Guide
GEO: how to make the AI cite you — and how to know whether it did.
GEO, or Generative Engine Optimization, is the set of practices that let AI models such as ChatGPT, Gemini, Claude and Perplexity find, understand and cite your content when answering questions. It depends on a direct answer, clear structure, an identifiable author, verifiable sources, original data and consistency across pages — and it is measured by putting questions to the AIs and recording whom they cite.
What GEO is (and what it is not)
GEO stands for Generative Engine Optimization: optimising content for the engines that generate answers, not only for those that return a list of links. The term appeared between 2023 and 2024, when it became clear that a share of searches ended inside a model's answer, with no click on any site — and that the sites cited in that answer took the attention that used to go to the first result.
GEO is not a technical trick. There is no hidden tag, no magic word in the code, no way to "declare" that you are the right source. What exists is content a model can retrieve, understand and attribute: a passage that answers the question completely, a structure that separates a definition from a step and from a comparison, an author with a verifiable credential, sources that are still live, and a set of pages that do not contradict one another.
It is also not a synonym for AEO, although the two acronyms get confused. AEO (Answer Engine Optimization) came first, aimed at the answer boxes in search engines — that highlighted passage at the top. GEO is broader: it deals with models that write a new answer from several sources and decide which of them to name. Anyone already working for the answer box started ahead; anyone who never structured content starts from zero on both.
GEO — Generative Engine Optimization (GEO) is the practice of structuring content, authorship and sources so that generative artificial intelligence models find, understand and cite that content when answering users' questions.
How the AIs choose whom to cite
No vendor publishes the full criteria, and anyone promising the formula is selling something. What can be stated comes from the observable behaviour of models with search, the public documentation of search engines, and repeated testing. That is what follows — and only that.
Diagram of how an AI answer is put together.
| What counts | Why we know | What to do |
|---|---|---|
| The passage answers the whole question | Answers cite short, self-contained paragraphs, not whole pages. | Open every article with an answer of up to 60 words that makes sense out of context. |
| The page is retrievable | Models with search start from the search engine's index; what isn't indexed never enters the answer. | Ensure indexing, a stable URL and content in the HTML, not only after JavaScript runs. |
| There is an identifiable author | Google's public quality guidelines describe experience, expertise, authoritativeness and trust; AI answers favour pages with explicit authorship. | Sign with a real person, with a credential and their own page, marked up as Person. |
| Claims have sources | Models prefer material that already rests on verifiable references, because it lowers the risk of the answer. | Cite an official body, literature or primary data, with a live link. |
| The content is distinguishable | When ten pages say the same thing, the answer cites the one with a figure, a method or an angle the others lack. | Include an original contribution in every article: your number, your case, your comparison. |
| The entity is consistent | The same name, with the same credential, across several domains raises confidence in the attribution. | Keep name, credential and markup identical across all your sites. |
The passage answers the whole question
Answers cite short, self-contained paragraphs, not whole pages.
Open every article with an answer of up to 60 words that makes sense out of context.
The page is retrievable
Models with search start from the search engine's index; what isn't indexed never enters the answer.
Ensure indexing, a stable URL and content in the HTML, not only after JavaScript runs.
There is an identifiable author
Google's public quality guidelines describe experience, expertise, authoritativeness and trust; AI answers favour pages with explicit authorship.
Sign with a real person, with a credential and their own page, marked up as Person.
Claims have sources
Models prefer material that already rests on verifiable references, because it lowers the risk of the answer.
Cite an official body, literature or primary data, with a live link.
The content is distinguishable
When ten pages say the same thing, the answer cites the one with a figure, a method or an angle the others lack.
Include an original contribution in every article: your number, your case, your comparison.
The entity is consistent
The same name, with the same credential, across several domains raises confidence in the attribution.
Keep name, credential and markup identical across all your sites.
One honest caveat: the models change, and what holds today may change in six months. That is why the part that doesn't change is measurement — if you ask every month, you notice the change before you lose ground.
GEO × SEO
| What counts | SEO | GEO |
|---|---|---|
| Unit | The page that appears in the list | The passage that gets cited |
| Core signal | Links and relevance | Clarity and verifiable authority |
| Metric | Position and clicks | Share of answers and citations |
| Structure | Headings and keyword | Direct answer, definition, FAQ, table |
| Author | Recommended | Decisive |
| Source | Helps | Required |
| Consistency across pages | Solved with a canonical | A coherent entity across every domain |
Unit
SEO: The page that appears in the list
GEO: The passage that gets cited
Core signal
SEO: Links and relevance
GEO: Clarity and verifiable authority
Metric
SEO: Position and clicks
GEO: Share of answers and citations
Structure
SEO: Headings and keyword
GEO: Direct answer, definition, FAQ, table
Author
SEO: Recommended
GEO: Decisive
Source
SEO: Helps
GEO: Required
Consistency across pages
SEO: Solved with a canonical
GEO: A coherent entity across every domain
They are not rivals. GEO is the SEO that survived the arrival of AI: the good practices still hold, and some that were optional became mandatory.
The six signals that count
Direct answer
The first paragraph answers the question in the title, on its own, in up to 60 words. That is the passage the AI copies.
- Before
- In this article we're going to talk about anxiety, a subject that worries a lot of people and deserves attention. Read on to understand it better.
- After
- Anxiety is the body's response to a perceived threat, with physical and mental symptoms. It becomes a disorder when it is disproportionate, persists for months and disrupts work, study or relationships. The treatment with the strongest evidence combines psychotherapy and, in some cases, prescribed medication.
