Pratik Goswami

Posted on May 22 • Originally published at Medium

I Read Google’s Official AI Search Guide So You Don’t Have To. Here’s What Actually Matters

#webdev #ai #google #productivity

Google just wrapped up its Google I/O 2026 and after spending some time going through all of Google's official documentation on Search Agents, I decided to create a one-stop guide to things I took away from all the information. While I originally posted on Medium, I wanted to share it here as well because the Dev.to community tends to care more about these stuff.

Agentic Search Optimization, Answer Engine Optimization, Generative Engine Optimization - the terminology changes depending on who you follow, but the panic underneath it is the same: Google has changed, agents are reading your site, and if you are not optimizing for it, you are invisible.

So developers have been doing what developers do. They are adding llms.txt files. They are "chunking" their content. They are rewriting copy specifically for AI models. Some are even paying for tools that promise to get their site cited by ChatGPT.

Here is the thing. Google just published an official guide on how to optimize your website for generative AI features on Google Search. I read it cover to cover. Most of what the internet is selling you isn’t what search engines are looking for. This article highlights what can actually boost your visibility to such agents.

First, a Quick Recap on Why This Matters

If you read my previous article, Your Portfolio Is Invisible. Here’s How I Fixed Mine, feel free to skip this section as you already know the core idea. Your site has two types of readers: humans and machines. And increasingly, it is the machines that decide whether the humans ever find you.

SEO (Search Engine Optimization) is about helping crawlers like Googlebot index and rank your content so people find you in search results.

AEO (Answer Engine Optimization) is about helping AI systems form an accurate and confident understanding of your site, so they can surface it when someone asks a relevant question.

The difference matters more now than it did a year ago. Google is actively building Search Agents. These are autonomous systems that can perform tasks on behalf of users. Whether it be booking a reservation, comparing products or researching services; to do any of that, the agent first has to read, understand, and trust your site. If it cannot do that, it will not recommend you and that is the whole problem.

What Google Actually Says

Google has created a guide titled Optimizing your website for generative AI features on Google Search. It is the official position from Google Search Central and it serves as a guide to optimize a website for the new Search Agents.

From Google’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.

It means the same foundational principles that have always made a site rank well - original content, clean structure and good performance - are still what matter. The difference is in how those principles are applied in a world where an AI model is summarizing your page rather than a user clicking through to read it.

Part 1: What Actually Moves the Needle

1. Non-Commodity Content

This is the section that gets the most emphasis in the Google guide and the least attention in most AEO tutorials.

Google uses the phrase “valuable, non-commodity content” and it is worth understanding what they mean.

Commodity content is a summary of information that already exists. It answers a question accurately, but it could have been written by anyone or generated by an AI model. The content has neither a point of view nor a first-hand experience. It does not tell you something you could not find in the first five results for the same query.

Non-commodity content is the opposite. It can include a first-hand review, case study with real numbers or an expert take that reflects genuine experience in the field. Something that provides a unique viewpoint.

Here is the uncomfortable truth. The content most developers are cranking out to “rank for AI” is precisely what AI models are trained to ignore. If you are writing blog posts that restate what the documentation already says, you are producing commodity content. The irony is that the shortcut most people reach for, using AI to generate more content faster, tends to produce the exact type of content that generative AI systems are trained to look past.

Your perspective is the signal. The depth you bring from actually having built, shipped, and debugged something, is what makes your content worth citing.

2. Clear Technical Structure

This part will be familiar if you read the SEO article, but the reasoning has a new layer to it now.

Google’s AI systems use the same underlying infrastructure as traditional search. That means the same technical requirements apply. A page has to be indexed and eligible to appear in Google Search before it can appear in AI Overviews or be used by a Search Agent.

The checklist looks like this.

Semantic HTML. This is still the highest-leverage change most developers can make. When your markup uses <article>, <section>, <nav>, <h1> through <h3>, and <main> correctly, a crawler reads your page like an outline. Without them, it reads a flat document and has to guess what matters. The Google guide specifically notes that while perfectly semantic HTML is not required, it is generally a good idea because it helps multiple types of users, including AI systems that parse and navigate your page structure.

Crawlable content. If your content only renders after a user interaction, a state toggle, a tab click, or a modal open, it effectively does not exist to a crawler. Googlebot crawls the initial DOM, not the post-interaction state. Audit your site for content that is conditionally rendered and refactor it so the important information is present in the initial HTML.

Sitemap and robots hygiene. A sitemap provides a guide to the crawlers about every page on your site so nothing gets missed. A robots file tells crawlers which pages they are and are not allowed to visit. Both are small files that have an outsized impact.

