SEO Didn’t Disappear. The Search Environment Got Bigger.
Share
SEO is dead. Now it's GEO, or AIO, or answer engine optimization — and depending on who's selling the service, there is usually another acronym waiting behind the last one.
A note on my own usage before I go further: when I say AIO here I simply mean optimization for AI-assisted search and answer experiences. I'm not treating it as a standardized technical term, because it isn't one yet.
Some of the excitement is justified. Search really has changed, and fairly dramatically. People ask longer questions. Google answers more of them directly. ChatGPT searches the web and returns answers with links and citations. Products, videos, social content and ordinary websites increasingly compete inside the same discovery environment.
But the claim sitting underneath all of it — that the foundation of search disappeared the moment AI arrived — has never made much sense to me. A search system still has to find information, understand what it's about, and decide what to surface. Google's current guidance is unusually direct on its side of this: its generative Search features are built on the same core ranking and quality systems as the rest of Search, and site owners get pointed straight back to foundational SEO practice.
SEO didn't vanish. The environment around it got considerably bigger.
The foundation is still there
Strip away twenty years of tactics and SEO was always solving a fairly basic problem. A search engine needed to discover a page, access it, understand it, and work out when it deserved to appear. That's why crawlability, internal links, page structure, clear subject matter, useful content and external authority became important in the first place. None of that becomes irrelevant because a system can now assemble an answer before showing its sources.
What's strange is how much of the current GEO conversation skips straight past it. I keep meeting businesses hunting for an AI-search tactic while the website underneath is difficult to navigate, poorly structured, or so vague that it takes five minutes to work out what the company sells. There isn't much point optimizing the attic while the foundation is still moving.
For Google, ordinary Search eligibility comes first: a page has to be indexed and eligible to appear with a snippet before it can qualify for generative features. Google has also shipped a Search generative AI control in Search Console. Inclusion is the default. That control finished rolling out to all websites worldwide on 31 August 2026, although the help documentation still notes that not every property has caught up. Where it's live, it governs whether a site's links and content can appear in and help ground those generative experiences.
OpenAI handles crawling on its own terms. Any public website can potentially appear in ChatGPT search, and OpenAI recommends allowing OAI-SearchBot if you want content discoverable and usable for search-backed answers, citations, summaries and snippets. Opting out limits that use, though OpenAI notes a URL can still appear as a navigational link in some circumstances.
None of this is the interesting part of the conversation. It simply didn't stop mattering.
What changed is the shape of the question
A traditional search looked like this:
best lightweight rain jacket
Now someone can ask this instead:
I need one shell for hiking, travel and wet winter weather. I run hot, I hate bulky jackets, and I want something that packs into a small bag. What should I actually look for?
There are four or five questions buried inside that one request. Google's generative systems can use query fan-out to run related searches around a broader question, and ChatGPT search may rewrite a prompt into one or more targeted searches, issuing additional ones as needed. I don't think every marketer needs to turn that into a systems diagram. The practical consequence is enough on its own: the customer can now ask the whole question instead of compressing it into three words.
So the old test — can we rank this page for this keyword? — is still worth applying. I just wouldn't stop there anymore. The better test is whether, once the customer asks the entire question, the company has enough genuinely useful information around the subject to remain relevant. That version exposes thin content very quickly.
Keywords still matter. The obsession matters less.
There's been an overcorrection here too. Some people now talk as though keywords are irrelevant. Of course words matter — people use language to explain what they want, and companies use language to explain what they offer. What matters less is building a page around every microscopic wording variation. Google says outright that its generative systems can connect queries with relevant pages even where the exact phrasing doesn't match, and warns against producing masses of pages to catch variations or anticipated fan-out queries.
A premium design-build company does not need five thin service pages chasing five versions of essentially the same phrase. It needs to explain clearly what it builds, the kinds of projects it handles, how its process works, where projects tend to go wrong, and why anyone should trust it with an expensive piece of work.
I care much more now about the questions sitting underneath the keyword. What does the customer need to understand? Where do they hesitate? What are they comparing? What do they misunderstand until somebody who actually does the work explains it? And what does this particular company know that a generic article cannot tell me?
That last question keeps getting more important, because average information has become incredibly cheap.
Blogging didn't die. Commodity blogging ran out of excuses.
This one still frustrates me. People say blogs don't work anymore as though a generic 700-word article titled Five Reasons to Remodel Your Kitchen and a detailed piece of firsthand knowledge are the same object. They are not. Generic blogging was already weak. AI simply made generic information cheaper to produce and removed the last excuse for tolerating it.
Now imagine a remodeling company publishes What we found after opening the walls in three older hillside homes. That can contain real observation. Unexpected framing. Old wiring. Moisture. Structural changes nobody documented. The discovery that added two weeks to the schedule. The thing homeowners consistently underestimate before demolition begins. That is information the company acquired by doing the work, and nobody can generate it from a prompt.
Google's guidance is quite direct on this point. It asks for unique, useful, expert-led, non-commodity information, and specifically contrasts firsthand material with content that reworks what's already available or could easily be assembled from existing sources.
