In a recent LinkedIn Live, Muck Rack’s VP of Data & Intelligence Matt Dzugan and Zeno Group’s Global Head of Analytics Michael Brito teamed up to dig into what vibe coding is and what it actually looks like in practice.
Attendees asked questions about how to get started with vibe coding, and Matt and Michael answered them all.
If you missed the LinkedIn Live, catch it on-demand here.
Q: In PR and comms, where and how do you draw the line between AI-generated direction and human-verified accuracy?
Unsurprisingly, both Matt and Michael agreed that a human layer of verification is almost always necessary when using AI for PR and communications work.
“It's extremely important,” Michael said. “Because there's a lot of AI slop out there—it's all over LinkedIn.”
Michael mentioned it’s extremely obvious to him when he receives an email or sees a post on LinkedIn that’s totally AI-generated. Matt added that there are some serious tells that someone is using AI: a phrase like “it’s not this, but that” or overusing the em dash.
At Zeno Group, Michael and his team have a governance model they follow that outlines how they use AI and when/how they verify information.
A best practice Matt likes to follow is if you’re asking your chatbot to do any kind of subjective work where it offers suggestions, it’s probably a good idea to have a human look it over with a fine-tooth comb.
Q: What are some downsides and pitfalls to be aware of with vibe coding?
A major pitfall is cost, and more specifically, future cost.
Vibe coding is great—and can be done using free tools—but Matt shared that the tools get better and more useful with paid accounts.
“As these models get better and better, it costs more money per what's called a token,” Matt explained. “A token is essentially just a word that goes in or out of the model, and even as it's thinking, and it's maybe not showing you any output, but it's thinking behind the scenes, that's all [using] tokens.”
A big industry conversation at the moment is about how expensive these tools are possibly going to become.
“Something to be aware of is make sure you know how to analyze within the admin panel of any of these tools,” Matt said, to keep an eye on how much building an app or tool might cost you.
In addition to cost, Michael mentioned another pitfall to be aware of is bias.
“There's still a bias of the engines,” he said. “The models were developed by people, and those people have bias, and the models have bias.”
Tiny changes in wording or phrasing can trigger different biases, so be aware of that, and certainly run outputs through a human layer before running with it or sharing with a client.
Q: How should PR and comms teams deal with low‑quality or AI‑generated “slop” sites?
It’s not a secret; there’s plenty of AI slop out there on the web, but is it actually breaking through the LLMs?
The good news, according to Matt, is that they really aren’t.
“We don't see those kinds of bad acting content producers show up in AI citations much. Fortunately, those sites, although they do exist, don't seem to get a lot of pickup,” he explained.
This is something that Muck Rack monitors through its Generative Pulse tool.
Muck Rack is passionate about verifying the media outlets and journalists that appear most across the various LLMs i.e. how visible various outlets are to the AI systems.
Muck Rack’s new AI visibility badges, shown as small icons throughout Muck Rack, highlight the outlets and journalists that appear most frequently as cited sources in AI-generated responses.
“The worst thing to possibly ever do would be to claim that somebody—a real human writing an article—would be to inadvertently claim that the content was AI-generated,” Matt said.
Behind the scenes, Matt shared that Muck Rack is aiming to categorize and systematically look at the content where outlets disclose AI use because many outlets are combining AI and human writing in published works.
Q: Beyond productivity hacks, what are some use cases for PR and comms pros using AI?
Here are a few ways Michael shared he’s using AI in his comms work.
- To identify new media outlets. Michael likes to go beyond his usual targets to find outlets they may not have considered before or outlets that are currently highly influential in certain models.
- To identify stakeholders. Michael mentioned some tools already do this, but if you’re looking to identify stakeholders in different areas like for crisis comms, executive comms, or in academia or NGOs, AI can often give you a solid starting point to build on.
Similarly, Matt has recently become interested in using AI for “synthetic audiences,” AI-simulated digital twins of real market segments. Matt said PR pros can create a prompt describing in detail a person or demographic and then ask AI to analyze how they might respond or react to a specific type of message.
“Is it perfect? No,” Matt said. “But it’s cheaper than a focus group.”
Michael added that there’s a lot of free data out there, too, to help with audience research.
“Do a search in Perplexity or Claude for something like ‘GenZ audience behaviors, file type: PDF,’” he suggested.
The output will be a ton of research you can review and download and then upload the PDFs or findings into a project.
“There's a lot of really innovative things,” Michael said. “It's not perfect, to Matt's point, but it's a starting point, especially for somebody who doesn't have a big team or budget.”
Q: What tools should I use for vibe coding?
This is really up to you; there are many different tools you can use, including ChatGPT, Claude, Perplexity and more.
In the example Michael shared during the LinkedIn Live, he mentioned that he prefers Perplexity.
“I prefer the Perplexity Comet browser,” he said. “You can do the same thing in ChatGPT and Claude.” However, Michael noted he pays a license fee of $20/month to access various parts of the tool.
For more tips and insights from Matt and Michael, check out the LinkedIn Live.
