SEO & AI

GEO: get cited by ChatGPT, Gemini, Perplexity

GEO explained simply: what Google and OpenAI recommend, 7 concrete actions to be cited by AI systems, and how to measure your results over time.

Published on 17 October 2026 · 5min read

GEO (Generative Engine Optimization) covers practices that raise your odds of being cited or summarised in answers from ChatGPT, Gemini, Perplexity or Google AI Overviews. Good news: there is no secret recipe. Google itself says classic SEO fundamentals still apply, and a few writing habits make a real difference.

What is GEO, and where does the term come from?

The term comes from an academic paper, “GEO: Generative Engine Optimization”, published in 2023 by researchers from several universities (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande) and accepted at KDD 2024. The authors report that GEO can increase visibility by up to 40% in generative-engine answers.

The study tested rewrite tactics: citing sources, adding quotations, adding statistics, improving fluency, simplifying language, using a more authoritative tone or technical vocabulary. One result is especially clear: classic methods such as keyword stuffing work poorly. Important caveat: results come from a research bench and vary by domain. Treat them as leads, not guarantees.

What Google says officially

In its documentation on AI features in Search, Google writes that there is “no additional requirement” to appear in AI Overviews or AI Mode, and no “special optimization”. It recommends sticking to the basics:

Allow crawling: robots.txt, and your CDN or host, must not block bots.
Create useful, reliable, people-first content.
Offer a good page experience.
Make sure structured data matches visible text.
Keep important information as text.

Google also explains that these features may use “query fan-out”: several related searches run across subtopics, which can surface a wider set of links. Covering a subject in depth, with clear sections per sub-question, gives you more chances to be cited.

Seven concrete actions to be cited by AI

Answer first. Put a direct answer in the first 100 words, then expand.
Structure. Use H2 and H3 headings phrased as real questions, lists and tables: engines extract answers more easily from well-bounded blocks.
Support claims with sources. Citing reliable references and verifiable figures is among the tactics studied. Add outbound links to official references.
Show who is speaking. Name the author or company, publish and update dates, and mark them up (BlogPosting JSON-LD with datePublished and dateModified).
Let bots in. Check robots.txt and firewall rules. At OpenAI, OAI-SearchBot is used so sites can appear in ChatGPT search results, while GPTBot is used for model training—two distinct settings.
Keep essentials as text. A table or fact only visible in an image, PDF or late-loaded component is harder to understand.
Refresh. Update guides when rules, prices or figures change, and show the update date.

Here is a robots.txt example that allows appearance in ChatGPT search while refusing training:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Whether to allow training is a strategic choice per company: OpenAI’s bot documentation describes each agent and its role.

How to measure results

There is still no standard “rank” metric inside an AI answer. A simple honest method:

Prepare 20 questions your customers actually ask.
Ask them monthly in ChatGPT, Gemini and Perplexity and note whether your site is cited, and with which excerpt.
Watch traffic in analytics and Search Console, and spot crawler visits in server logs.

Example: rewriting a paragraph so it is citable

Before: “Our agency offers innovative solutions to improve your online presence with a tailored approach.”

After: “A site loads faster when its images are compressed and its main content is server-rendered. Google treats an LCP of 2.5 seconds or less as good (web.dev). We apply both levers on every site we ship.”

Why the second version helps: it has a precise claim, a verifiable figure, an identifiable source and a named entity. An engine can reuse the passage in an answer. The first version says nothing you can cite. For the threshold detail, see our Core Web Vitals guide.

Checklist before you publish an article

The direct answer appears in the first 100 words.
H2 and H3 headings match real questions.
Every figure has a cited source with a link.
Author or company, publish date and update date are visible.
JSON-LD markup matches displayed text.
Important content is text, not only an image.
The site is in the XML sitemap and not blocked in robots.txt.
A FAQ section answers the most common follow-up questions.

What not to do

Stuff keywords into text: the GEO study shows this works poorly.
Publish generic mass-produced unverified content: Google asks for useful, reliable, reader-first content.
Add structured data that does not match visible content.

Our approach at Data Forge

We apply these principles on our own pages: server-side rendering so text is readable by every bot, JSON-LD structured data (Organization, BlogPosting, breadcrumb), XML sitemap, visible publish dates, and images with declared dimensions. Want to audit or rebuild a site for classic search and AI engines? Talk to us about your project.

A solid technical base starts with performance: read our guide on Core Web Vitals and INP. Treat citation monitoring as a recurring editorial ritual rather than a one-off experiment. When an assistant cites a competitor, compare their structure: direct answer, sources, dates, and clarity of ownership. Improve your page on those axes before chasing new keywords. Prefer original examples from your Algerian market context—payment rails, delivery constraints, local law—so summaries that quote you remain useful. Keep claims conservative: engines amplify confident wording, and overstated statistics damage trust when readers check the underlying source. Review your top ten commercial pages quarterly with the same twenty-question panel and store screenshots of cited excerpts for the team.

Frequently asked questions

Does GEO replace SEO?

No. Google says the same fundamentals apply to its AI features. GEO extends SEO with writing practices (direct answers, sources, figures) and machine readability.

Does blocking GPTBot stop you appearing in ChatGPT?

According to OpenAI, GPTBot relates to model training and OAI-SearchBot to appearance in ChatGPT search results. They are separate settings to configure by goal.

How do you know if an AI cites you?

Regularly test your key questions in several assistants, compare displayed sources, and keep a month-to-month record.

Are structured data enough?

No. They help engines understand the page, but Google says they must match visible text. Content remains central.