Who Tells the Robots What to Do?

It's a reliable genre. Every few months, a wave of posts appears announcing that search as we know it is over. The writing is urgent. The logic is not always present.

The model numbers change. The claims don't. This summer it's an AI-first interaction layer. Before that it was SGE. Before that, AI Overviews, RankBrain, the mobile-first index, Hummingbird, Panda. Each one was going to end SEO as we know it. Each one required a new acronym and a new set of consultants who understood the acronym.

The irony that runs through almost all of them is the same one Vladimir keeps asking about. If every query becomes an AI-generated answer, if search is now an interaction rather than a list of links — where does the AI get its information?

Not from nowhere. From the web. From pages that exist, that load, that are structured clearly enough for a machine to read, assess and cite. From content someone had to write, organise, and make findable.

Someone has to tell the robots what to do.

The acronym cycle

The latest version goes by a few names. GEO — Generative Engine Optimisation. AEO — Answer Engine Optimisation. The pitch is that you need to get cited by AI. Get your content into those boxes. Appear in ChatGPT's answers. Show up in Perplexity.

As if ChatGPT is a channel you can just turn up on.

There's a version of this conversation worth having. AI-generated answers are changing how users interact with search results. Zero-click is growing. CTR on informational queries is under structural pressure — not because those pages are underperforming, but because the answer is being served before the click. That's a real shift and it needs a real response.

The response being sold, though, is mostly rebranded SEO with a new name on the tin.

Think about the man on the high street holding the golf sale banner flag. You've seen him. Probably ignored him. Maybe glanced at it. He's visible, he's there, he exists in your peripheral awareness as you walk past.

Now imagine someone selling you that flag waving as a marketing strategy. Your brand, cited in a ChatGPT response. Mentioned. Present. Technically in front of people.

Did anyone buy anything?

Nobody knows. Nobody measured it. The flag was waving, which felt like progress.

You're selling answers hoping someone buys another question. And that someone has to feel curious enough, or uncertain enough, or ready enough to actually come to your site to ask it. That's a much thinner thread than a click from someone who already wanted what you had. If you've answered the question so completely that nobody needs to click, you've done Google's job for free.

GEO, as it's currently being sold, is often the flag. The citation exists. The AI mentioned you. Whether that mention did anything — whether it sent a human being to your site, whether that person engaged, whether they converted, whether the assisted value of that touchpoint is positive or negative compared to what you spent achieving it — that question tends not to appear in the pitch deck.

Visibility is not value. It never has been. Ranking isn't data, it's a perspective. A citation in a ChatGPT answer is a perspective too. Useful, possibly. Measurable, if you set the tracking up properly. Worth reorganising your entire content strategy around?

Ask the man with the flag how many golf clubs he's sold.

Google said so themselves

In May 2026, Google published official documentation titled "Optimising your website for generative AI features on Google Search." It is worth reading in full. The short version is this: AEO and GEO are not separate disciplines. They are SEO.

Google went further. The guide specifically names the tactics currently being sold as essential AI optimisation work — llms.txt files, content chunking, AI-specific rewriting, special schema markup — and confirms that none of them are required. The systems already understand long pages. They already extract relevant sections without needing content broken into fragments. The markup people are charging to implement isn't doing what it's being sold to do.

This is not the first time Google has done this. The keyword density industry. The link-building industry. Each time a new set of services forms around a perceived technical shortcut, Google eventually publishes something that makes the shortcut redundant. The fundamentals remain. The shortcut doesn't.

We've been here before

This isn't a new habit. The industry has always reached for the metric that's easiest to report rather than the one that's most accurate.

Last-click attribution was the previous version of the same comfortable fiction. Default in GA4, default in almost every analytics platform before it. It works by giving all the credit for a sale to the last channel the customer touched before converting. Simple. Clean. Reportable in a slide deck without any caveats.

There is a version of your analytics that makes paid search look brilliant and organic look mediocre. Most people are looking at that version right now.

