How we use AI, and why it’s the least interesting part of our process

Sooner or later, it comes up on every call with a new client.

Sometimes it’s asked outright. Sometimes it’s slipped in near the end, after timelines and before goodbyes: “So...do you guys use AI?”

It’s a reasonable thing to ask. We’ve all read the content they’re worried about: smooth, upbeat, grammatically perfect, and completely empty. The writing equivalent of a 2-star hotel lobby. If you’re paying an agency to speak for your brand, you want to know you aren’t paying for that.

But we’d suggest the question is slightly off target. The better one is: “How do you make sure what I get is good?”

Our view is that where the raw material comes from matters less than what happens to it afterward. An interview transcript isn’t an article. Neither is a folder of research notes, and neither is a paragraph a chatbot produced in four seconds. They’re all inputs. What we’re accountable for is the output.

That includes getting things right, which is nothing new for us. Checking facts has always been part of the job. Many of us learned it in newsrooms, where a wrong number or a misattributed quote meant a correction in print and an awkward conversation with an editor. Every claim, figure, and quote in a Campfire piece gets verified by a person, and it always has, whether it came from an interview, a report, or a client’s sales deck.

AI simply raises the stakes. Generative tools can be confidently, fluently wrong. They’ll produce a statistic that doesn’t exist and attribute it to a study nobody ran, in a sentence so polished you want to believe it. That doesn’t change what we do. It just means we keep doing it, every time.

‍

Our standard

Everything that lands in the client’s inbox has our fingerprints all over it.

Not a prompt at the start and a quick tidy at the end, with an LLM doing the interesting part in between. Editorial judgment shows up at every stage: in deciding what the piece is really arguing, who it’s for, what goes first, what gets cut (usually more than you’d think), and how the reader should feel and do when they reach the last line.

To be clear about what this isn’t: we haven’t banned AI, and we aren’t hiding it. It’s a promise about quality and intention. Whatever tools were open along the way, the finished piece is the result of deliberate choices by people who took the time to understand your business and the people you’re writing for.

What we rely on is judgment, the kind that takes years to build. Our team has spent a long time writing for B2B SaaS companies, and many of us started out as journalists. That combination gives you strong instincts for what makes a story, what a reader actually needs, and what won’t survive someone asking “says who?” or “so what?”

‍

The “zero draft”

When our writers use AI to generate text, we have a name for what comes out. We call it a zero draft. It’s not a first draft and definitely not a final one. It’s lower than both: rough material to react to and pull apart.

The distinction matters to us more than it might sound. A real draft reflects decisions about argument, structure, emphasis and voice. A zero draft hasn’t made those decisions, or has made them the way a committee would: safely, generically, and without much conviction. Sometimes its most useful contribution is showing us exactly what we don’t want to say.

When AI does save us some time on the mechanics, we don’t pocket it. It goes back into the work that needs a person: chasing a better example, asking one more follow-up question in an interview, or editing a paragraph until it says exactly what it needs to say, really well.

‍

One standard, many workflows

Across the team, we have very different working styles.

One writer works line by line. She takes rough AI output and goes through it sentence by sentence, questioning and rewriting as she goes. By the end, her draft often looks nothing like where it started. The zero draft was something to fight with, and she always wins.

Another starts with the source material. She rarely bothers with a zero draft unless there’s something substantial to feed it, like interview transcripts or original data. Without that, she finds generated text more distracting than useful. When she does reach for AI, it’s at a much smaller scale: testing a transition, tightening a sentence that’s drooping.

Sometimes the job is synthesis rather than writing. On projects built from several long interviews, we’ll use AI to sort and cluster ideas across the transcripts, so patterns and tensions surface faster. 

Very different processes. What they share is intention.

‍

Showing our work

There’s one more thing we do that, frankly, says more than any policy could. We leave our reasoning in the draft.

Open a Campfire draft and you’ll often find comments in the margins: why we went with this hook, why this structure, and why we cut the section you were probably expecting.

Clients notice and some have told us how much they valued those notes. They made the thinking behind each piece easy to follow, and just as importantly, easy to push back on, which is the point. Every piece of writing is a series of choices. When you can see the choices, you can see that someone made them, intentionally, with you in mind.

‍

This was never really about AI

When we sat down as a team to write an AI policy, the conversation kept drifting away from AI.

Eventually someone said it: “We’re not talking about AI. We’re talking about culture.” The tools will keep changing, probably faster than any policy can keep up with. 

What we want to stay fixed is how we work:

• High standards on every piece, whatever its size. A LinkedIn post gets the same care as a 3,000-word guide.
• Anticipating what you need before you ask. The best round of feedback is a short one, because we saw your questions coming.
• Doing the things you wish we’d done the first time, like an extra source or a headline option, and definitely the note flagging a claim legal might want to see.

When new writers join us, that’s what we spend our time on: what we expect and what good looks like here. Which tools they prefer comes up eventually but it’s the least interesting part of the conversation.

‍

What you can count on

So, do we use AI? Sometimes, in some ways, for some things.

Here’s what matters more:

• A person makes the editorial decisions on every piece.
• Nothing goes out that reads like a one-shot prompt.
• Our reasoning is visible in the draft when it’s useful.
• The work sounds like you. Not like us, and not like a machine.

If you’d like to know more about how we work, or you’ve got a project in mind, get in touch. We’re happy to talk process. We just suspect we’ll spend most of that conversation talking about your goals.

‍