You've Been Writing for Readers. Now You Need to Write for Machines Too

Intro

You spent years learning how to write for people. Short sentences. Active voice. A hook in the first paragraph. You know the rules.

But there's a new audience reading your content - and it doesn't care about your hook.

AI systems - ChatGPT, Perplexity, Google's AI Overviews, and others - are now processing and summarizing web content at massive scale. When someone asks one of these tools a question in your industry, it pulls from pages it can confidently parse, extract, and cite. The question is: is yours one of them?

The frustrating part is that great writing for humans can be nearly unreadable for machines. Not because the content is bad - but because the structure makes meaning hard to extract.

Here's what that looks like in practice, and what you can do about it.

The Problem Isn't What You're Saying. It's How You're Saying It.

When a language model reads a page, it doesn't experience your content the way a human does. It doesn't benefit from context built up over paragraphs. It doesn't pick up on implied meaning. It extracts - pulling individual sentences or sections and using them to construct an answer.

That changes what "good writing" means.

Consider these two versions of the same sentence:

Version A: "Our solution helps teams stay on the same page and get more done."

Version B: "Acme's project management software reduces cross-team miscommunication by centralizing task tracking in a single dashboard."

Version A sounds warmer. Version B sounds more corporate.

But Version B is the one an AI will quote. It names the company, specifies what the product does, and identifies the problem it solves - all in one sentence that can stand completely alone. Version A, extracted from context, means almost nothing.

Writing for Extraction: A Simple Framework

You don't need to rewrite your entire site. You need to think about each major piece of content slightly differently.

The core question is: if someone lifted this paragraph out of context, would it still communicate something clear and specific?

Here's a practical way to apply that test:

1. Name your entities explicitly. Don't write "our platform" - write the platform's name. Don't write "this approach" - say what the approach is. AI systems build meaning from named entities. Pronouns and generic references don't give them anything to hold onto.

2. Define before you qualify. Lead with the facts, then add nuance. "Bounce rate dropped 40% after restructuring the navigation" is extractable. "After making some changes to how users moved through the site, engagement improved significantly" is not.

3. Make each section answer one question. The best content for AI retrieval is organized around discrete questions: What is this? Who is it for? What problem does it solve? How does it work? Each section should be able to answer its question independently.

4. Avoid hanging context. A sentence that starts with "As mentioned earlier..." or "Building on the above..." becomes meaningless when extracted. Restate what's necessary. Repetition that feels redundant to a human reader is actually helpful for a machine.

The Structural Layer Below the Writing

Even perfectly written content can fail to reach AI systems if the structure underneath it doesn't support machine readability.

This is where things move beyond writing style into a technical layer that most content creators haven't had to think about before.

A few things that matter here:

• Schema markup - structured data that tells AI-enabled search systems not just what's on your page, but what type of thing it is. For platforms like Google's AI Overviews, schema has a well-documented impact on how content is interpreted and surfaced. For LLMs like ChatGPT and Perplexity, the relationship is less direct - but clear structure and semantic signals still make your content easier to process and represent accurately.

• Heading hierarchy - clear, descriptive headings (H1, H2, H3) give AI systems a map of your content. Vague headings like "More Information" or "The Next Step" don't communicate anything useful.

• llms.txt - a proposed convention (not yet an official standard), similar in concept to robots.txt, that gives AI crawlers explicit guidance about your site's structure and what they're allowed to use. It was introduced by researcher Jeremy Howard and has been adopted by several major AI crawlers - but adoption is still uneven, and many sites that have the file haven't verified it's actually being read correctly.

Tools built specifically for this layer - like the ai visibility tool Geordy - exist precisely because this isn't something most content teams can manage manually. Geordy automatically generates the structured formats AI systems use to understand and cite your content: schema, llms.txt, YAML, Markdown, and more, served from your own domain without touching your existing site.

What This Looks Like in Practice

Let's take a real-world example. Say you run a small accounting firm and you've written a solid explainer page about quarterly tax filing for freelancers.

Right now, when someone asks ChatGPT "when do freelancers need to file quarterly taxes," your page may not be among the ones cited - even if it's well-written and ranks well on Google. A few common reasons:

• Your key information is embedded in flowing prose that's hard to extract out of context.

• Without a clearly structured FAQ section, it's harder for AI systems to identify which part of your page directly answers a specific question.

• Your firm's name and specific services aren't linked clearly to the claims being made.

The fix isn't rewriting the whole page. It's adding a clearly structured FAQ section with explicit schema markup, making sure your firm name appears near every key claim, and ensuring your headings are question-based rather than topic-based.

"Quarterly Filing Deadlines" becomes "When Are Quarterly Tax Deadlines for Freelancers?" Same information. Much more extractable.

The Audience Has Expanded - Your Writing Should Too

This isn't about abandoning everything that makes content readable for humans. Good writing is still good writing.

But the definition of "good" has expanded. Content that works today needs to serve two audiences: the person reading your page, and the AI system that may cite your page in an answer someone else receives - without them ever visiting your site at all.

That second audience has different needs. It can't be inferred. It can't hold context. It extracts and represents.

Write for both. Structure for both. And verify that the technical layer underneath your content actually supports what you're trying to communicate - because right now, for most sites, it doesn't.

That's a gap worth closing.