Cassandra Mellen
During a recent episode of Pitch, Please, Faith Cedela and Cassandra Morse shared hard-won insights on one of the most misunderstood parts of working with AI today: how to write prompts that actually produce something usable. The discussion highlighted a simple but inconvenient truth most people don't want to hear. AI isn't the problem. Bad prompts are.
"Your output is only as good as your input," said Cassandra. "You can't just put something in there and say, hey, I want an awesome post on supply chain. It's all going to sound the same."
Faith and Cassandra covered why generic prompts create generic content, the biggest mistakes people make when using AI, how to write prompts that generate better articles, social posts and pitches, and why context matters more than most people realize. Whether you're a marketer, PR professional, business owner or someone who's just been quietly hoping AI would figure it out on its own, this one's for you.
The Biggest Mistake People Make with AI (And It's Not What You Think)
Here's the scenario: you open ChatGPT, type something like "write me a LinkedIn post about leadership," and receive something that reads like a motivational poster had a LinkedIn account. You post it anyway because the deadline was twenty minutes ago. We've all been there.
But here's the thing nobody really wants to sit with: when the output is vague and forgettable, it's not the AI's fault. It's yours.
"The number one issue is vague instructions," Cassandra said. "AI doesn't know how to pick up on context. You have to give it the information."
AI doesn't have opinions. It doesn't have experience. It cannot tell that your client would never, under any circumstances, use the phrase "synergistic growth opportunities." You have to tell it that. You have to tell it everything. The moment you expect expert-level results from minimal input, you've already lost.
The second mistake people make is assuming the first output is the final output. AI drafts need editing. They need human review. They need someone with actual subject matter expertise to look at them and say "yes, this is right" or "no, this is completely wrong and also weirdly confident about it." Skipping that step is how content that sounds like it was written by a polite robot ends up on your client's website.
Why AI Content All Sounds the Same
If you've noticed that a lot of AI-generated content feels interchangeable right now, you're not imagining it. The reason is simple: without real voice data, AI defaults to the same generic middle ground every single time.
"If you don't have background information, your content is going to sound the same," Cassandra explained. "I've taken meeting transcripts of clients talking, pulled their LinkedIn content, created knowledge documents for each of them. When they use the agent, they tell it who they are, what kind of post they need, and it spits something out in their voice, their tone, in an area they're actually an expert in."
This is the difference between AI-assisted content and AI-generated slop. When you feed the tool real material, including transcripts, approved blogs, writing samples, LinkedIn posts, even a download from a recorded session, it can pattern-match against something authentic. Without that, it's just doing its best impression of what it thinks a business person sounds like. And we've all read enough of those posts to know how that goes.
What Every Good Prompt Actually Needs
Strong prompts aren't complicated. They just require a few things that most people skip because they're in a hurry.
Every prompt should include who the content is for, what the goal is, what voice or tone to use, any source material you can provide, and the format you want at the end. That's the whole framework. The more specific you get on each point, the better your output.
Faith walks through her pitching process this way: "I go into chat and immediately I say, this is for my client, here's who they are, and I copy their LinkedIn profile or something from their bio. Then I take it a step further and copy one of their approved blogs. Sometimes I'll download a transcript from one of our live LinkedIn sessions with the client and put that in too. Then I say, answer this pitch using my client's voice. And we actually get a pretty solid pitch."
Platform specifics matter too. What works on LinkedIn in terms of length and format is completely different from Instagram or a media pitch. Telling AI which platform you're writing for, what character limits apply and whether you want links included isn't overthinking. It's just the information it needs to do the job correctly.
Real Prompt Examples: What Fails and What Works
Here are side-by-side comparisons pulled directly from the conversation, because seeing it is more useful than describing it.
For a social post:
Bad prompt: "Write me a LinkedIn post about leadership."
Better prompt: "Write a 250-word LinkedIn post for a manufacturing CEO about a leadership lesson she learned during a supply chain disruption. Make it bold and slightly controversial. Here's a transcript from a recent interview with her to match her voice and tone. Do not add any hashtags."
For a media pitch:
Bad prompt: "Write a pitch for a journalist about my client's supply chain expertise."
Better prompt: "Write a 150-word pitch responding to this journalist query [paste query]. My client is [name], a manufacturing CEO with 15 years of experience in [specialty]. Here is her LinkedIn bio [paste] and an approved article she wrote [paste]. Match her voice. The journalist wants a response in three to five bullet points from an expert source. Keep it conversational, not corporate."
