SEO in the age of AI answers: understanding GEO

AI answers: what AI Overviews, ChatGPT, and Perplexity change for SEO

With Google's AI Overviews, ChatGPT, and Perplexity, part of all searches now returns a directly written answer, no longer just a list of links.

In practice, the user often gets their answer without clicking. The sources still exist, but they appear as supporting links (the citations) next to or below the generated answer. The consequence for a publisher: being cited as a source becomes a goal in its own right, distinct from simply ranking among the blue links.

Two useful points to keep in mind, verified against Google's official documentation:

• AI Overviews are grounded in Google's search index. To be eligible as a supporting link, a page must be indexed and able to appear in Google Search with a snippet.
• No special structured data or dedicated file is required for these AI features: Google points back to existing SEO best practices.

In other words, the entry point remains largely the same: an indexable, high-quality site. What changes is the form of visibility (being cited in an answer) and the share of clicks you can expect from it.

SEO and GEO: difference and continuity

The term GEO comes from an academic paper, GEO: Generative Engine Optimization (Aggarwal et al., Princeton and others, arXiv:2311.09735, submitted on November 16, 2023, published at KDD 2024). It defines GEO as a framework for improving the visibility of content within the answers of generative engines.

The simple distinction:

Continuity matters more than rupture. Mainstream AI engines often rely on a search index (Google documents this for AI Overviews), so a site that cannot be indexed cannot be cited. The SEO fundamentals (useful content, crawlability, clean technical setup, authority) therefore remain the foundation of GEO.

The GEO paper reports that its methods can increase visibility by up to 40% in generative answers, with the most effective tactics being the addition of source citations, attributed quotes, and statistics. An honest caveat is in order: this experiment was run on a system imitating Bing Chat (validated on Perplexity) in 2023-2024, and not on Google's AI Overviews. The figure is serious in its context, but it should not be generalized to all engines in 2026.

How AI engines choose their sources

There is no single public recipe. What is established, and what is not:

Established (primary sources):

• Google AI Overviews: selection grounded in the search index. A page must be indexed and eligible for a snippet to serve as a supporting link. A documented nuance from serious analyses: the selection does not follow the exact order of the web ranking, and pages in low positions are sometimes cited.
• No required schema for Google's AI features: the official documentation is explicit.

Plausible but not official:

• Perplexity works on a RAG principle (web retrieval then reranking), favoring fresh, factual, well-structured, and authoritative sources. The general mechanism is widely described and coherent, but the quantitative details (index size, number of steps) come from unofficial third-party blogs: treat them with caution.

The reasonable common denominator: clear, factual, well-structured content, on a crawlable site that holds authority on its topic. This converges with SEO rather than breaking from it.

Five honest levers for GEO: direct answer, structure, structured data, E-E-A-T, crawlability

Five levers increase the probability of being understood, indexed, and judged relevant: answer directly, structure, mark up with schema.org, build authority (E-E-A-T), and ensure crawlability and indexability. None guarantees a citation.

• Answer questions directly. A short, self-contained, accurate answer at the top of the page is easier to reuse than a long preamble. This is consistent with the tactics validated by the GEO paper (citations, quotes, sourced statistics).
• Structure clearly. Explicit headings, lists, tables, short paragraphs: they make extraction easier for engines as well as for readers.
• Structured data (schema.org). An established role for content understanding and eligibility for rich results in classic search. Note: no proven causal link between schema.org and being cited by AI. Use it as a comprehension best practice, not as a promise of citation. Details in my guide on llms.txt and structured data.
• Authority and E-E-A-T. Real expertise, author signals, sources, mentions, and citations elsewhere on the web. It is slow to build but durable.
• Crawlability and indexability. A site that is blocked or poorly indexed cannot be cited by engines grounded in a search index. Check robots.txt, indexing, and load time.

For the version specific to conversational assistants, see being cited by ChatGPT and Perplexity. To automate part of this editorial and technical routine, see AI automation.

What I do not control

Honesty requires setting the limits:

• No citation guarantee. You can make content eligible and relevant, but never force an engine to cite it.
• Volatility. Generated answers vary depending on how the question is phrased, the moment, the engine, and its updates. A page cited today may no longer be cited tomorrow.
• Opacity. The exact selection criteria are not fully public, and some figures that circulate (citation rate by tactic, market shares of AI engines, schema-related multipliers) have no reliable primary source, so I do not repeat them.
• Hard to measure. Tracking your presence in AI answers remains a manual craft, see the FAQ below.

Going for Growth is run by Fabien Cavanna, a sole operator. This site applies these principles itself (structured data, llms.txt file), not as a guarantee of citation, but as technical consistency. I present these topics for what they are: solid practices on the fundamentals, and an emerging field where caution beats promises.

Frequently asked questions

Further reading

• Being cited by ChatGPT, Perplexity, and AI
• llms.txt and structured data for AI
• An AI agent in production: the Opposio case study