Don’t Write for AI. Write So AI Can’t Misunderstand You.

There's a growing idea in communications circles that AI should be treated as a stakeholder. The argument goes: if AI systems don't pick up or correctly interpret your content, your reach suffers. So you'd better optimize for them.

From my perspective the observation is right. The conclusion isn’t. That AI affects how content travels doesn’t make it a party with interest, any more than Google’s algorithm made PageRank a stakeholder. Something can shape outcomes without having a stake in them. 

But dismissing the observation entirely would also be a mistake. The information environment is changing in ways that matter for anyone trying to build a narrative, position a company, or get a message to stick. For startups and VCs especially, who often get one shot at defining how they’re perceived, these changes are worth understanding clearly.

The ground has shifted: zero-click search and AI gatekeepers

58% of Google searches now end without a click. AI overviews answer the question before your link appears. A few years ago, the main question in communications was: how do we get journalists and platforms to carry our story? That question hasn't gone away, but it's not enough anymore.

Search engines and AI tools increasingly provide direct answers instead of links. And a growing share of what circulates online, including likes, shares, and comments, is generated by automated systems rather than real people. The public information space is noisier, more synthetic, and harder to read than it was five years ago.

On top of this, there are now new gatekeepers. LLMs and AI-powered search tools don't just deliver your message; they interpret it, summarize it, and pass it on, often without attribution. A potential investor, customer, or partner might form an impression of your company through an AI-generated answer before they ever visit your website. What that answer says depends entirely on how clearly and consistently your positioning is represented across everything you've published.

The double compression problem

Here’s what I think most people miss: your message is getting compressed twice.

On the production side, as more companies use AI to draft content, messaging converging toward the same register. LLM-assisted writing produces ideas that are semantically less distinct from each other. Everyone starts sounding the same.

On the reception side, AI tools summarize, shorten, and reinterpret your content before it reaches your audience. Weak narratives get diluted. Generic positioning becomes invisible.

And there’s something happening on the audience that makes this worse. There's a Microsoft study (presented at CHI 2025) showing that higher trust in generative AI correlates with less critical thinking. So if your audience is already processing less deeply, and AI is compressing your message before it reaches them, weak positioning gets diluted twice over. 

I see this play out often. A fund that describes itself as "backing the builders of the next economy" might not surface at all when an LP asks an AI tool about climate tech investors. But a fund clearly and consistently associated with early-stage energy storage and grid infrastructure will. The margin for vagueness is shrinking from both ends.

What actually needs to change

The goal isn't to optimize for AI. It's to be impossible to misread.

There's a growing field called generative engine optimization (GEO) (if you are woking in the communications/marketing space, you might have heard that a lot by now) that focuses on tactical stuff like schema markup, structured data, and keyword placement to increase visibility in AI-generated answers. Search interest in it has grown sharply, and I get why. But for communications professionals, I think the frame is wrong. GEO optimizes the signals machines read. It doesn't change what those machines say about you once they've read it. That's not a tooling problem. It's a clarity problem. And clarity has always been the work of communications, not search optimization.

Write to survive compression

If your story only holds up when told in full, in your own words, in the right context, it's not strong enough. "We're reimagining the future of work" compresses to nothing. "We help remote-first companies manage contractor payments across 50+ countries without the compliance overhead" survives being cut in half. The question worth asking is not just "does this sound good?" it's "does this still mean something when someone else is doing the telling?"

Be specific enough to be wrong

If your positioning is so broad that no one could argue with it, it's also too broad to be remembered, by humans or machines. When a general counsel asks Perplexity which AI tools help law firms with contract review, a startup that describes itself as "transforming the future of legal work" is invisible to that query. A company that clearly states it helps litigation teams extract and compare clauses across large contract volumes is just easier to find. Specificity is a risk, sure. But vagueness is a bigger one.

Make it machine-readable and human-understandable

Precise language over buzzwords. Clear structure. Unambiguous claims. Your content needs to work for a procurement lead querying for compliance tools and a journalist backgrounding a story. This is where visibility in AI-generated answers actually starts. With saying what you mean.

The human layer is where it still matters

AI might surface you. But it doesn't close anything.

The decision to invest in a startup, to sign a contract, to recommend a product, that still happens in a conversation, a meeting, a referral. What moves someone to act is the feeling that this company sees the world a certain way and isn't afraid to say so. You can't optimize for your way to that. You have to actually mean it. And that's precisely the part of this work that isn't going anywhere.