What I Learned Writing for a CEO

Two LinkedIn posts. 40,000 impressions each.

Comments, likes and reposts from the UK AI Safety Institute, the European Commission's AI Office, IBM researchers, policy scholars from Stanford and Harvard.

I wrote both of them.

I ran the thought leadership operation for Kush Wadhwa, founder of Trilateral Research, a Responsible AI company. My task was to position him as a voice in responsible AI. In practice, that meant I owned the research, the angles, and the full drafts. Kush shaped the conclusions (pushing back on framing, asking "how will a CTO actually interpret this?"), and handled line edits.

Here's what that taught me.

How the Best Post Came Together

It's easier to talk about problems people are already experiencing, than to get them invested in something new.

The AI governance post started with my own confusion. I was trying to parse sources from different sectors—regulators, researchers, vendors—and kept hitting the same problem.

Everyone used different terms for the same concepts. "Explainability" to the engineers, "transparency" to legal, "interpretability" to the regulators.

Before drafting any thought leadership posts, I made it a habit to browse the latest research, specifically the latest Arxiv posts.

When I came across the IBM AI Risk Atlas, I saw it mapped the whole terminology problem I had been struggling with. My angle was born: AI governance doesn't have a technology problem. It has a language problem.

Kush pushed on the conclusion. My first draft was too academic. He asked what a CEO reading this at 7am would actually do with it. We landed on: "We can't fix AI governance's language problem, but we can start by getting our own teams speaking the same language first."

The post hit 487 reactions and 60 reposts. But what mattered more was who engaged—PhDs, IBM researchers, policy experts, CTOs. The exact audience we wanted to remember the name "Kush Wadhwa".

What Made It Work

Both viral posts followed the same structure: counterintuitive hook, research tie-in, numbered takeaways, natural company connection, actionable close.

But structure wasn't the differentiator. Research was.

Most LinkedIn content is opinion. These posts had something else: academic papers and recent studies most people hadn't seen. For the AI governance post, I pulled from the IBM AI Risk Atlas and regulatory research on terminology gaps. For a post reframing MIT's "95% of AI projects show no ROI" statistic, I dug into the Productivity J-curve and the Solow Paradox.

This took time—half a day to a full day per post, mostly research. But it's what made the content accessible. Anyone could engage, not just specialists. That was the difference between these posts and the ones that didn't break through.

The Constraint That Shaped Everything

Trilateral works across public and private sectors. That meant I couldn't call out specific companies or individuals, even when it would've made for a sharper hook. The content had to be edgy enough to stop the scroll, but appropriate enough that a government client wouldn't flinch.

That tension actually helped. It forced me toward structural arguments instead of hot takes. "AI governance has a language problem" works because it names a system-level issue, not a villain.

The Non-Obvious Lesson

Here's what I'd tell anyone ghostwriting for an executive: define the relationship before you start writing.

It's easy to define logistics—what do you send, when do you send it. But the creative latitude—what to do if you disagree on a take or headline—is much more fraught. If you know a headline performs better than their suggested revision, can you say so? Do you have standing to push back?

Executives are equipped with intuition. I know Kush's edits made my work sharper. But sitting in the ambiguity of not knowing who has the final say on the headline builds friction that compounds over time. When you own the strategy, you need space to execute it. Clarify that upfront.

What I'd Do Differently

The two successful posts ran back-to-back. LinkedIn's algorithm rewarded that consistency. After the first hit, the second had tailwind.

Looking back, I should have protected that momentum more aggressively. Kush's LinkedIn had become something with real pull—an audience that showed up for a certain type of post. Same voice, same structure, same quality, same rhythm. When those conditions shift, the algorithm notices before you do.

The framework worked. The challenge is keeping everything else from getting in its way.