For a blog post:
Bad prompt: "Write a blog about AI in PR."
Better prompt: "Write an 800-word blog post for a PR agency's website targeting communications professionals. The topic is how to use AI tools without losing your brand voice. Tone should be direct, slightly irreverent, practitioner-focused. Avoid bullet-heavy formatting. No em dashes. Here are two approved articles from this site to match the style [paste links or text]."
The difference between these isn't effort. It's about thirty extra seconds of specificity that saves you twenty minutes of editing on the back end.
Can AI Replace Expertise? No, and Stop Asking
This needs to be said clearly: AI is a tool, not a strategist. It doesn't have experience. It cannot replace the kind of thought leadership that comes from someone who has actually lived through the thing they're writing about.
"We still need the human touch," Cassandra said. "Even if the pitch isn't exactly the same topic, AI can sense their tone and how they talk. But giving it those pieces, the transcript, the approved blog, the real voice? That alone is gold."
The PR teams doing this well right now are combining both. They're capturing real voice through live conversations, recorded sessions and actual writing samples, then using AI to process and produce first drafts that already sound like the person. The approval process moves faster because clients recognize themselves in the content. Pitches land better because they include genuine perspective instead of AI-assembled filler.
The Real Risk: Losing Your Voice Entirely
Here's what's actually at stake if you over-rely on AI without putting your own expertise back in. Your content becomes indistinguishable from everyone else's content. It looks the same. It sounds the same. Readers scroll past it because there's nothing in it that required a human to actually think or experience anything.
"I have seen so many filler articles out there that all look the same," Faith said. "I don't even bother reading them because I know it was not written by someone. It didn't come from someone's expertise. It was clearly just a copy-and-paste prompt. And I think that could damage business."
For PR professionals, this is especially high stakes. A pitch that sounds like a bot wrote it goes straight to the bottom of a journalist's inbox. A pitch that sounds like a real expert with a specific point of view on a relevant topic is the one that gets picked up. The content living on your clients' sites is building or eroding their credibility every time someone reads it.
This is why thought leadership still matters more than ever, and why getting executives to show up on record, whether through LinkedIn content, live sessions or interviews, gives you material that AI genuinely cannot manufacture.
How to Use AI Without Sounding Like a Robot
Faith and Cassandra have built a process at Winsome around exactly this. Monthly live LinkedIn sessions with executive clients, recorded and transcribed, give them a bank of real quotes, real opinions and real voice to work from. AI then helps turn that raw material into polished content that the client actually recognizes as something they would say.
"We push our executives to do live LinkedIn at least once a month," Faith explained. "We get actual quotes from them, we can pull blogs from it and use that for pitching because this is actually their advice, their thoughts on a topic. This is not AI. And we are giving quality pitches that have a better chance of getting picked up."
The workflow matters. It's not just dropping a transcript into ChatGPT and hitting send. It's telling the tool how to use that information, what format you want, who the audience is and what the goal of the content is. Every single time.
Tools like ChatGPT can absolutely match voice and tone when they're trained on real examples rather than asked to invent a personality from scratch. The technology is capable. The gap is always on the prompting side.
One AI Tip to Start Using Today
Two actually, because neither Faith nor Cassandra could pick just one.
Cassandra's tip: "Once you have an output, ask for feedback, but tell it not to just please you. Ask it to stress test the answer. Ask how confident it is in what it's giving you. A lot of times it'll hallucinate, and this will make it go back on track."
Asking AI "what's missing from this?" or "where could this be wrong?" after you get a draft is one of the most underused prompting moves out there. It forces a second pass and catches the overconfident nonsense before it ends up in a client email.
Faith's tip: "Give AI your client's actual voice with real examples. Don't just say make me a leadership post for LinkedIn. Give it words from your client. That's where people are missing the mark."
When the content already sounds like the person, the approval process gets faster, the edits get lighter and the final product is something worth actually publishing. That's what AI is supposed to do. Not replace the thinking, but make the whole process more efficient for everyone involved.
AI works best as a collaborator with clear instructions and real material to work with. Give it context, give it voice, give it direction and you'll get content worth editing. Skip those steps and you'll keep getting very enthusiastic posts about leadership that sound like absolutely no one you've ever met.
Need help building an AI-assisted content process that doesn't make everyone sound the same? Let's talk.