instruction to be able to simulate their customers. Like for example, someone says someone knows their ICP is a CFO.
They literally have a prompt, which can actually even be a mega prompt instructing the AI to be a CFO in their specific market for a specific industry and with specific psychographic and demographic profiles. But even then the AI is kind of guessing from all of its knowledge on the internet, right? So it's never going to be super specific. It's never going to be super relevant to what you actually do. So the key thing is...
Steffen Horst (09:42.574)
Mm-hmm.
Chris (09:56.252)
What I always do is create, if you create synthetic personas, always base them on real human data, which to me is lots of interview transcripts, competitor data, internal insights, like we talked about before, those three areas, internal, external, and market. Once you create your personas out of that specific and very relevant information, then the AI is much more able to actually give you responses that are
Obviously not hundred percent accurate and still biased, but at least when you're looking for directional validation, those can be 70, 80 % accurate, which I think it's super valuable because they're fast, they're cheap, than having to spend months conducting research done. then there's also, obviously there's lots of platforms that actually do it for you way better. Platforms that I typically use, it's called Rally.
Steffen Horst (10:41.326)
Yeah.
Chris (10:52.54)
And those platforms, what they do, they actually have built-in systems and algorithms that are able to map out your personas without all the typical biases or training instructions that the commercial AI models have, right? So those are way more accurate. So if you want to scale synthetic research and be able to use it at scale and with higher quality responses, definitely look into some of those synthetic research platforms.
Steffen Horst (11:22.764)
Yeah, let's stay with systems and tools. Now, what does a real AI assisted messaging system look like inside a B2B marketing team? And then as a second question, how do research messaging and copy generation actually connect?
Chris (11:31.142)
Yeah.
Chris (11:37.34)
So as far as the system, what I typically do is I create... The first one is different projects or different chats. So in my tool, the tool that I mentioned, I usually create different projects to start with. The first one is going to be your marketing assistant project, which is the one that kind of supports you in the decision-making process, in brainstorming.
in coming up with the synthesis and analysis of your research from the other raw data. And that's kind of the first bucket that you want to have. The second one is the Copywriter. And for that, I would have different folders if you can have folders or different projects for each of your channels, right? So the Copywriter takes in the strategy documents that you created with the other LLM and then is able to actually write first drafts or copy for you.
Then you can also have different other types of let's call agents or folders. You can have one for sales teams. You can have one for product teams that can actually strategize. But I think that the, of the best ways to organize this is first again, start with the raw knowledge and one general assistant, and then specialize as you move forward in the process with the finer information, more specific relevant information. And.
different chats and different separate projects for each of your tasks.
Steffen Horst (13:11.022)
Now with this AI team of experts, people might think, you know what, I can get rid of certain people. I don't need to hire certain people for specific roles. you give your thought on what does these AI-assisted agents replace and where is the human input still required to make it work properly?
Chris (13:32.956)
Yeah. Yeah, so I would say, so I have a framework that I call path framework, P-A-T-H. And the first one when you want to create this another persona is the preparation. So again, putting together all the research, creating your materials, your context. The second one is articulation. So instructing the AI to wear the shoes of your personas.
The third one is the T, so testing. So once you have all of that in place, you can actually start asking questions, seeing what objections your personas could have to any messaging or any points that you want to bring up, or even test different pieces of messaging, different variants. And one of the most critical steps is the final step, which I call harmonize the age, which basically means take all the findings that you've gotten with your AI work.
but then bring it back to your human research. So it's a continuous loop, right? So once you know what the AI gave you, then bring it back. When you have more customer interviews, try and validate what the AI told you and kind of never accept what the AI tells you as like absolute truth. I would always ground it in human insight first.
Steffen Horst (14:42.189)
Yeah.
Steffen Horst (14:47.278)
Mm-hmm.
Steffen Horst (14:51.136)
Interesting. What are the biggest red flags that a team that is using AI badly that you see? What are the biggest red flags?
Chris (15:02.938)
Yeah, probably as I mentioned, dumping everything in ChachiPT and expecting great results. That's probably the main one. Another one, asking ChachiPT or another LLM to sound like or respond or simulate the behavior of my persona without any sort of instruction and expecting ChachiPT to get data from Reddit on what your persona is without any specifics.
