Stop Using AI to Write Like a Marketer. Use It to Think Like Your Customer.

The blog nobody reads

The blogs I read for work are not bad. That is what's wrong with them. A technical founder opens ChatGPT, spends twenty minutes, and ships an 800-word post called "The Ultimate Guide to API Rate Limiting." Clean, structured, an intro and three subheadings. Almost nobody reads it, because nobody was searching for it. It's the kind of title a marketer writes to sound authoritative on a topic, not the sentence a real buyer typed into Google at 11pm while debugging a production issue.

That's the tell. The post answers a category, not a question. And buyers don't arrive with categories — they arrive with problems, half-formed and specific, usually already 74% through their own research before they'll even talk to you.

AI didn't cause this failure. It just made it cheap to produce at scale. The real opportunity isn't using AI to sound like a marketer explaining a topic. It's using it to simulate the actual person googling "why is my webhook silently failing" at midnight, and writing to that.

Your buyer already did the research without you

Most companies still optimize their website for a buyer who doesn't exist anymore—someone who's still deciding, who needs you to educate them, who will bend to your narrative. But that's not what's happening.

By the time a prospect reaches out, they've already made the core decision. They've read ten, fifteen pieces of content. They've sketched out what they need. They know roughly who's in the running. The sales conversation, when it happens, isn't where the choice gets made. It's where the choice gets confirmed.

You can see this in what buyers actually say they want. Most would rather never talk to a salesperson at all. When they do get cold outreach, they punish mismatches—wrong angle, wrong person, wrong moment—by ghosting. So the old playbook, where a warm intro gets you a chance to "educate" the prospect, assumes facts not in evidence.

Here's where it gets harder: the person doing the research and the person writing the check are often different people. Sometimes they're very different people. The technical person needs proof it works. The budget owner needs to know it won't hemorrhage money. The executive above them wants to know why this matters for the business. They're not having a debate in your sales calls. They're each reading different articles, different case studies, different pricing pages, forming separate opinions. One content strategy has to do the work of three conversations that will never happen.

So your website isn't really a sales tool anymore. It's a proxy for all the conversations your buyer is having without you.

So point the model at the question, not the copy

Here's the tell. Ask an AI to "write a blog post about why our API monitoring tool is better than Datadog," and you get five hundred words of confident nothing — bullet points, a table, a call to action. It reads like copy because you asked for copy.

Ask it something different: "You're a VP of Engineering at a 40-person startup, evaluating monitoring tools, with a $15k annual budget. What do you Google first? What three objections stop you from buying? What proof would actually change your mind?" Now the model has to reconstruct a decision process instead of decorating a conclusion. It'll tell you the VP is Googling "Datadog alternatives cheaper," worrying about migration cost and vendor lock-in, and wants to see a real incident dashboard, not a pricing page.

That output isn't the essay. It's the map. And it works because a language model is a decent simulator of common reasoning patterns — it's seen thousands of forum posts, reviews, and Slack threads from people exactly like your buyer. You haven't. You've been staring at your own product for fourteen months, which is precisely what makes you the worst person in the room to guess what confuses a stranger. With 72% of B2B buyers starting their research online before ever talking to a vendor, the thing worth optimizing isn't your prose. It's whether you understood the question before you answered it.

The authenticity objection

Here's the real objection: if people already distrust AI, won't using it in marketing feel like you're tricking them? The worry isn't abstract. Qualtrics found consumer comfort with AI-generated marketing dropped from 57% in 2023 to 46% in 2024. Most marketers I've talked to say authenticity is their biggest headache with generative AI. So it's worth asking whether the tool defeats the purpose.

But the thing is, I don't think those consumers are wrong about what bothers them. What erodes trust isn't that something was AI-written. It's that you can tell. A piece optimized for keywords instead of people. Hollow reassurance masquerading as insight. The same person who ignores an AI-written ad will read a founder's blog post three times if it actually says something true.

So the fix isn't to ban the tool. It's to hide it. Use AI where readers never see it—to organize messy notes, to draft the first pass on something you'll rewrite completely, to test a premise. But the part that leaves your mouth should sound like it came from your brain. That's not a purity requirement. It's just practical. People can tell the difference. They prefer the version where you didn't try to hide the fact that you used help.

Thinking is the part that doesn't scale

Here's the thing that took me too long to see. I kept treating AI as a words machine, because words used to be the bottleneck. They aren't anymore. Anyone can produce a thousand competent sentences before lunch. The bottleneck moved upstream, to the question the words are supposed to answer, and that question lives entirely inside your buyer's head, not yours.

Buyers already did the work of moving on without us. They review something like 11 pieces of content and settle 57% of the journey before a sales conversation starts, according to that 2024 survey. They're not waiting for our post. They're checking whether we understood their problem before they showed up.

So the founders who win won't be the ones who publish the most. They'll be the ones who use the model to sit in the buyer's chair, ask the uncomfortable question, and answer it straight, even when the honest answer isn't flattering. That's the part that still requires a person.

Sounding like a company is free now. Being understood by a customer never was.