google ai search guidance | liam hayward

So, I’ve had a little think…

Google released its guidance on AI search back in May, and it caused a stir in the SEO community. My LinkedIn feed was pretty much entirely posts with screenshots from SEOs gleefully highlighting various passages that debunk GEO/AEO practices as “hacks” and “myths”.

As part of an industry both dominated by Google and constantly uneasy about the impact of AI, I too breathed a small sigh of relief at the validation that I’d been doing the right thing by not drinking the kool-aid and rebranding myself as a GEO specialist (I know very little about rocks).

The feeling of vindication was immense. This was going to be the most beautiful day of my life. My tesco meal deal tasted better than any meal you and I have ever tasted… you get the point.

As the initial high tailed off, I realised something didn’t quite sit right with me. It was too perfect - it was mollifying and pacifying. It reminded me of the unspoken pact SEOs think Google made with them: do as we say and you’ll be rewarded. Blessed are the SEOs, for they shall inherit the SERPs.

The advice seemed to tick all the boxes, giving SEO practitioners a tummy rub and telling them they’re doing well, whilst mocking the GEO and AEO tactics: “look how stupid they are, can you believe they thought this would work?”

I’m not saying I think the Google AI search guidance isn’t genuine. It’s that it’s pretty much exactly what the SEO community wanted to hear (and already believed). It’s so weirdly perfect and placating that I can’t help being suspicious and asking what it might be trying to achieve.

It also came just days before Google I/O 2026, where Google heralded a new era for AI search and welcomed us to the agentic Gemini era: a near future in which agents, not people, do the browsing. Hold that thought.

by google, for google

I’m not the first person to point out that the guidance is by Google, for Google: the definition of self-serving. Mike King’s breakdown reads it as the continuation of a pattern that’s been excellent for Google and lousy for the people doing the work.

Read it and you can feel the industry’s shoulders drop. It busts the AEO and GEO “myths”, waves off the panic about needing new tricks like an llms.txt file, and reassures a jittery profession that nothing has really changed: it’s still just SEO, you’re all doing fine.

But underneath the reassurance sits one request, made over and over until it stops sounding like advice: make unique, non-commodity content. In a document barely 2,500 words long, “unique” and “helpful” each turn up nine times, with “non-commodity” thrown in for good measure another five. Here it is, in its own words, over and over:

“Create valuable, non-commodity content for your audience”
“Creating content that people find unique, compelling, and useful…”
“Providing a unique point of view”
“a first-hand review provides a unique perspective based on personal experience”
“Create the content yourself based on what you know about the topic…”
“Don’t just recycle what others on the internet have already said, or could easily be produced by a generative AI model.”
“Creating non-commodity content that’s helpful, reliable, and people-first”
“Be sure that you’re writing non-commodity content…”
“provides unique expert or experienced takes that go beyond common knowledge and the ordinary.”
“creating unique, valuable content”
“Focus on developing unique, expert-led content that provides value beyond common knowledge.”

The obedient SEO in me knows Google has always told us to create helpful, reliable, people-first content, but this advice, repeated ad nauseam, is starting to sound more like a plea than a suggestion.

the ai ouroboros

It’s worth setting that plea for unique content against what Google actually does with the stuff once it has it. It’s taken it for its AI Overviews with minimal attribution, driving a massive decline in click-through rate to the sites that created it. Meanwhile the internet that AI feeds on is filling with machine-written content: Ahrefs looked at 900,000 fresh web pages and found roughly three quarters already carried AI-generated content, a looming problem for the future of the web.

We’re soaring toward an AI ouroboros, with LLMs eating their own generic spiel and regurgitating it over and over.

It’s not just a hunch, though. While I was still writing this post, Graphite published a study on “AI search collapse” (covered on the SEO handbook), the timing almost too neat to be true. It found that once models start retrieving their own output, 79.6% of simulations collapse toward the same uniform, homogenised answer. Worse, this isn’t some far-off worry: the same research projects that 42.7% of the references ChatGPT pulls in are now themselves AI-generated. The snake has found its own tail, nom nom nom.

So when I read the messaging about unique writing injected throughout the guidance, the tin-foil-hat SEO in me thinks they’re actually worried about running out of nutritious, non-commodity content for their answer robot to eat. It reads like something engineered to go viral among SEOs, and underneath the applause it keeps repeating the same instruction: keep pumping out unique content, and maybe you’ll get a click when their AI regurgitates it later.

never send a human to do a machine’s job

The cherry on top is what the guidance tells you to do next: go and ready your site for “agentic experiences”, the browser agents and search agents that visit on your users’ behalf, the very ones from that I/O stage, running 24/7 in the background. So it praises you for writing “people-first” content in one breath, and in the next points you toward a future where an agent does the reading and your visitor never arrives. We spent years fretting about zero-click search. This is worse: no click, no visit, not even the courtesy of an impression. So much for inheriting the SERPs.