Google’s AI Search Guide, the Ahrefs Schema Study, and the AI Content Panic: What’s Actually Worth Your Attention

Google published its first official guide to optimising for AI search features on 15th May 2026. Four days earlier, Ahrefs released a study tracking 1,885 pages that added schema markup, finding no meaningful impact on AI citations. The AI content debate continues to rage. And the SEO industry is doing what it always does: splitting into camps, claiming vindication, and selling products off the back of it.

I’ve been working in search since 2003 and on the web since 1997. This pattern is familiar. The same tribal dynamics played out around toxic backlinks in 2012-2014, when Google’s Penguin update convinced half the industry that disavowing links was essential and the other half that disavow was snake oil. Agencies built entire service lines around link audits. Most of that work was unnecessary. The same thing happened with mobile-first indexing around 2018, when “your site will disappear if you don’t redesign for mobile” became a sales pitch for responsive redesigns that were often already in progress. And E-E-A-T from 2018 onwards spawned an industry of author bio optimisation, EEAT audits, and “topical authority” consulting, most of which fundamentally misunderstood what Google was actually evaluating.

Every cycle follows the same arc. Google publishes something measured. The industry overreacts. Consultants with products to sell shout loudest. And the people quietly doing good work don’t change much, because the fundamentals rarely do.

Here’s what I think is actually worth paying attention to this time, and what you can safely tune out.

Key takeaways if you don’t have time for the full article

• Google confirmed its AI features use retrieval-augmented generation from the existing search index. Good SEO feeds AI visibility directly.
• Schema markup isn’t required for AI search. Google said this explicitly. But they also said to keep using it as part of your overall strategy.
• The Ahrefs study found no citation uplift from adding schema, but only tested pages already getting 100+ AI citations. Whether schema helps undiscovered pages get found is a different question the study can’t answer.
• Google didn’t say stop using AI for content creation. They said stop producing commodity content. Those are different statements.
• If you’ve been writing comprehensive content that covers a topic thoroughly, you’ve already been doing what Google now calls “query fan-out” coverage. They gave a name to something good content creators have done for years.
• Every study and every reaction in this space has commercial incentives behind it. Read accordingly.

Does schema markup help AI citations? What the Ahrefs study actually shows

The Ahrefs study is more rigorous than most SEO research out there. They tracked pages that added JSON-LD, matched them against control pages, and ran four separate statistical tests including difference-in-differences analysis. The finding: adding schema produced no meaningful uplift in citations across Google AI Overviews, AI Mode, or ChatGPT. AI Overviews showed a small 4.6% decline that was statistically significant but practically tiny.

Credit where it’s due. Control groups, multiple tests, transparent limitations. That’s rare in SEO research.

But the study has constraints that the headline obscures. Every page in the dataset was already getting 100+ AI citations before schema was added. These pages were already inside the consideration set. The study answers “does schema help pages that are already being cited get cited more?” It cannot answer “does schema help pages get into the consideration set in the first place?” Those are fundamentally different questions.

The sample is also self-selecting, not randomised. Market research can draw representative conclusions from 1,000 respondents because the sampling is random against a well-defined population. The central limit theorem makes population size largely irrelevant once you have enough random samples. But the Ahrefs sample comprises pages whose owners chose to add JSON-LD during a specific window, tracked by one tool’s crawler. That’s a convenience sample with inherent selection bias. Pages that add schema tend to be mid-way through broader SEO work. Content changes, technical fixes, internal linking updates. The study acknowledges it can’t separate schema from these co-occurring changes.

Ahrefs also cited a searchVIU experiment where AI chatbots were asked to extract data from a test page. None of them parsed JSON-LD during the live fetch. That’s a valid finding about one specific retrieval pathway. But live chat retrieval is a lightweight, speed-optimised operation. It grabs visible text and moves on. Expecting it to parse, validate, and reconcile JSON-LD against page content in real time is like expecting Google to render JavaScript on the first crawl pass. The heavier processing, including schema parsing and entity extraction, happens during indexing, where the system has time and resources. Testing one pathway and drawing conclusions about all of them is a methodological stretch.

And the incentive question applies. Ahrefs sells backlink analysis, content tools, and Brand Radar (featured prominently in the study). Their conclusion, that the sites getting cited invest in content quality, authority, and links, describes their product suite. The study is well-executed. The conclusion is also commercially convenient.

Three days after the Ahrefs study, Google published guidance saying schema isn’t required for AI search but remains valuable for your overall strategy. Meanwhile, the pro-schema camp had already been making bold claims well before either publication landed. Stackmatix had published that “content with proper schema markup has a 2.5x higher chance of appearing in AI-generated answers,” with no source. Digital Applied had referenced “internal Google documentation” confirming schema influences AI Mode source selection, with no link. Multiple consultancies were already positioning schema audit services as essential for AI visibility. The inflated claims didn’t emerge from the evidence. They predated it. The pattern is familiar on every side: stake a position, wait for anything that can be selectively interpreted as support, sell services.

