What I Think About Google’s “Optimizing your website for generative AI features on Google Search” Guidance

It’s been a while since I’ve written a blog post on my website. This is partly because my current job as VP of digital marketing takes a lot of my time and focus.

Having said that, I felt like Google’s recent document on GEO, “Optimizing your website for generative AI features on Google Search,” presents a good opportunity to have a discussion.

For those close to this topic, you’ll already notice a point of contention in what I’ve written — I called the document a “GEO” document, when Google’s authors say right there in the document that “optimizing for generative AI search is optimizing for the search experience, and thus still SEO.”

I won’t take the time to summarize the document here, because if that’s what you’re looking for, you can find that analysis already from smart SEOs like Marie Haynes:

Rather, what I want to focus on is some of the SEO/GEO tactics I’ve been employing and how they align with Google’s guidance.

In the document, Google first talks about RAG (retrieval augmented generation) and query fan-out. In my work, I’ve concentrated on creating content to feed grounding queries (the type you can see in Bing Webmaster Tools now).

Basically, I look at content not so much as a destination for users but as an input to AI assistants when they’re doing web search. That said, I still believe in optimizing for users first, so it’s a balance.

I also believe in mapping query fan-outs to the buyer’s journey — you can see such queries in tools like Profound or using other tools like Qforia — and trying to fill in gaps along that journey with owned or earned media.

Having said that, the content needs to be helpful, reliable, and people-first, i.e., non-commodity content.

This is Google’s next point in their article — follow SEO best practices for content creation to achieve AI visibility success. This includes:

Providing a unique point of view — I do this by having an expert review the content (or author it) to get their personal insights, including original data (especially charts), or drawing on firsthand experience (like I’m doing here, telling you about how I run SEO/GEO campaigns).

Creating helpful content — I do this by understanding what an audience is looking for and satisfying that search intent, but doing so with a unique point of view, as discussed above.

Organizing content in a helpful way for readers — I do this by following a logical heading structure (though I didn’t really take that advice in this article), writing in a BLUF style (bottom line up front), using semantic triples (subject, predicate, and object), and taking advantage of lists, charts, or other formats that break up the content into readable sections. (Yes, “readable sections” means semantically coherent “chunks” or passages.)

Adding valuable images or video — I firmly believe in using custom images for pages or articles. With ChatGPT Image 2.0, there are no excuses for not generating a unique image — bonus points if you took the photo IRL. And of course, certain AI assistants love YouTube or organic social content in a video format, so if you have the resources to produce this, definitely do so.

Focusing on what users want — This means not creating content for the sake of production, but because it fulfills an actual need or knowledge gap in the world. It’s hard work to find genuinely untapped topics, but you can reference reviews, CX tickets, Reddit threads, or other internal sources to find such insights. And if you must, you can always create a better version of content than what’s currently available — more helpful information consumable in less time.

Not using AI to create scaled content in an abusive way — I never (I mean hardly ever) use AI tools to generate content for my blog. Everything is 100% written by me. Having said that, AI assistants like Claude are great for writing assistance, and they’re widely used by content professionals. The trick is to use them smartly, creating content that contributes value, rather than just rehashes what’s already been said.

The next section of Google’s document focuses on technical structure.

The first step is ensuring content is crawlable and indexable, so basically don’t block AI crawlers in robots.txt or throw a noindex tag on a page. Pretty straightforward but still essential to check.

You can verify your site’s robot.txt configuration using free tools like CrawlerCheck.

You also want to use semantic HTML tags if possible. Here are some common ones, compliments of Gemini:

Another area of focus is JavaScript. Basically, you want to follow best practices, but that may only get you so far. Not all AI assistants render client-side JavaScript, so if your site is loading critical content that way, consider an alternate solution, like server-side JavaScript.

Other technical optimizations include good page experience, like passing Core Web Vitals, especially on mobile. Slow loading pages can trigger a 499 error, which Mike King has written about extensively.

There’s also a note in Google’s document about duplicate content. Nothing new here except that if you’re creating markdown file versions of your pages, I do wonder if those could be seen as duplicate, so just something to consider. And if they’re not, they could be considered as cloaking, so something else to consider. That said, I know people are experimenting with this tactic, so we’ll see where it all lands.

Google’s document also mentions that Google Business Profiles and Merchant Center feeds can be used to influence AI visibility. The standing advice here is to fill these out as completely and accurately as possible, and keep your GBP active over time, like by responding to reviews or even posting updates.

Lastly, we have a bunch of myths that Google busts in their documentation.

The first is an LLMs.txt file. They say it’s not needed. Personally, I have one on my website generated by Yoast, and I’ve experimented with them on other sites. I believe these files can be read by certain agents, Claude Code, or even ChatGPT, so while you don’t have to create an LLMs.txt file, there’s not really a downside to doing so, except for your time in creating and maintaining it.

Google also says special markdown versions of pages isn’t needed. Maybe if you have a really bad page-load time and your content isn’t getting chosen by AI assistants, you might give this a try as an experiment, but it’s not like a general best practice or anything. I personally haven’t used it for website content, though I have turned documents into markdown to feed them into AI assistants.

There’s also a note about writing specifically for AI systems. The document calls out using long-tail keywords specifically as an unnecessary optimization — if done for its own sake. Personally, I still advise writers to use semantic triples and BLUF for semantically coherent passages. They may do this anyway for the sake of readers, but I believe it does help with making content more extractable for AI assistants.

Another note is on inauthentic mentions, basically the spammy backlinks of the AI era. I’ve always been an advocate for quality and authenticity when it comes to backlinks, and I believe the same applies to mentions. Basically, if it’s easy to get, or anyone can get it, it’s unlikely to have much value. It’s those rarer instances where a legitimate brand mentions you (hopefully for a positive reason) that carry weight.

Lastly is a point about fixating on structured data. My opinion is that structured data doesn’t really contribute to AI visibility in the short term. I believe most schema gets stripped in training data or converted to tokens like any other text in the retrieval stage. Having said that, structured data contributes to entity clarity in knowledge graphs, and that’s the real and long-term value of it. I advise to use all eligible schema types, like organization, person, FAQ, how to, product, review, etc. with the goal being to create entity clarity and build a knowledge graph presence.

That about covers it. For advanced SEO/GEO practitioners, this article probably didn’t introduce you to much new information, but that’s all right. My opinion on SEO/GEO is that following the fundamentals well is what works, not any particular shortcut or hack.

I encourage you to use tools like Profound and Bing Webmaster Tools to understand your AI visibility, spot gaps with competitors, and focus one earned and owned media strategies to fill them.

I’ll return to this article to update it’s information or improve the formatting as I have time.

Until next time, enjoy the vibes:

Thanks for reading. Happy optimizing! 🤗

Related posts