Typed structure
Definition, steps, comparison and FAQ in recognisable blocks. Running prose forces the model to guess where the information begins.
- Before
- A 1,500-word text in continuous paragraphs, with no subheadings, no list and no table.
- After
- A definition at the top, three numbered steps, a table comparing the options and five FAQs at the end.
Identifiable author
A real name, a verifiable credential and a page of their own. "Editorial team" is not an author.
- Before
- By the Content Team · no page, no credential, no markup.
- After
- By a professional with board registration, an authority page on the site, Person markup and sameAs to ORCID and Lattes.
Verifiable sources
Every claim that isn't obvious points to an official body, a publication or primary data — with a link that still works.
- Before
- "Studies show that most patients improve."
- After
- "A review published in {exemplo} reports improvement in {n}% of cases" — with a link to the publication and the date accessed.
Your own data and method
What you measured, the case you ran, the criteria you use. It is what makes your page the cited one, rather than the one that summarised the same subject.
- Before
- A summary of what the first five results already say.
- After
- A figure you measured, a case from your own operation, or a decision criterion only your practice produces.
Consistency across pages
One main article per topic and satellites with distinct angles. Eight similar texts compete with one another and confuse the attribution.
- Before
- Eight articles on the same subject, with near-identical titles, across different sites in the same network.
- After
- One complete guide, satellites with their own angles, all linked to the guide, with no cannibalisation.
How to measure
Without measurement, GEO is opinion. The measure is simple to describe and tedious to run by hand — which is exactly why it usually doesn't get done.
- 1
Build a set of questions
Ten to twenty real questions from your field, per topic, in the languages of your sites. Questions a client would ask, not keywords.
- 2
Ask the four models
ChatGPT, Gemini, Claude and Perplexity answer the same question differently. Record the whole answer, the date and who was cited.
- 3
Repeat every month
The value isn't in the snapshot, it's in the series. A single measurement can't tell you whether you went up or down.
- 4
Work out the share
In how many of the topic's questions does your brand appear? And the competitor? That ratio is your share of answers.
- 5
Turn the gap into a topic
Every question where the AI cited someone else and not you is a missing article. It is the most objective work queue there is.
Common mistakes
Volume with no author
Publishing a lot, signed by nobody. It multiplies the cost and builds no authority.
Repeating the keyword
Stuffing the text with terms doesn't help the model understand; it makes the text worse for the person and for the machine.
Hiding the answer at the end
Three paragraphs of introduction before answering. The model cites whoever answers first.
Dead sources
A link that stopped existing drags down the credibility of the whole page. Check them periodically.
Twenty identical articles
The network competes with itself and none of the texts accumulates authority.
Measuring once and stopping
With no series, you can't tell whether the change you made had any effect.
GEO checklist
Twelve items. An article that meets all of them has a real chance of being cited; one that meets half depends on luck.
- A direct answer of up to 60 words at the top, which works out of context.
- A title that is a question or a definition, not a pun.
- An explicit definition of the main term, in one sentence.
- FAQs with real questions from your audience, not invented ones.
- A table whenever there is a comparison between options.
- An author with a name, a credential and their own page, marked up as Person.
- A "reviewed by" when the topic affects health, money or rights.
- At least two verifiable sources, with live links.
- One original contribution: your data, your case or your method.
- Links to the topic's main article and to its siblings.
- Valid markup for the article, the author and the FAQ.
- A question set measured every month, recording who was cited.
How CDE does each item
This is the only product block on the page. Each line of the checklist above matches something the system does by default — and that you can verify by opening any published article.
| Checklist item | What CDE does |
|---|---|
| Direct answer | The studio generates the direct-answer block at the top and QA fails an article without it. |
| Title as a question or definition | The brief defines the question the article answers before any text exists. |
| Explicit definition | The article's typed structure has a definition block. |
| Real FAQs | The questions come from what the audience asks in search and in AI tools, not from imagination. |
| A table for comparisons | The article type determines the blocks; a comparison generates a table. |
| Author with a credential | Nothing publishes without an authorised expert; the authority page and the Person markup ship with it. |
| Reviewed by | Risky topics require a qualified reviewer before publishing. |
| Live sources | Allowed sources are configured per field and the link is checked. |
| Original contribution | A required field in the brief: without it, no text is generated. |
| Internal links | The brief's link plan is checked by QA. |
| Valid markup | The markup is generated with the article and validated before publishing. |
| Monthly measurement | The question set runs every month across the four models, and the gap becomes a topic. |
Direct answer
The studio generates the direct-answer block at the top and QA fails an article without it.
Title as a question or definition
The brief defines the question the article answers before any text exists.
Explicit definition
The article's typed structure has a definition block.
Real FAQs
The questions come from what the audience asks in search and in AI tools, not from imagination.
A table for comparisons
The article type determines the blocks; a comparison generates a table.
Author with a credential
Nothing publishes without an authorised expert; the authority page and the Person markup ship with it.
Reviewed by
Risky topics require a qualified reviewer before publishing.
Live sources
Allowed sources are configured per field and the link is checked.
Original contribution
A required field in the brief: without it, no text is generated.
Internal links
The brief's link plan is checked by QA.
Valid markup
The markup is generated with the article and validated before publishing.
Monthly measurement
The question set runs every month across the four models, and the gap becomes a topic.
FAQ
What people ask about GEO.
Find out which questions the AI should already be citing you in.
The check measures your presence in the four models' answers and shows the three highest-impact actions.