What Google’s AI optimisation guide actually says about content and rankings

The most useful parts of Google’s guide are the technical explanations. Google confirmed that AI features use retrieval-augmented generation (RAG) from their existing search index, and described “query fan-out,” where the AI model generates concurrent sub-queries to fetch additional results. If someone searches “how to fix a lawn full of weeds,” the system might simultaneously search for herbicide recommendations, chemical-free approaches, and prevention strategies.

This is genuinely helpful to understand. But if you’ve been writing content that comprehensively covers a topic, addressing the questions your audience actually has, anticipating related concerns, covering the angles that a thorough treatment requires, you’ve been doing query fan-out coverage for years. Google just gave it a label. Good topical content has always meant thinking beyond the single query and addressing the web of related questions that surround a subject. Nothing about this requires a new strategy. It requires continuing to do what competent content creators already do.

The “non-commodity content” section is where the guide gets more interesting and more problematic. Google says to create content with a unique point of view, based on genuine experience, that goes beyond common knowledge. Write from what you know. Provide first-hand perspective rather than summarising what’s already out there.

In principle, sound advice. In practice, several problems that nobody in the excited commentary is raising.

Unique perspectives lack corroboration. LLMs are trained on consensus. If 95% of sources say one thing and your genuinely novel insight says another, retrieval systems are architecturally less likely to surface your content. Google is telling you to be unique while building AI systems that structurally favour conformity. The guide doesn’t acknowledge this tension.

Expert experience is also a content pattern, not proof of expertise. “In my fifteen years of working in this field” is a sentence anyone can write. Any competent copywriter or AI model can produce that framing. Google’s quality systems cannot verify lived experience at scale. They can evaluate signals that correlate with expertise, like consistent publishing history, citations from authoritative sources, and entity recognition. But the content-level markers of experience are trivially replicable.

And the advice itself is self-serving in ways worth noting. Unique, non-commodity content makes Google’s job easier. It reduces deduplication problems. It gives ranking and quality systems clearer signals. It produces a better search index. The advice benefits publishers too, but Google has always steered content creators toward practices that simplify crawling, indexing, and ranking. This guide continues that pattern. Understanding this doesn’t mean ignoring the advice. It means reading it with appropriate context.

Does Google’s guide mean you should stop using AI for content?

No. And Google didn’t say that. But a significant portion of the industry heard it that way.

The guide says to meet quality standards, follow spam policies, and create non-commodity content. It links to Google’s existing guidance on AI-generated content, which has consistently said method of production matters less than quality of output. The distinction that actually matters: AI-generated content that restates what’s already widely available online is commodity content by definition. That’s what Google is warning against. AI-assisted content, where someone with genuine knowledge uses tools to produce work they couldn’t have scaled or articulated as effectively on their own, is a different proposition entirely.

The debate is stuck on “should you use AI or not” when the relevant question is “does this content add something that didn’t exist before?” If the answer is yes, the production method is secondary. If the answer is no, it doesn’t matter whether a human wrote it longhand. Commodity content is commodity content regardless of who or what produced it.

How to optimise for AI search: what actually works in 2026

The same things that have worked for the twenty-three years I’ve been doing this professionally.

Build technically sound sites. Ensure your content is crawlable, indexable, and well-structured with semantic HTML. Implement schema markup as infrastructure, because the cost is low, the downside is zero, and the potential upside exists across multiple pathways even if no single study has proven a direct AI citation effect. Don’t let anyone tell you schema is a magic lever. Don’t let anyone tell you it’s worthless either.

Create content that genuinely serves the people you’re trying to reach. Cover topics thoroughly. Bring actual expertise to what you publish. If you’re writing about something you know well, you’ll naturally produce the kind of non-commodity, fan-out-friendly, expert-led content that Google’s guide describes. If you’re writing about something you don’t know well, no amount of schema or optimisation will fix that.

AI visibility matters. Usage of AI search is growing rapidly, and being absent from those conversations has real consequences. But treat AI citation as one signal within a broader visibility strategy rather than a standalone metric to optimise for in isolation. The measurement infrastructure is still maturing, and the link between citation and business outcomes is not yet well understood. Build for it. Monitor it. Don’t let it become the only thing you measure.

And stop reacting to every new study or guide as though the ground has shifted beneath your feet. The platform evolves. The delivery mechanisms change. The principles of building something useful, making it findable, and earning trust with your audience have not changed in the entire time I’ve been in this industry. They’re unlikely to start now.

Be sceptical of anyone selling certainties. The people shouting loudest usually have products attached.