AI can write the post. It can't be you.

I have watched founders paste a bland brief into a model, receive a polished paragraph, and look briefly relieved-as if the hard part of marketing had been solved by fluent sentences. The sentences were fine. The problem is that fine sentences are now abundant. What remains scarce is the thing the model cannot invent without you: a point of view earned from contact with reality.

AI can write your content. It cannot replace your perspective. That distinction sounds philosophical until you publish a month of generated posts and notice they leave no residue. People skim, nod, forget. The work did not fail because it was badly written. It failed because nothing in it could only have come from you.

This piece is about why perspective is still the moat, when AI content feels empty even when it is competent, and how to use drafting tools as labour-not as a substitute for judgement. Related: your competitive advantage was never typing faster.

Fluency is no longer rare

For a long time, “good writing” was a proxy for thinking. If someone could structure an argument, they had probably done the thinking. That proxy is collapsing. Models produce structure on demand. Smooth openings. Balanced lists. Sensible conclusions. The surface of competence is cheap.

What is not cheap is deciding what deserves to be said. Perspective is a selection problem: which tension you notice, which trade-off you refuse to paper over, which customer sentence you treat as more important than the category slogan. Selection requires taste formed by experience. Generation requires pattern matching. Confusing the two is how brands start sounding like everyone else at once.

• Fluent prose can now be produced without a fluent mind behind it.
• Readers still detect when nothing is at stake in the argument.
• Perspective is what makes one piece worth forwarding when ten others cover the “same” topic.
• If your draft could be published by any competitor with a different logo, you do not have a perspective problem-you have a commodity.

Why AI content often feels empty

Empty is not the same as incorrect. Empty means the piece never risked a claim that could make someone disagree. It summarises consensus. It hedges where a practitioner would choose. It names benefits without naming costs. It reads like a meeting minutes document of what “the industry” already believes.

Models are trained on averages. Averages are useful for orientation and dangerous as a brand voice. When you ship average thinking in elegant packaging, the packaging becomes the product-and packaging does not compound. For how trust actually forms, see people trust people before they trust marketing.

If your content strategy is “ask the model what to think,” you have outsourced the only part of the job that creates differentiation.

A framework: notice → stake → shape → ship

1. Notice: collect what only you are positioned to see

Perspective begins before the blank page. It begins in sales calls, support tickets, failed experiments, awkward product decisions, and the sentence a customer used that no competitor’s landing page repeats. Keep a raw log. Not polished insights-raw notices. Models can rearrange notices. They cannot attend your Tuesday demos.

2. Stake: force a claim that could be wrong

A stake is a sentence that commits. “Consistency beats brilliance” is a stake. “Here are five tips for growth” is not. Before you open a drafting tool, write one stake in plain language. If you cannot, you are not ready to generate-you are ready to research. Curiosity still beats prompt tricks; more on that in the best prompt is still a good question.

3. Shape: use AI as a craftsperson’s assistant, not an author

Once the stake exists, generation becomes useful: expand examples, propose counter-arguments, tighten structure, produce alternate openings. The model works inside a fence you built. Without the fence, it will build a very neat paddock in the middle of nowhere.

1. Write the stake and the audience in one sentence each.
2. List two lived examples the model cannot invent (names changed as needed).
3. Ask for structure options-not final copy.
4. Draft or expand only after you choose the structure.
5. Edit for voice until a colleague could identify it as yours without looking at the byline.

4. Ship: publish the perspective, not the polish

Polish without perspective is a well-ironed empty shirt. Ship the claim. Accept that someone will disagree. Disagreement is often a sign you said something. Silence is what commodity content receives. Cadence still matters-perspective compounds when it is repeated with integrity, not when it is reinvented weekly. See the internet rewards consistency more than brilliance.

A fictional contrast: two B2B newsletters

Founder A asks a model for “a weekly newsletter about productivity for startups.” It arrives competent: tips, frameworks recycled from public blogs, a cheerful CTA. Open rates drift down. Replies are rare. The list grows; belief does not.

Founder B keeps a notice ledger from onboarding calls. One week the stake is: “Most ‘productivity’ tools fail because founders use them to avoid deciding what not to build.” The model helps expand examples and tighten the structure. The essay names a painful trade-off. Three founders reply with stories. One asks for a call. Same category. Different residue.

Perspective as a moat (without the mythology)

A moat is not a slogan about uniqueness. It is a reason a peer prefers your articulation when alternatives exist. Perspective becomes a moat when it is specific, repeated, and tied to proof. Specificity comes from contact. Repetition comes from systems. Proof comes from product and customer reality-not from adjectives.

This is why organic growth still has to be earned even when drafting is free: organic growth isn’t free-it’s earned. Free words do not equal earned attention. Attention still tracks belief.

• Specific: names a constraint other posts avoid.
• Repeated: the same spine across formats, not a new persona weekly.
• Tied to proof: examples, outcomes, or observed patterns.
• Owned: language you would still use in a sales call without flinching.

Where teams lose the plot in practice

The failure mode is rarely “we used AI.” It is “we used AI as a substitute for contact.” A product marketer who has not sat in three recent sales calls will ask a model to invent buyer psychology. The draft will sound plausible. It will also miss the objection that actually kills deals. Plausible is the enemy of useful.

Another failure mode is perspective tourism: one week you are the bold contrarian, the next you are the helpful explainer of industry news, the next you are a tip machine. Models make tourism easy because every costume is available on demand. Readers do not build trust with costumes. They build trust with a recognisable spine they can predict and quote.

1. Audit the last six published pieces for a single spine you could name in one sentence.
2. If you cannot name it, write the spine before the next draft-not after.
3. Replace one “tips” piece with one stake drawn from a real conversation.
4. Keep AI for expansion and structure once the stake and examples exist.
5. Measure replies and forwards, not only impressions, for four weeks.

What research is useful for

For how people evaluate credibility online, Nielsen Norman Group remains practical on trust heuristics. Harvard Business Review regularly covers differentiation and brand meaning without reducing everything to content hacks. The Stanford Encyclopedia of Philosophy entry on points of view is denser than a growth tweet-and occasionally useful when you need language for why “seeing-as” matters. None of these replace sitting with customers.

How this shows up in my work at nau

I help founders build product stories and publishing systems where tools accelerate craft without replacing taste. We start with stakes and evidence, then use generation for structure and scale. If you need systems more than another campaign spike, start with why I care more about systems than campaigns.

And when the work gets operational-what to automate versus what to own-read AI should handle the repetition; you should keep the judgement.

Keep the scarce part scarce

AI will keep getting better at producing paragraphs. That is not a threat to people who have something to say. It is a threat to people who mistook paragraph production for strategy. Protect the notice. Force the stake. Let tools shape the labour. Ship the perspective. That is still the advantage-and it was never available in the autocomplete.