Steffen Horst (15:30.86)
Hmm.
Chris (15:32.144)
The other one, again, rely on the output without actually confirming it and validating it. What I always do, I also have another role in a messaging, AI messaging assistant system. I think it's super important. It's the editor. I basically have a prompt that I use whenever I'm writing copy inside one of those copywriting chats or projects.
Steffen Horst (15:58.265)
Mm-hmm.
Chris (15:58.434)
Once I have all the context, the copy has been written. I have this prompt which goes through literally 10 steps of editing sequentially, one after the other. But you, as the copywriter, are the one that at each step are able to look at the different variants that it proposes, the different edit recommendations, and make the choices. It's kind of like a choose your own adventure prompt in the sense that you are the one supervising the output. And with your knowledge of the audience,
Steffen Horst (16:08.6)
Mm-hmm.
Steffen Horst (16:22.936)
Mm-hmm.
Chris (16:28.176)
product, you're able to make those decisions. And I think that's super important because after a while, if you stop making decisions, your brain kind of atrophies, right? So you want to still make it work in a way as much as possible, even though you're obviously, you should be able to scale and compound the productivity with AI, but still you want to do to make some of those decisions. And I think it's an important step being able to edit and fine tune.
Steffen Horst (16:57.164)
Yeah, yeah. How do you course correct without derailing the momentum? Is that what you just talked about? That's what you apply or are there any other questions?
Chris (17:06.648)
Yeah, I would say it's that final step, the harmonization, which is a lot of times if you don't correct with real human data and you don't close the loop, then the AI starts injecting bias and you want to make sure that each loop you're course correcting. And that's what a lot of those platforms that do it for you already do it built in, which is super helpful.
But obviously, if you want to do it in-house, it's still possible to least get some directional accuracy.
Steffen Horst (17:39.341)
Yeah, yeah. Now, many SaaS teams want contractors and agencies to write on brand, but they struggle to scale messaging. How can AI help teams create brand-consistent copy across writers without losing nuances like voice and judgment?
Chris (18:00.38)
To me, that comes down to having a messaging framework. A lot of companies skip messaging framework. Maybe they know what they do. They can express their positioning in a way just because they have lot of conversation, especially if they're sales led. They have a lot of conversation with clients, customers. They are in touch with their market, but they are not able all the times to translate that into a messaging framework, which is what allows you to scale messaging, even with AI.
If I create a Positioning Canvas, which is typically one page, super simple, but at least it defines the differentiation. And then I feed it into ChudgePT. ChudgePT is probably going to guess and assume a lot of the copy variants, all of the angles that you are going to use in your copy. And so it's a lot based on assumptions again. So you want to make it clear and crystallize.
What your strategic narrative? What's your point of view? What's your, how is your sales pitch structured? What are your key messaging pillars that you're going to use both in conversion copy, but also in your LinkedIn content, for example. What's your brand voice guidelines? So all of those are kind of like Lego blocks that help you scale messaging, both as a human, but also as a, with AI, because once you feed it the right stuff, then you get the right output.
Steffen Horst (19:22.138)
It sounds like all comes back in the end to kind of the setup, right? Kind of the input. The more time you spend in the beginning on providing the LLM with information about you, company, as precise as possible, the better your output in the end is going to be. And then the more on end it's going to be. The more room you leave, the...
Chris (19:26.074)
Yeah, exactly. Yeah.
Chris (19:40.572)
Exactly. Yeah.
Steffen Horst (19:45.474)
the more you're lax in kind of describing and providing information, the wider the area of error is going to be in the output.
Chris (19:49.551)
Yeah.
Chris (19:54.04)
Yeah, I literally have had a conversation today with a potential client when at the proposal stage, they literally asked me like, why do we need to do customer interviews? Like, can we not just skip them and do I don't know, look at reviews or other stuff competitors? And I literally told them like, this is the most important part of everything. Like if we do any type of research, that's customer interviews, even if you do five, at least.
That's basically the human element and the reaction, the live feedback that you need to actually make the copy that sounds human, even using AI. It's all going through that. It's 70 % of my work research.