That's where business blogging should have been heading anyway. The goal was never to maintain a publishing quota. It's to turn real knowledge into something useful and permanent.
Say what you actually sell
A company website should probably be the clearest version of the business anywhere on the internet. Oddly enough, a lot of them aren't. You can read the homepage, the about page and half the services section and still be unsure what the company does, because everything is transforming, elevating, innovating and delivering solutions.
Fine. But what do you sell?
For a professional service, clarity might come from strong service pages, documented expertise, case studies and useful analysis. For a product company, from specifications, comparison content, testing, reviews and demonstration video. For a contractor, from project documentation, customer questions, service information and evidence of completed work. There's no reason those three businesses should have the same content mix — which is exactly where the current GEO conversation gets too tidy.
An e-commerce company is not a consulting firm
Take a premium outdoor brand selling technical equipment. Its product information is structural: materials, dimensions, variants, price, availability, photography, shipping information, customer reviews, product feeds. Google points e-commerce businesses toward Merchant Center and accurate product data for exactly this reason, because product information can surface in both conventional and generative Search experiences. If that layer is weak, I wouldn't start by ordering twenty blog posts about outdoor adventure. I'd fix the product layer.
Then there are the questions product data answers badly. How does it perform after six months? Does the expensive material really last longer? What does it look like packed into a vehicle with everything else? Does the sizing run the way the chart suggests? That's where field testing, comparison guides, real customer feedback and video earn their place.
Now take a business advisory firm. There's no product feed. Nobody needs to watch an unboxing. What's being sold is judgment. If the entire site amounts to some version of "we help companies achieve their full potential," a prospective client has almost nothing to evaluate. I'd far rather read the firm explain where succession plans break down, what causes owners to misjudge valuation, how operational weaknesses surface during a transaction, and what it learned from engagements that went sideways.
One of those companies needs intellectual depth. The other needs visual proof. Handing both the same GEO content package is where strategy quietly turns into a checklist.
Somebody else has to say it too
A business can call itself the best company in its category fifty times across its own pages. It's still the business making the claim. Reviews, legitimate publications, professional references, independent product reviews, customer discussion, interviews and relevant links produce a different kind of evidence entirely. Google specifically warns against seeking inauthentic mentions around the web as a generative Search tactic, which tells you how common the attempt has already become.
There's no clever shortcut hiding in there. You have to give other people legitimate reasons to talk about the company.
Social media belongs in this system too, though its importance swings wildly by category. A visually driven consumer brand may depend heavily on Instagram, YouTube, TikTok or creator content. A specialized engineering consultancy probably shouldn't spend half its marketing budget trying to become a lifestyle channel. Social can create awareness, distribute expertise, and demonstrate a product in ways a static page cannot. It can also become a very expensive treadmill when nobody has decided what job it's supposed to perform.
If I were building today
I'd work from the business outward. Make sure the website can be accessed and understood. Make the company painfully clear about what it sells, who it serves and what it actually knows. Build the pages closest to revenue before filling the site with supporting content. Then document useful expertise, and give independent sources legitimate reasons to validate it.
Distribution comes after that. Not because social, video and email are unimportant, but because they work considerably better when there's something substantial to distribute. I'd rather watch one useful piece of work travel across five appropriate channels than watch a business operate five unrelated content factories.
Measurement is getting less theoretical
This part has improved quickly. Google launched dedicated Generative AI performance reports in Search Console on 3 June 2026, beginning with a subset of UK site owners and completing a worldwide rollout on 31 August. The reports show impressions, which pages appeared, countries, devices and visibility over time. They don't show clicks, they don't show queries, and they certainly don't expose Google's retrieval or ranking logic.
Bing arrived first and offers a different view. Its AI Performance report entered public preview in February 2026 and expanded in June with intent classification, topic grouping, citation share and comparison tools. It shows citation activity, cited URLs and sampled grounding-query phrases across supported Microsoft AI experiences. Microsoft is explicit that citation counts are not page rankings, authority scores, placement indicators or measures of importance. Useful visibility, yes. A map of the algorithm, no.
Both are worth switching on now, if only so there's a baseline sitting there before somebody asks what happened.
What's actually left
GEO does add a genuinely useful perspective. Traditional SEO spent a lot of time asking whether a page could rank. Now we also have to ask whether information is accessible enough to be retrieved, whether a system can understand it accurately, whether the source adds something distinct, and whether the broader web supports the same story. I like those questions.
I'm far less interested in the supposed hacks. Google says there's no special AI schema required for its generative Search features, no requirement to chop pages into tiny AI-friendly chunks, and no Google Search advantage from publishing an llms.txt file.
Which leaves the harder work. Build something worth finding. Be clear about what it is. Publish actual knowledge. Give independent sources reasons to validate it. Distribute it where it makes sense.
The industry will rename the process again. The underlying work is harder to rename, because most of it was there all along.
Sources: Google's AI features and your website guidance and its Search Console generative AI performance reports announcement; Microsoft's AI Performance public preview and its June 2026 expansion; and OpenAI’s crawler documentation for OAI-SearchBot.