ChatGPT Search and Google AI Overviews are changing SEO: Habr AI released a guide to GEO

On Habr AI, a detailed guide to GEO—optimizing content for LLM search—was published. The main idea is simple: ChatGPT Search, Perplexity, and Google AI Overviews increasingly answer users directly, so the battle is no longer just for a click in search results, but for inclusion in the model's response.

Why GEO Emerged

The author cites these markers of search restructuring in 2024–2026:

• AI blocks from Google seized the top of the screen
• ChatGPT Search reached over 100 million active users in search mode by April 2026
• Perplexity reached 15+ million queries per day
• Zero-click searches reached 65%
• Some websites are already losing 20–40% of organic traffic

At the same time, a new channel emerges—AI-referral, when a user arrives via a link from an AI model's response.

GEO, AIO, AEO, and LLMO are related terms, but they mean the same thing: content must be convenient not only for the reader and search robot, but also for the language model that assembles an answer from fragments of different pages. Unlike old SEO, where you could win through keywords, meta tags, and link building, clarity of meaning, expertise, and fact extractability are now more important.

How Models Choose

The article breaks down a typical RAG scheme: the system receives a query, searches for relevant pages, ranks them, cuts text into 200–500 token blocks, inserts the best pieces into a prompt, and only then generates an answer with citations.

The key consequence: it's not the entire article that competes, but each individual paragraph. If a needed fact is hidden deep, diluted with filler, or not structurally separated, it may simply not get into the model's context, even if the page itself is strong.

The article shows that old SEO patterns like keyword density, thin pages targeting a single query, and link schemes work poorly in LLM search. The spotlight shifts to signals that help the model quickly understand the text's meaning and trust it. In other words, the winning material is not where a keyword phrase is repeated more often, but where the thesis is stated directly, supported by figures, and easily translatable into an AI answer without loss of context.

"Optimization for algorithm gives way to optimization for understanding".

• Domain and author authority: E-E-A-T and accumulated citation history increase trust.
• Structural extractability: clear headings, definitions, FAQ, and schema markup help extract facts mechanically.
• Semantic density: less filler, more independent semantic units in each paragraph.
• Verifiability: dates, figures, studies, and specific links increase the chance of citation.
• Freshness: current data receives priority in systems working with real-time web indexing.

What Editors Should Do

Practical recommendations sound fairly straightforward. The article advises giving a direct answer in the first 200–300 words, moving definitions to the beginning of sections, adding FAQPage, HowTo, Article, and Person schema, and assembling materials into topic clusters instead of scattered pages targeting a single query. Special emphasis goes to the author page: byline, update date, profiles, and publications are needed not for aesthetics, but as machine-readable trust signals.

The first step for an editorial team is to check 10–15 key queries in ChatGPT Search and Perplexity and see who these services are citing right now. Next, it makes sense to set up a separate AI traffic segment in GA4, go through top materials, and add direct answers, facts with dates, and clear confirmations to them.

Next, restructure the content core into pillar pages, cluster pages, and explicit internal links so that the topic is read as a unified expert body rather than a random collection of publications.

The author suggests measuring progress with new metrics: presence in AI Overviews, volume of AI-referral traffic, frequency of domain citations, and share of answers where the brand appears. This is an important shift for editorial teams and content teams: KPIs can no longer be reduced to just SERP position and overall organic traffic. You need to understand how often language models choose your material as the baseline source for an answer.

What This Means

GEO does not cancel technical SEO, but shifts the goal of optimization. Winners will not be those who inserted keywords more precisely, but those who publish dense, verifiable, and well-structured expert content. For media, blogs, and B2B websites, this is a direct signal: if material is not suitable for paragraph-by-paragraph citation, it will increasingly not be visible to users.