Google Finally Told Us How to Show Up in AI Search. Here's the Playbook.

Google just published an official guide on how to optimize your site for its generative AI features, the ones powering AI Overviews, AI Mode, and answers in Gemini. If you have spent the last year reading breathless threads about secret AI ranking factors, the guide is almost funny in how calm it is. The headline, in Google's own framing, is that its AI features are rooted in core Search ranking and quality systems. There is no separate AI algorithm to game. Good SEO is the AI strategy.

That sounds anticlimactic, but it is genuinely useful, because it tells you what to stop doing as much as what to start. Here is the playbook, and how it maps to what Potatometer already scores.

How Google's AI actually picks content

Two mechanisms do the work:

1. Retrieval-augmented generation (RAG). Google's normal ranking systems retrieve relevant, fresh pages, and the AI then reads specific passages to compose an answer, with clickable links back to sources. If you cannot rank, you cannot be retrieved, and if you cannot be retrieved, you cannot be cited. 2. Query fan-out. A single user question gets expanded into several related sub-queries that run at once, each pulling its own results. This is why breadth matters: the model is quietly asking five questions, not one, and you want to be a credible answer to as many of them as possible.

The practical takeaway: AI visibility starts with classic retrievability. Crawlable, rankable, trustworthy pages are the price of entry.

What Google says to actually do

Write genuinely original, people-first content. Google leans hard on distinctive perspective: first-hand experience, original reporting, expert insight, the things a model cannot synthesize from everyone else's pages. Commodity content that restates what is already out there gives the AI no reason to surface you specifically.

Keep it crawlable and technically clean. Meet the standard Search technical requirements. Follow JavaScript SEO best practices if you are on a heavy framework, because AI features use publicly accessible, crawlable content. Reduce duplicate content. Keep page experience solid across devices. None of this is new, which is the point.

Structure content for humans. Clear headings, real paragraphs, sensible sections. Google explicitly frames this as readability for people, not as "chunking for machines" (more on that below).

Use structured data as part of the mix. It is not required to appear in AI features, but it supports rich results and helps Google understand your page. For ecommerce and local businesses, Merchant Center feeds and Google Business Profiles feed the same machinery.

Support text with strong visuals. AI features can pull in relevant images and video, so high-quality multimedia that genuinely supports your content can help.

What Google says to stop doing

This is the part worth printing out. Google's mythbusting section is unusually direct:

• Do not bother with llms.txt or special AI markup. Google Search ignores them. If you added an llms.txt file expecting Google to read it, it does nothing for Google. (Other AI systems are a separate conversation, but for Google specifically, it is inert.)
• Do not artificially "chunk" your content into fragments you imagine an AI will digest more easily. Write for people.
• Do not rewrite content specifically for AI. There is no special AI writing style that wins.
• Do not manufacture inauthentic mentions of your brand across the web. Fake buzz is a risk, not a tactic.
• Do not assume you need a dedicated page for every long-tail keyword variant, or that perfect semantic HTML is a prerequisite. It is not.

The pattern is clear: Google is pushing back on a cottage industry of AI-specific hacks and pointing everyone back to substance.

Where this leaves the "llms.txt" debate

This is the spiciest bit, so let us be precise. Google says its Search systems ignore llms.txt. That is a statement about Google, not a universal verdict. Some other engines and crawlers may use it, and the standard is young. But if your entire AI plan was a text file in your root directory, Google just told you it is reading your actual pages instead. Optimize those.

How this maps to your Potato Score

If you have run a Potatometer scan, most of Google's guide is already staring back at you:

• AI Crawlability checks whether GPTBot, Google's crawlers, and friends can even reach you. That is the retrieval step Google describes.
• Structured Data and Content Structure line up with Google's structured-data and readability guidance.
• Content Depth is our proxy for the original, substantive, people-first content Google keeps emphasizing.
• Our GEO score exists precisely because being rankable and being quotable by an AI are related but not identical jobs.

One honest caveat Google's guide implies and our own data agrees with: being cited in an AI answer is not the same as getting traffic. AI Overview citations behave more like a mid-page result for clicks. Treat AI visibility as brand presence and mental availability, then measure real impact through brand-search lift and conversions, not raw referral counts.

The one-sentence version

There is no AI optimization secret. Publish original, helpful content on a crawlable, well-structured, fast site, skip the AI-specific gimmicks, and you are doing the thing Google's own guide asks for. Run a Potatometer scan to see which of those boxes you are missing.

Source: Google Search Central, AI features and your website.

← Back to Blog