Good page experience. Google’s guide explicitly links page experience as an AI search signal. Thus, Core Web Vitals like LCP (Largest Contentful Paint), which measures how long it takes for the largest visible element to load, has a major impact. A slow site is not just a UX problem, it is a visibility problem.

3. Structured Data (The Right Way to Think About It)

There is a nuance here that most articles miss. While structured data is not required for generative AI search, it is still a good idea to use it responsibly as it is a part of improving the overall SEO strategy. In other words there is no special schema.org that can give you an edge with AI search, but it does help with being eligible for rich results on Google Search.

Generative AI search is built on top of the same index as traditional search. A site that is well-optimized for rich results tends to also be better represented in AI responses. Structured data gives crawlers explicit, machine-readable facts about your content rather than making them infer those facts from your prose.

A Person schema on a portfolio tells Google your name, your role, your employer, and your social profiles in one structured block. A Product schema on an e-commerce page gives price, availability, and reviews in a format that leaves nothing to interpretation.

For a personal site, the most useful schemas are Person, WebSite, and SoftwareApplication if you are showcasing projects. For a commercial site, focus on Organization, Product, FAQPage, and LocalBusiness if applicable.

The goal is not to stuff your page with schema, rather to shine the spotlight on your most important facts.

Part 2: The Myths Worth Dropping

I want to go through it because these myths are widespread, and I believe acting on them is a waste of time at best, counterproductive at worst.

llms.txt is not a signal

This has been circulating as the hot new AEO tactic. The idea is that you create a machine-readable file at your domain root describing your site for LLMs, similar to how robots.txt works for crawlers. Google’s guide is clear to highlight that you do not need to create llms.txt or any other "special" markup for AI. While they will be crawled and indexed, they don’t provide any advantage to improving visibility.

Chunking content does not help

The theory is that breaking your pages into smaller pieces helps AI systems process them more easily. Google explicitly highlights this as a myth, as their systems can handle multiple topics on a single page and surface the relevant piece to users. There is no ideal page length for generative AI search. So make pages for your audience, not for the AI.

You do not need to write for AI keywords

AI systems can understand synonyms. They understand the general meaning of a query and match it to relevant content even when the phrasing does not match. You do not need to worry about capturing every variation of how someone might search for your topic.

Seeking inauthentic “mentions” does not work

Some tools and guides push you to get your brand mentioned in blog posts, forums, and discussion threads across the web, with the reasoning that it will help AI systems associate your brand positively.

Google addresses this directly. Their core ranking systems focus on high-quality content while others focus on blocking spam. Their generative AI feature depends on both. Thus, pushing for bulk mentions may be harming visibility instead of boosting it.

Part 3: The Part Most Guides Are Not Writing About Yet

AI agents are autonomous systems that can perform tasks on behalf of users - not just answer questions, but take actions. A browser agent, for example, can visit your website, analyze visual renderings (like screenshots), inspect the DOM structure, and interpret the accessibility tree, all to gather the information before reporting back to the user.

Google’s agent-friendly guidelines come down to one principle: if an agent cannot see it clearly and click it confidently, it will skip it.

Google links to a separate guide on agent-friendly website best practices. Although it is worth reading alongside this, here is a summary of the best practices:

Use <button> and <a> tags for interactive elements. If you must use a <div>, give it role="button" and a tabindex. Agents are trained to recognize semantic HTML as actionable.
Set cursor: pointer in CSS on clickable elements. This is treated as an explicit actionability signal.
Keep your layout stable. Agents that take screenshots get confused by elements that shift position across pages, like a cart button that appears in a different location per product category.
Always link <label> tags to their inputs using the for attribute. This tells the agent what a form field is actually for.
Make sure interactive elements are larger than 8 square pixels. Visual analysis filters out anything smaller, so a tiny button effectively does not exist to an agent.
Remove transparent overlays and ghost elements. If a layer covers an interactive element, even a fully transparent one, the agent may discard the node underneath it entirely.

Additionally, the Universal Commerce Protocol (UCP) is emerging as a standard that will allow Search Agents to do significantly more on commercial sites. If you run an e-commerce site or any site where a user might want to complete a transaction, this is something to track.

This is not something you need to act on today. But it is the direction things are heading, and being aware of it now means you are not late to the party.

The Summary: What to Actually Do

The foundation has not changed.

Build a site that is technically clean, fast, crawlable, and can be indexed.
Create content that has a real point of view and provides something a generic summary cannot.
Make your most important facts explicit through structured data.