I used to Google things and scroll through search results.
But now, I’m asking questions and expecting a tailored answer.
Somewhere over the last year or two, that seems to have become the default for most people I speak to, and it changes something really important for marketers. Your reader, more often than not, might not be human at all.
You’ve probably noticed it already.
You ask ChatGPT a question, or Google throws up an AI Overview before you’ve even scrolled, and somewhere in that answer is a summary of what a brand, a blog, a product page said, minus the visit to the actual page.
You don’t click through. AI does the reading of the website on your behalf, and decides what content was worth repeating.
That’s not a small shift. For twenty years, writing for the internet has meant writing for two audiences: the person reading it, and the algorithm ranking it.
Now there’s a third, an entire system that reads your content, decides whether it’s worth quoting, and then paraphrases it back to someone who may never see your site at all.
So the question I keep getting asked by clients, and keep asking myself, is: Does this mean we need to write differently?
The honest answer is yes and no, and the “no” bit matters more than most of the current advice out there admits.
The reader you didn’t know you had
Generative engine optimisation, or GEO traces back to a 2024 academic paper that found certain changes to content could boost its visibility in AI-generated answers by up to 40%, though they were honest that this varied a lot depending on the topic.
It’s still a fairly young idea.
Nobody’s fully agreed on what to call it yet, you’ll see GEO, Generative SEO, AI visibility and even “answer engine optimisation” used almost interchangeably, which tells you the industry is still figuring this out in real time, same as we all are.
But is this actually new, or just SEO with a new coat of paint? Here’s where it gets a bit more interesting, and where I think a lot of the current advice oversells itself.
Google’s own developer documentation on AI Overviews says, in plain terms, that “there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimisations necessary.”
Their position is that this is just good SEO, same as it always was, dressed up with a new acronym.
Meanwhile, Semrush and HubSpot have both published detailed GEO guides telling marketers to restructure content specifically for AI extraction, add more quotes and stats, tighten headings, and build authority signals.
Semrush’s guide, published earlier this year, even goes as far as saying content with quotes and statistics sees roughly 30 to 40% higher AI visibility.
So who’s right? Probably both, depending on what you’re actually trying to fix.
If your content is genuinely thin, badly structured, or unclear, Google’s advice is still valid. If you take the time to write something genuinely helpful for your audience, the rest tends to follow.
But if your content is already decent and simply isn’t built in a way a model can easily lift and quote, that’s where the craft-level changes start to matter.
It’s the same discipline, with a slightly different reader looking over your shoulder.
What actually changes?
If there’s a genuine difference between writing for people and writing in a way that also earns a spot in an AI answer, it lives here, at the sentence and paragraph level, not in some new overarching strategy.
This is the bit worth bookmarking, because it’s more about tightening habits most of us already know, but don’t always stick to when a deadline’s looming…
So, here are a few tips for how to write content that’s more likely to get picked up and quoted, whatever you make of GEO (or whatever you’re calling it) as its own discipline:
- Lead with the answer, then the context, not the other way round. Models tend to pull self-contained sentences out of your content, so if the actual point is buried under three sentences of scene-setting, it might just never get quoted.
- Write in chunks that can stand alone. A paragraph that only makes sense if you’ve read the one before it is a paragraph that won’t survive being lifted out and reused.
- Use real structure, not just nicely written prose. HubSpot’s guide makes the point well: AI-driven search tends to favour content that’s “extremely structured via bulleted lists, structured headings, and listicle-style articles.” That doesn’t mean turning every post into a listicle, but it does mean your headings should mark a real shift in the point being made, not just break up the page for the sake of it.
- Trade vague marketing language for specifics. Named numbers, named sources, named examples. A model has nothing to extract from a sentence that just says “innovative solutions for modern businesses.”
- Build in the signals that say a human with real expertise wrote this: original data, a real quote, a named author. That’s the same E-E-A-T thinking Google’s had in its guidelines for years, it’s just doing double duty now.
It’s bigger than your own website
Here’s the part that surprises most clients when I bring it up.
Being visible in AI answers isn’t only about your own copy.
These models are drawing on the wider web’s opinion of you, forums, review sites, Wikipedia, YouTube comments, Reddit threads, not just what you say about yourself.
Semrush flags something genuinely counterintuitive here: unlinked brand mentions, the kind that never used to count for much in traditional SEO, seem to carry real weight with AI systems.
At the same time, having a Wikipedia presence matters disproportionately, simply because of how much AI training data is pulled from it.
For a lot of businesses, that reframes the conversation. It’s not just “is our website well written,” it’s “what does the internet, collectively, seem to think of us.”
Digital PR and community presence stop being a nice-to-have next to content and start being part of the same job.
Where you can start
If you want a starting point rather than a full audit, I’d focus on this, in this order:
- First, check whether your key pages answer the actual question in the first two sentences, not the fifth paragraph. If not, move the answer up and push the context down.
- Second, look at your longest, least structured pages and ask whether a person skimming, or a model scanning, could pull a clean, accurate sentence out of any of them.If not, break it up: real headings, shorter paragraphs, a list where one actually fits.
- Third, find the two or three places where a generic claim could become a specific one, a real number, a named client outcome, a direct quote. If you can’t find any right now, that’s usually a sign there’s no original thinking on the page yet.
- And fourth, start paying attention to what gets said about you off your own site. You can’t fully control it, but if you don’t know what’s out there, that’s the gap to close first, before you can judge whether it’s helping or hurting you.
A caveat worth being honest about
I’ll say the thing most GEO content won’t: a lot of the statistics doing the rounds right now, including some I’ve used above, are still early and largely untested independently. That doesn’t make them wrong, but it does mean the models themselves are changing fast enough that what works today may not hold in six months.
I’ve chased a moving target like this before, a few years back it was writing for voice search and featured snippets, and most of that effort aged about as well as milk left out on a warm day.
Here’s what’s actually stuck with me, though.
Everything in this piece, leading with the answer instead of burying it, writing in chunks that stand on their own, being specific instead of vague, caring about what gets said about you off your own site, is the same advice good marketers have been giving for years.
Google will tell you that’s just SEO. Semrush and HubSpot will tell you it’s GEO.
I’d call it just good writing, wearing a new coat because a new kind of reader turned up.
Write for the human first. It turns out the machine has been reading over their shoulder this whole time, and, funnily enough, it’s judging you by the same standard your best readers always did.
So no, you don’t need to learn a whole news skillset to write for robots. You just need to stop writing like one.
If you’re not sure whether your own content would pass the test, or would like support with your content marketing, get in touch, we’d love to talk it through.