GEO isn't a replacement for good writing — it's proof that good writing always mattered. Here's what shifts when you write for AI search, and what stays the same.
The thing about search engine optimization (SEO) — or, at least, what we've been able to put our finger on with regards to search engine optimization, which has always been the fastest moving target in web design — is that it's always been hidden in a black box, which means we've never really been confident in what works and what doesn't. But there have always been a lot of OPINIONS on what works and what doesn't. Many of those opinions have been right. Many of those opinions have been right for a certain time period. And many of those opinions have seemed right due to common sense.
Which means we've bounced back and forth, a lot. As editors, we've followed the guidelines until the guidelines changed — whether that change was fueled by the algorithm itself, or by new research and analytics — and then we adjusted our thinking around those changes.
And now, we find ourselves on the precipice of a new change. Well, that's not totally correct; we find ourselves knee deep in the middle of a complete overhaul of how the web works, and particularly about how web content works. We find ourselves at the start of the answer engine revolution, and the instinct for most editors is to drop everything, find out the new rules, and follow them.
There’s a silver lining, here; this time, the rules all look a lot more familiar. More than that, they look like the rules we should have been following all along.
So, what's changed? What's new? And what's been there the whole time? What, exactly, is the difference between SEO and generative engine optimization (GEO)?
What does GEO mean for working editors?
Picture an editor. Her name is, I dunno — Sierra. She's staring at a service page that she's updated three times in the past year. The page isn't bad — it's thorough, and it answers the question, but it's not performing the way it used to. She knows this because someone in a meeting last week told her that web traffic is down and that we need to start doing something about "GEO," and now she's supposed to figure out what that means and fix the page.
The short answer — and the one Sierra probably needs to hear — is that GEO isn't as foreign as it sounds. Generative engine optimization is the discipline of structuring content so that AI-assisted search tools can find, understand, and cite it. And the core of what it asks for is the same thing good editors have always been tasked with: write clearly, answer real questions, and don't make the reader work harder than necessary to find the point.
How AI search changed what it means to "rank"
For roughly two decades, SEO was about signals. It was about making sure you put the keywords in the right places and making sure you link from credible sources. Page load times, character counts, and backlinks were things we hyperfocused on, and while a lot of traditional SEO is editorial in nature, a fair amount was mechanical. We wanted to put the right thing in the right boxes to get a position in the results, regardless of whether it was written for the human being who might be reading it.
Position mattered, because visibility meant clicks, and clicks meant traffic.
Simple enough … until now, when generative search breaks that equation. When someone asks Claude or ChatGPT or Google's AI-assisted search a question, they're not picking from a list of results — they're getting an answer assembled from whatever sources the model decided were credible and clear enough to quote. There's no position one to aim for, not really. There's "cited" and "not cited" — and traditional ranking signals, while still relevant, don't guarantee you end up in the answer.
The shift toward generative search has thrown the entire conversation on its side. We've moved away from the idea of a very complicated math equation toward something a bit more real — toward something a lot more simple, at least for those of us writing and creating content: the thing that makes AI cite you instead of someone else, as far as we can tell right now, is simply good writing.
It's a little humbling. The same things that made content easy for people to read have always made it easier for machines to process, which is to say that GEO didn't invent new requirements — it just raised the stakes on the ones that were already there: things like clear structure, specific answers to real questions, and plain language. In other words, content that explains what it means instead of assuming you already know.
What to avoid when shifting our mindset from SEO to GEO.
A lot of the content sitting on a lot of websites right now — content that ranked fine under traditional SEO — was written in ways that made it harder to read. Editors weren't doing this on purpose; they were responding to the incentives in front of them: write to the keyword, structure for the algorithm, and optimize for the metric, which unfortunately is exactly what the job asked for.
But traditional search was forgiving enough that you could get away with a lot of it. You coud bury the answer in paragraph four because you wanted to "set up" the point, and it didn't matter because Google would still surface the page based on keyword density rather than readability. Or, you could use a heading like "Our Approach" — a heading that doesn't really give you a clue into the actual content of the following paragraphs — because the algorithm was looking for other clues.
