Google just settled the GEO debate. The hard part starts now.

On May 15, 2026, Google published its first official guide on optimizing for AI search. After two years of speculation, vendor pitches and contradictory advice, the company finally went on record.

The headline was blunt. AEO and GEO are not separate disciplines. They are SEO.

The mythbusting section named tactics that an entire cottage industry has been selling: llms.txt files, content chunking, AI-specific schema, rewriting pages in machine-friendly prose. Google said all of it is unnecessary for AI Overviews and AI Mode.

For retail teams, the news is not that the rules are simpler. It is that the rules are now public, definitive, and harder to execute than ever. Because what Google said actually works is the one thing most catalogs cannot produce at scale: non-commodity content.

What Google actually said

The guide draws a clear line between what matters for AI search and what does not. Here is the official position.

What does not work

•   llms.txt files have no impact on AI Overviews or AI Mode.

•   Content chunking is not needed. Google's systems parse multiple topics on a single page natively.

•   There is no special AI schema. Existing structured data is enough.

•   Rewriting content in stilted, machine-friendly prose hurts more than it helps.

•   Seeking inauthentic mentions across blogs and forums is filtered out by core ranking systems.

What does work

•   Non-commodity content built on first-hand experience, original data and a real point of view.

•   Multimodal assets. Original images, videos and diagrams are increasingly cited in AI responses.

•   Technical hygiene. Crawlable, indexable, fast, mobile-ready pages remain the floor.

•   E-E-A-T signals applied across the full site, not just YMYL pages.

The non-commodity content shift is the real story

Buried inside the guide is the concept that will reshape retail content over the next 24 months: non-commodity content.

Google defines it directly. Commodity content is generic, interchangeable, and could have been written by anyone. The example used in the guide is "7 Tips for First-Time Home Buyers." Non-commodity content is built on direct experience, proprietary data and a specific point of view. It is the kind of content AI cannot generate from training data alone, which is precisely why AI systems cite it.

Apply that distinction to a retail catalog and the implications are immediate.

A product description that lists six attributes copied from a manufacturer feed is commodity. Newtone's analysis of 30 enterprise retail catalogs found that 72% of PDPs across the sample met Google's definition of commodity content. They restated facts available on dozens of other pages, with no original framing, use case detail or proprietary information. We covered this attribute depth problem in detail in Why 60% of your product catalog is invisible to AI search.

Across the same dataset, the PDPs that earned AI citations in ChatGPT, Gemini and Perplexity shared three traits: enriched attribute depth (15 to 20 fields versus the catalog average of 6), use case language specific to the buyer context (not the product), and integrated customer signals like reviews, fit data or care insight.

In other words: the pages that get cited are the pages that could only have come from the brand.

Category pages: where most retailers will lose

If PDPs are the harder problem at volume, category pages are the harder problem at quality.

Google's guide is explicit that commodity, restated content gets deprioritized. Most retail category pages are precisely that. A 60-word intro, a grid of products, and either no FAQ block or a generic one copied from competitor sites. Nothing that could only have come from the brand.

Newtone's audit data shows the median enterprise retail category page contains 83 words of editorial content. Across the same sample, category pages that earned AI Overview citations averaged 420 words of original editorial. Not keyword stuffing. Genuinely useful, brand-specific content: sizing guidance, material insight, styling context, regional fit notes, seasonal logic.

This is the gap. Retail teams know category pages need depth. They know FAQ-style content drives AI citations. They know freshness matters because LLMs deprioritize content older than two months, as we documented in Your category pages have an expiry date.

They just cannot produce that depth, at that quality, across hundreds of pages, in ten languages, every quarter.

What Google said vs what most vendors are still selling

The gap between Google's official guidance and what AEO and GEO vendors have been pitching is wide. This is what the documentation says about the most common tactics on the market.

The pattern is consistent. Most AI search tooling sold to retail brands in the last 18 months addresses tactics Google just said do not matter. The signals Google did confirm are the ones that take real content effort, at real catalog scale.

The execution problem nobody is solving

Google has just removed the excuse. There is no AI shortcut. There is no special file, no special schema, no clever rewrite that unlocks AI visibility.

What works is the hardest thing in retail content: original, expert, brand-specific writing across thousands of pages, refreshed continuously, localized natively, optimized for SEO and AI search at the same time.

This is a math problem before it is a content problem.

We have written about this at length in The content execution gap costing retailers their AI search rankings. A typical enterprise retail brand publishes 300 new products per week, manages 400+ category pages requiring refresh every two months, operates across 5 to 10 markets, and runs a content team of 6 to 12 people. The math does not work.

Agencies cannot scale to it. Newtone's analysis of agency pricing across 14 enterprise retail contracts shows an average cost of 180 euros per category page brief, 8 to 12 day turnaround, and brand drift after the third batch.

Internal teams cannot hire to it. The median content team grows 12% year over year while catalog volume grows 40% to 60%.

Custom GPTs and generic AI tools plateau at roughly 70% quality. The remaining 30% takes 23 minutes of human editing per asset, multiplied by every page in the catalog, every refresh cycle, every market. This is the shift we explored in Why retail content teams are moving from AI tools to AI infrastructure.

Google just told the industry what to build for. The question is whether you can actually build it.

What this changes for retail content teams

Three implications follow from Google's guide if you run content for a retail brand.

Stop paying for GEO-specific tactics that do not work

If your team or your agency has invested in llms.txt generators, AI chunking tools or proprietary AI schema packages, that spend is no longer defensible. Redirect it to where the signal actually is: original content, multimodal depth, and technical fundamentals.

Treat AI search optimization as content infrastructure, not a workflow

The teams that win will not be the ones writing better briefs. They will be the ones that have built a system to produce expert-level, brand-specific content across the entire catalog, in every language, at a refresh cadence the LLMs will actually reward. This is infrastructure work, not workflow work, as we argued in Generative engine optimization, a new visibility frontier for enterprise retail.

Measure non-commodity output, not page count

Update your content KPIs. Number of new pages and number of refreshed pages are no longer sufficient. Measure how much of your published content qualifies as non-commodity by Google's own definition: original framing, proprietary attribute depth, brand-specific point of view. Anything else is being filtered out of the AI surface.

Where Newtone fits

Google has just confirmed what we have been building toward.

Newtone is content engineering infrastructure for retail. We do not sell GEO hacks or AEO tactics. We build the AI content creation suite, integrations and workflows that let enterprise brands like Chanel, ASICS, Sandro, Nissan, Adanola and Victoria Beckham produce on-brand, non-commodity content at the scale and quality their catalog and markets require.

Trained on each brand's editorial guidelines, SEO rules, and existing content, our models generate publish-ready PDPs, category pages, FAQs, blog content and localized assets with the brand voice consistency, attribute depth and originality Google has now confirmed as the actual ranking signals for AI search. See how we approach this in our SEO and GEO use case.

The strategy is settled. The execution gap is what matters now. That is what Newtone closes.