Don't use AI to write: AI accelerates; it doesn't replace thinking

The B2B question we hear across LatAm is simple: “Can I just use AI to write the blog, white paper, or services page and ship it?” Somos Gente Digital’s short answer: AI does not replace thinking or writing the argument; it accelerates research and feedback.

The essay that sparked this angle is Don't use AI to write by Paul Bakker (August 2026). We attribute carefully: no invented quotes or badges.

At SGD we are a human web/tech B2B agency team. We use AI every day. We also insist on people-first content: useful, original, and human-reviewed — not thin spam or “N words” with no owner.

Writing forces you to think deeply about the problem you are solving and what you want readers to understand. That is why writing is hard — and why it is useful. If you generate a document from bullet points (or a prompt) and accept it with light tweaks, you skipped the thinking step.

The issue is not that AI tools are bad at writing (often they are good). They are bad at thinking and inventing new ideas with your judgment. That is where you come in.

Productivity in writing is not shipping the most pages in the least time — just as programming is not producing the most lines of code. It is the impact of the ideas on your team, company, or readers. Does a fluffy 60-page strategy help more than a sharp three-page one that hits the real problem?

This is not dogma. It is about not replacing your thinking with a model’s “thinking.”

Where it helps (before and after you write):

• Research: sorting data, finding patterns, comparing sources.
• Preparation: understanding the problem before drafting the solution.
• Feedback: “read this — what questions would you have?”, “what is the main takeaway?” — like asking a coworker to review, not to write for you.

Where it hurts:

• Asking the model to write the document from scratch from bullets and publishing with almost no human review.
• Measuring success only by volume (“N words / week”) without clarity, originality, or real usefulness.
• Skipping the argument: if you cannot explain the idea out loud, the generated text does not have it either.

Practical tip (aligned with Bakker): ask for questions, not automatic “fixes.” Feedback should put you back in thinking mode — the same as a good coworker review.

Our flow is not “prompt → publish.” It is a cycle with a human owner:

1. Brief: a real business question, audience, one idea per URL.
2. A human writes the argument: thesis, evidence, examples, honest limits.
3. AI asks questions: gaps, objections, clarity, what still needs citation.
4. Hermes SEO·GEO·AEO review: people-first signals, structure, citable FAQs, useful links — no thin AI spam.
5. Publish with judgment: honest title, useful excerpt, aligned meta.

In practice: brief → a human writes the argument → AI asks questions → Hermes SEO·GEO·AEO review. That looks more like custom software and services with judgment than shipping “N words” with no human owner.

If the contrast with AI speed without a domain map resonates, we also wrote about vibe-coding without a map: there the map (and the Requirements Manual as a working guide) stops you from reinventing poorly. Here the map is thinking the argument before asking for prose.

Before we publish a B2B post, we run simple signals (SGD SEO Norma / people-first / E-E-A-T — no invented badges):

• Answers a real B2B business doubt (not an empty keyword).
• Carries an original SGD take — not only a generic third-party summary.
• Honest title: promises what the piece delivers.
• Citable FAQs: questions a client would actually ask, with clear answers.
• Useful internals: services, product, or related posts when they help — not filler.
• One idea per URL: do not mash three topics to “win length.”

Publishing “N words” without human review clashes with helpful content and how SGD works. We prefer fewer, clearer URLs.

So I should never use AI to write?

Use it for research and to challenge your draft. Do not use it to skip thinking the argument or to publish text with no human owner.

What do SEO·GEO·AEO mean here?

SEO is traditional search optimization. GEO is about how generative engines cite you. AEO (answer engine optimization) aims at clear, citable answers. At SGD, Hermes reviews those signals without turning the post into spam.

Can I ask AI to “improve” the text?

Better: ask for questions. Automatic fixes push you into passive editing. Feedback should return you to thinking.

Does this contradict using AI in engineering or marketing?

No. We use AI daily. The line is replacing human judgment in the argument — the same spirit as vibe-coding without a map: speed without understanding the problem is not progress.

How do you avoid thin content / AI spam?

Human brief, one idea per URL, original evidence, real FAQs, Hermes review, and not measuring success by word count alone.

What if my team already ships from prompts?

Start by reclaiming the brief and the human draft. Use AI as a questioning reviewer. Cull thin URLs. Prefer clarity over volume.

If you want B2B content that feels written by people — because it is — and a flow where AI accelerates without emptying the argument, let’s talk. At SGD, content is part of the product, not a page generator.