The problem is that customers don't behave like that. A person who finds a product through an organic search, sees a retargeting ad three days later, clicks a paid link and then buys — that sale gets credited entirely to paid. Organic gets nothing. The assisted journey, the initial discovery, the trust built across multiple touchpoints — none of it appears in the number. Paid search looks like a genius. Organic looks like a drag on the budget. Budgets shift accordingly.

The data to tell a more accurate story already exists in GA4. Most people are not looking at it.

There's a practical dimension to this that rarely gets mentioned in the GEO pitch. Out of the box, GA4 doesn't track AI referral traffic properly. Visits arriving from ChatGPT, Gemini, Perplexity or Claude get dropped into Referral by default — if they get attributed at all. A significant portion arrives with the referrer header stripped entirely and disappears into Direct. Without deliberate configuration — a custom channel group, a maintained regex, regular source auditing — most GA4 setups are completely blind to what AI platforms are actually sending.

So the industry is selling optimisation for a channel that most businesses cannot measure. The flag is waving. Nobody is counting who walked past.

The GEO citation is the same mechanic dressed in different clothes. A mention in a ChatGPT answer feels significant because it's new and visible and has a number attached to it — your brand appeared in this many AI responses this month. That number will get put in a report. It will get compared to last month's number. Someone will decide it needs to go up.

Whether any of it connects to revenue will remain, as with last click, a question nobody is quite asking loudly enough.

The point is not that AI citations are worthless. They may not be. The point is that a single metric reported without context will systematically distort the decisions made around it. We know this. We've known it for years. We just keep finding new metrics to do it with.

Standing on the shoulders of giants, then charging you for the view

There's a detail that tends to get lost in the noise. The transformer architecture that underpins ChatGPT — the technical foundation of how it processes language, generates answers and understands context — came from a Google Research paper published in 2017. "Attention Is All You Need." Google's own BERT followed in 2018. OpenAI built GPT on those foundations.

The techniques powering the thing that's supposedly killing Google were invented by Google. The best part of a decade ago.

It compounds further when you look at what's actually happening with ChatGPT's search function right now. Testing of ChatGPT 5 suggests it has moved away from relying primarily on Bing's index and is now pulling from Google's. The product competing with Google search is, at least in part, running on Google's index. The irony is structural, not incidental.

So the pitch being made — restructure your content strategy, learn new skills, pay for new services, optimise for the AI citation — is asking you to pay for access to a system built on Google's intellectual groundwork, increasingly powered by Google's index, in a market Google still dominates by a margin that hasn't meaningfully shifted. We've covered that in detail in our search market share breakdown — the headline figure worth knowing: all AI search engines combined account for less than 0.28% of actual search referrals. Google isn't going anywhere.

Standing on the shoulders of giants, then charging you for the view.

The personalisation argument finishes it off. Google has been personalising search results since at least 2009. Location, device, search history, behaviour — results have never been a single universal list. The direction of travel is more personalisation, not less. Which means the idea of optimising for the AI citation is not just unproven, it's structurally incoherent. There isn't one answer to rank in. There are millions of answers, serving millions of different users in millions of different contexts.

The flag isn't even waving at the same people.

What the robots actually need

Machines — whether they're traditional search crawlers or large language models — need content that is clear, structured, authoritative and attributable. They need to know who wrote something and why they're qualified to write it. They need to be able to parse a page without ambiguity. They need signals that a piece of content is the right answer to a specific question.

That is an SEO brief. It has always been an SEO brief.

The framing changes. The underlying requirements don't. E-A-T became E-E-A-T. Featured snippets became AI Overviews. PageRank became something more complex and less transparent, but still fundamentally about authority and relevance.

Every time the interface changes, the posts appear. The clock is ticking. The summer is coming. SEO is over.

The robots are still there, reading the web, trying to figure out what to say.

Someone still needs to give them something worth reading.