This changes with AI-assisted search that depends on generative LLM (large language model) because, unlike a traditional search crawler that's just looking for signals, a generative model is actually trying to use your content — it needs to understand what you're saying well enough to hand it off to someone else as an answer. Vague, jargon-heavy, and poorly structured content suffers and gets skipped because the model can't reliably quote something it can't clearly parse. Which means as we adjust how we think about writing for the web, we need to be aware of some new traps — traps that may already be in place, because until recently they were never penalized. To move into the future, we need to:
- Avoid burying the answer. Long preambles that wind up to the point don't serve a reader who wants to know the answer — and they don't serve a model trying to find it, either. (This is especially hard for me, your humble author, who has made a living winding up to the point.)
- Avoid headings that label instead of describe. "Overview." "Background." "Our Process." These are organizational markers, not meaningful headings, and they mean very little on their own. A heading like "How we structure a content audit" tells the reader — and the model — what the section actually covers. "Overview" tells nobody anything. (Bonus: this is also how to write for accessibility — some assistive devices help readers scan by headline, which makes those headlines even more important.)
- Avoid jargon that assumes prior knowledge. If your audience includes people who are still figuring out what they need, writing like they already know your terminology creates a gap. Your human reader isn't going to look it up. And you can't always count on the model having enough context to fill it in.
- Avoid keywords over questions. Content written around a keyword phrase ("content strategy services") rather than a real question ("how do I figure out what goes on my new website?") tends to answer a question that nobody asked.
What writing for GEO actually looks like in practice.
None of this requires a complete content overhaul. Most of it is just habit — a shift in what you're paying attention to when you sit down to write or edit.
Which is to say: the practical version is not very exciting, and it's going to look very very familiar:
- Answer first, then explain. Whatever the point of the paragraph is, lead with it. If someone could only read the first sentence, would they still have something useful? If not, rewrite the first sentence.
- Write headings that work out of context. Imagine your heading appearing in an AI-generated answer with no surrounding text. Does it still mean something? "How content modeling affects your CMS setup" works on its own. "Getting started" does not.
- Define before you abbreviate. Every time you write "CMS" without having first written "content management system," you're assuming knowledge. Sometimes that assumption is fine. Often it isn't. When in doubt, spell it out.
- Write FAQs for the questions people actually ask, not the ones you wish they'd ask. "What is content strategy?" serves the glossary. "Do I need a content strategy before I redesign my website?" serves the person who's actually trying to make a decision — and it's specific enough to be genuinely useful rather than just technically present.
- Write paragraphs that could be quoted whole. If an LLM pulled your paragraph and dropped it into an answer, would it hold up on its own? Self-contained, specific, and clear is what gets cited. Prose that relies on the three paragraphs before it doesn't travel well.
It's not an exciting list. More than that, it's not really a checklist at all — it depends on the grey areas of editorial judgment. That's sort of the point: the editorial habits that position you well for generative search are the same ones that make content useful to a real person in a hurry.
Good writing has always been GEO-ready
There's a version of this conversation that treats GEO as a brand new discipline that requires brand new expertise. We don't think that's the right frame, and we're a little suspicious of anyone selling it that way. The truth is: writing for GEO has already been here. It's called "writing for people."
GEO rewards editorial clarity and it rewards thinking about the person reading your content before thinking about the system indexing it. Editors who kept that discipline as they optimized for SEO are now quietly positioned as the most GEO-ready people in the room.
Which brings us back to Sierra — still at her desk, probably still a little worried. Here's what I'd tell her: "Your service page doesn't need a full GEO audit. Instead, it needs a clear first sentence, headings that fulfill the promise of what will follow, and paragraphs structured for understanding. That's most of it. The rest is just writing like someone's going to read it — because someone is. Possibly a person. Possibly a model handing that person an answer.
Either way, write it for them.
