Optimizing for Google AI Search: What Works for AI Overviews in 2026

Short answer: there is no separate “AI optimization” for Google. In May 2026 Google published an official guide to optimizing for generative AI features in Search, and its main point is simple: AI Overviews and AI Mode pull pages from the same index as regular search results, so the same SEO principles apply. Below is what Google actually recommends, which popular “GEO hacks” it explicitly calls unnecessary, and what we saw in our own Search Console report for moveiton.net.

How Google builds AI answers

According to Google, generative features in Search rely on two mechanisms:

• Retrieval-augmented generation (RAG, or grounding). The model doesn’t make up an answer: Google’s core ranking systems first retrieve relevant pages from the index, and the answer is built from their content — with prominent links to the sources.
• Query fan-out. The model generates several related queries and fetches additional results for them. For example, for “how much does a company website cost” it may separately search “what makes up the cost of web development” or “what a turnkey website includes”.

Hence the first practical requirement: to appear in an AI answer, a page must be indexed and eligible to be shown with a snippet. If it has noindex or nosnippet, it won’t appear in AI features either. Google also notes that the site must be included in generative AI features in Search Console settings.

What actually works: Google’s recommendations

1. Content no one else has (non-commodity)

Google explicitly separates “commodity” and “non-commodity” content. The first is a retelling of common knowledge that anyone or any AI model could write (“7 tips for…”). The second is first-hand experience, your own data, specific decisions and their outcomes. According to Google, this will influence your presence in generative search over the long run more than anything else.

For a web studio this means case studies with real numbers, breakdowns of our own projects, prices from our practice, and mistakes we learned from. Not “what is SEO”, but “what happened to a client’s traffic after moving off a template website”.

2. Structure for people, not for the algorithm

Paragraphs, clear headings, a logical flow. Google stresses that there is no ideal page length, and semantic HTML is nice to have but not required.

3. Images and video

AI features in Search can show images and videos from your page, which gives you extra entry points. People also increasingly search by photo with Google Lens and Circle to Search. For a web studio, that means case-study screenshots with descriptive captions and alt text, and short video walkthroughs of projects.

4. Technical basics and page experience

Googlebot must get the page with a 200 status code, without robots.txt blocks, with content in the HTML or correctly rendered JavaScript. The page should load fast on a smartphone, and the main content should be easy to tell apart from everything else. Nothing new here: these are Google Search’s standard technical requirements.

5. Business Profile and product feeds

For local businesses and e-commerce, Google recommends keeping your Google Business Profile and Merchant Center up to date: this data can also appear in AI answers.

What doesn’t work: GEO and AEO myths

For Google, GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization) are simply SEO. The guide has a separate section on what you don’t need to spend time on:

Advice from the internet	What Google says
Create llms.txt or special “AI files”	Google Search doesn’t use them. They neither hurt nor help — keep them for other services if you like.
Split text into small “chunks” for AI	Not needed: Google’s systems understand multiple topics on one page and show the relevant part.
Rewrite texts for AI and add every query variation	Not needed: AI understands synonyms and meaning. Mass-producing pages for every variation or fan-out query is treated by Google as spam (scaled content abuse).
Build up brand “mentions” across the web	Inauthentic mentions don’t help: ranking focuses on quality, and spam is filtered out.
Add special schema markup for AI	There is no special markup for AI. Structured data helps with rich results but isn’t required for AI answers.
Third-party tools with “internal Google metrics”	No third-party tool has access to Google’s internal ranking systems.

An honest correction

In the previous version of this article we claimed that key facts must appear in the first 150–200 words, and that without LSI structuring the Google-Extended robot “won’t add the page to the vector database”. That was wrong, and we are correcting it. Google-Extended is a robots.txt token that controls whether content is used to train Gemini models and for grounding in Gemini Apps and Vertex AI. According to Google’s documentation, it does not affect a site’s inclusion in Google Search and is not a ranking signal. There is also no “first 150 words” or “LSI within 14 days” rule anywhere in Google’s official materials.

What the Generative AI report in our Search Console showed

Since August 31, 2026, Google has made its generative AI performance report available to all websites. Here is what we saw for moveiton.net over three months (July 7 — October 6, 2026):

• 768 impressions in AI Overviews, AI Mode and Discover’s AI features. The report shows impressions only; there are no clicks in it.
• Noticeable growth from mid-September. This coincided with fixing our hreflang setup and adding location signals to service pages, although the report doesn’t prove a direct link.
• Pricing articles appear in AI answers most often: “Online store development cost in 2026” — 152 impressions, “How much does website development cost in 2026” — 81. Next come the home page (73) and the Russian SEO services page (45).
• Multimodal search (searching by image with Lens or Circle to Search) — 44 impressions with an average position of 2.1.

Our takeaway: generative search readily cites pages with specifics from real practice — prices, ranges, what drives the cost. Generic “about everything” texts rarely get there. So our next step is to rewrite portfolio case studies in a “challenge — solution — numbers” format and publish fewer but deeper blog articles.

Checklist: what to do for your website

1. In Search Console, check that your important pages are indexed and not blocked with noindex or nosnippet.
2. Open the Generative AI report and see which URLs already appear in AI answers. Strengthen those first.
3. Find pages that merely retell common knowledge and add your own experience: numbers, examples, decisions, prices.
4. Add your own images and videos with descriptive captions.
5. Check speed and the mobile version: Core Web Vitals in Search Console or PageSpeed Insights.
6. Update your Google Business Profile: the real business name without keywords, correct categories, a current description.
7. Don’t pay for “GEO optimization” built on llms.txt, “chunking” and mass-generated pages for queries.
8. Add annotations in Search Console on the day of every change so you can see what worked.

How to measure results

Your main tool is Search Console: the Generative AI report for AI features, the “Multimodal” search type filter in the Performance report, and chart annotations. To separate branded traffic, use a regex query filter with your brand name. And don’t stop at clicks: Google itself recommends looking at conversions — inquiries, calls, sales.

If you’d like us to look at your website from this angle, order SEO promotion or start with an audit: we covered typical technical issues in “How technical website errors drain your SEO budget”.