Steffen Horst (20:34.222)
Yeah, and I think a lot of companies miss out on that element. They just think, we know our customer, you know, and this is how a customer looks like. But in reality, you know, first of all, there are so many sub segments to customer a company has, right? And they all need to be addressed in a different way. They all need to be talked to in a different way. They all respond to different things. So getting a clear understanding of what that really looks like outside of your company bubble.
Chris (20:39.522)
Mm-hmm. Yeah.
Chris (21:02.492)
Mm.
Steffen Horst (21:02.606)
I think this is important, not only for this, it's in general important.
Chris (21:06.624)
Yeah, yeah. And you bring up a great point as far as perspective, because you said a lot of companies think that in other customers. And one of the biggest benefits of working with someone like myself is typically not a lot, not 100 % about the strategy, the writing, but it's also about the external outsider perspective that I bring in when I speak with customers, when I look at the product, when I speak with our teams. And so another
thing going back to AI as an input machine, even if you don't hire someone like myself and you do it internally, using AI to actually look at your interview transcript and give it your objective perspective, right? It's very different from what you might know as someone who works at the company. So that alone itself, I think it's super valuable to get that outsider or AI perspective from your data.
as long as it's real human data.
Steffen Horst (22:04.898)
Yeah. Now, looking ahead, how will synthetic research, personas, messaging, testing shape the next generation of GDM teams?
Chris (22:17.222)
So I definitely think because I spoke with a lot of those, some founders of those synthetic research platforms, a lot of the direction that they're taking is to make synthetic research and testing as like a constant and regular component of go-to-market. So a lot of those platforms are starting to implement workflows with APIs that...
literally constantly are able to test messaging, maybe even add creative on the go automatically so that as you are putting out stuff, you're able to test with your synthetic personas continually, which is what has been, I think has been missing a lot because a lot of companies treat the research as a one-off project and never did it again. So I think this is going to bring a lot of value to actually always have fresh...
Steffen Horst (23:05.517)
Yeah.
Chris (23:09.186)
new data and validating insights even before you launch your copy. So I think that's probably one of the biggest benefits of this. And yeah, in general, a better knowledge and being able to better serve your customers.
Steffen Horst (23:24.47)
Yeah, and I mean, in the end, you're setting something up that can be continuously used, right? I mean, I'm not saying don't continue feeding it with new information. And I think that was one of the sentiments you had throughout the conversations, like you need to continuously enrich the LLM and then provide more context as you create more information, right? It makes it more precise, will create an even better output.
Chris (23:34.353)
Yeah.
Chris (23:40.176)
Yeah, the context.
Chris (23:45.104)
Mm-hmm.
Steffen Horst (23:49.538)
But technically it is there, the setup is there, just keep using it and the output is created in a relatively short amount of time, compared to if you have a human being doing all the research and trying to get that information.
Chris (23:59.814)
Yeah.
Chris (24:04.324)
Yeah, yeah, I think probably one of the biggest advantages of companies would be who whoever understands at which touch points to install kind of mechanisms or feedback collection or data collection mechanisms, like where do you have a survey maybe that start collecting data, then then you automatically send to the like, is it after a sign up or in your emails or whatever, like whoever understands where to
Steffen Horst (24:07.95)
awesome
Chris (24:32.198)
filling the gaps with data collection and AI, I think it's going to win.
Steffen Horst (24:36.854)
Yeah, yeah. Well, Chris, thank you so much for joining me the Performance Delivered podcast and for breaking down how SaaS teams can use AI not as a shortcut, but as a multiplier for clear customer-driven messaging. For listeners who want to learn more about your work, your frameworks, or explore whether conversion alchemy is the right fit, where should they get in touch with you?
Chris (24:59.324)
Yes, so you can find me on my website, conversionalchemy.net, where I have a newsletter you can sign up. And also LinkedIn, I'm mostly on LinkedIn, posting almost every day and just having a chat with lots of lovely people.
Steffen Horst (25:13.144)
Great. Well, thanks everyone for listening. If you enjoyed this episode of Performance Delivered, Insider Secrets to Marketing Success, please subscribe and leave us a review on iTunes or your favorite podcast application. To learn more about Symphonic Digital and how we help brands unlock performance through full funnel digital marketing, visit SymphonicDigital.com or follow us on X at Symphonic HQ. Thanks again and see you next time.
.jpg)
.png)
