Does GEO replace SEO?

No. Generative engines search and read pages before answering, and many of them start from a traditional search index. Without crawling, indexing and authority there is nothing to cite. GEO is a layer on top of SEO.

Can I measure whether AI is citing my site?

Partially. Visits from chatgpt.com, perplexity.ai and similar show up as referrals in analytics, and Google Search Console includes AI features in the Web performance report. It is also worth manually testing the key questions of your niche from time to time.

Should I write for robots instead of people?

No. The techniques that work (direct answers, sources, numbers, clear structure) are the same ones that make a text better for human readers. Keyword stuffing, on the other hand, performed poorly in the GEO study itself.

Does blocking AI crawlers in robots.txt remove my site from answers?

It depends on the crawler. Many companies separate their training crawler from their search crawler. Blocking the training one (like GPTBot) does not block the search one (like OAI-SearchBot). The next article covers each of them.

GEO: how to get your site cited by ChatGPT, Perplexity and Google AI Overviews

When someone asks ChatGPT “which framework should I use for a fast company website?”, the answer is not a list of ten blue links. It is a paragraph with two or three cited sources. Being one of those sources is what GEO is about.

What GEO is, in one sentence

Generative Engine Optimization is the set of practices that make content retrieved, used and cited by systems that generate answers with AI, such as ChatGPT Search, Perplexity, Copilot and Google’s AI features in Search (AI Overviews and AI Mode).

The term comes from an academic paper by researchers at Princeton, Georgia Tech, the Allen Institute and IIT Delhi, published in 2023 and presented at KDD 2024 (Aggarwal et al., “GEO: Generative Engine Optimization”).

What the research found

The authors built a benchmark with thousands of queries and measured how much each source showed up in generated answers after applying different rewriting techniques. The findings that matter most to writers:

Technique	Observed effect
Citing reliable sources	Among the largest visibility gains
Adding direct quotes from experts	Among the largest visibility gains
Adding statistics and numbers	Among the largest visibility gains
More fluent, easier-to-read text	Moderate gain
Keyword stuffing	Little or no gain, sometimes worse

The best combinations reached roughly 40% more visibility on the study’s benchmark. The effect also varied by domain: fact-heavy content gained more from statistics, opinion content gained more from quotes.

The lesson is not “trick the model”. It is that language models prefer passages that already read like a well-supported answer.

What search engines officially say

It is worth separating research from official documentation. Google’s guidance on AI features in Search is direct: there is no special optimization and no specific markup. To be used as a source, a page must be indexed and eligible to show a snippet in regular search. The controls are the usual ones: nosnippet, data-nosnippet and max-snippet.

In practice, GEO starts with technical SEO. If the crawler cannot read your page, neither can the AI. The previous article’s checklist is still the foundation.

The techniques I apply

1. Answer first, context later

Models retrieve passages, not whole pages. If the answer only shows up in the fifth paragraph, the retrieved passage may not contain it. So:

the first paragraph after each h2 answers that heading’s question;
headings are written as clear questions or statements, not puns;
every article opens with key takeaways, like the box at the top of this page.

2. Verifiable facts, with sources

Concrete numbers (“LCP ≤ 2.5 s”), dates (“since March 2024”) and proper names (“KDD 2024”) are exactly what AI looks for to support a claim. Whenever possible, link to the primary source: official docs, research paper, repository.

3. One idea per block

Short paragraphs, lists and tables make each passage self-contained. A comparison table is practically a ready-made answer for “what is the difference between X and Y” questions.

4. Explicit authorship and dates

Who wrote it, what their experience is and when the text was last reviewed. This shows on the page (byline, date, “updated on”) and in structured data (author, datePublished, dateModified). Content with no owner and no date is harder to trust, for people and machines.

5. Real questions, short answers

A FAQ section with two-to-four sentence answers covers the question variants people ask assistants. Marked up as FAQPage in JSON-LD, it also makes the question → answer structure explicit.

6. Consistency across sources

Assistants cross-check information. If your site says one thing, LinkedIn another and GitHub a third, trust drops. Name, role, services and links must match everywhere. That is what sameAs in Person or Organization JSON-LD is for.

A before-and-after example

Before:

Performance is very important nowadays and can make all the difference for your business, so taking care of it is essential.

After:

Google considers a site fast when LCP stays at or under 2.5 seconds for 75% of visitors. Above 4 seconds, the experience is rated poor (source: web.dev).

The second passage has a number, a threshold, a criterion and a source. It is the kind of sentence a model can cite without rewriting it.

How to track it

Analytics: create a segment for referrals from chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com.
Search Console: traffic from Google’s AI features is included in the Web performance report.
Manual tests: once a month, ask assistants the 5 to 10 most important questions in your niche and note who gets cited.

GEO is still a young field and the rules change fast. What does not change is the principle: clear, verifiable, well-structured content is easier to cite. In the next article I cover the technical side: llms.txt, AI crawlers in robots.txt and structured data.