Skip to content
Etalo

Guide

How to get cited by ChatGPT

There is no button. There is a technical door that many sites close without knowing it, three editorial levers that have actually been measured, and a source mix moving so fast that no tactic pinned to a single site survives three months. Here is what is established, and what is not.

Founder of Etalo, SEO and generative visibility consultant

Read on 16 September 2026. Every study cited carries its date in the text.

The essentials in thirty seconds

To be cited by ChatGPT, your pages must first be reachable by the right crawler — OAI-SearchBot, which builds the ChatGPT Search index, not GPTBot, which collects training data. Then you need pages that answer a precise question on their own, with dated figures and attributed quotations: those are the two levers academic research has measured. And you need to measure, because the mix of sources ChatGPT cites changes from month to month.

  • OpenAI runs three crawlers, and they do different jobs: GPTBot (training), OAI-SearchBot (the ChatGPT Search index), ChatGPT-User (a page fetched when a user triggers browsing).
  • The most common mistake: blocking GPTBot believing it protects you, while leaving OAI-SearchBot open — or the reverse, blocking OAI-SearchBot and dropping out of the citation candidates without ever knowing why.
  • Two measured editorial levers (Princeton, Georgia Tech and the Allen Institute for AI, KDD 2024 conference, 10,000 queries): adding attributed quotations, up to +41% visibility; adding statistics, +25 to 30%. Keyword stuffing: no measurable effect.
  • The source mix is unstable: Reddit's citation frequency inside ChatGPT went from around 60% to around 10% in thirteen weeks (Semrush, July-October 2025).
  • Consequence: no tactic pinned to a specific domain holds. What holds is being readable, factual and verifiable — and measuring rather than assuming.

First, the technical door

Before any content strategy, a binary question: can ChatGPT read your pages at all? The answer sits in your robots.txt, and it often surprises — because OpenAI does not run one crawler but three, and their jobs have nothing in common.

Many sites blocked GPTBot in 2023, during the debate about model training. That was a legitimate decision. But it never had any effect on citations, because that crawler is not what feeds ChatGPT Search answers. Conversely, a site blocking OAI-SearchBot drops out of the citable pages, and nobody will tell it.

OpenAI's three crawlers, and what blocking each one does

All three are written into robots.txt like any other agent. Blocking them produces three different effects.

AgentWhat it is forBlocking it means
GPTBotCollecting data to train future models.Refusing to let your content serve training. No effect on ChatGPT Search citations.
OAI-SearchBotBuilding the search index ChatGPT uses to answer.Removing yourself from the citable pages. The most expensive block, and the least often intended.
ChatGPT-UserFetching a page live when a user triggers browsing.Preventing the reading of a page somebody explicitly asked ChatGPT to go and look at.

Source: OpenAI crawler documentation, platform.openai.com/docs/bots, read on 16 September 2026. OpenAI also publishes IP ranges for each agent: a declared user agent is not proof that the crawler is genuine.

What the research measured, lever by lever

Visibility gains measured on GEO-bench, a set of 10,000 queries, by researchers from Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi. Presented at the KDD 2024 conference.

Adding attributed quotations +41%
Adding statistics +25 to 30%
Stuffing the page with keywords no measurable effect
Source: Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024. The +41% is measured on the position-adjusted word count metric; the same lever is worth +28% on the subjective impression metric. The authors report up to +40% overall visibility, and +37% on Perplexity in real conditions.

What that study says, and what it does not

It says content that cites its sources, attributes its quotations and carries figures gets picked up more often than content that simply asserts. It also says keyword stuffing, a reflex inherited from search marketing, has no measurable effect on a generative engine — the authors explicitly conclude that methods need rethinking.

It does not say those figures transfer as they are to ChatGPT in 2026. The measurements date from 2024 and cover the generative engines of that time. What transfers is the meaning of the result: models pick up what is verifiable and attributed, not what is optimised.

And it says nothing about your market in particular. An average gain across 10,000 generic queries does not predict what happens on the thirty questions your buyers actually ask.

The source mix moves, and fast

Three measurements from the same field study: 230,000 prompts, over 100 million citations recorded across three engines between 14 July and 12 October 2025.

60 → 10%

Reddit's citation frequency inside ChatGPT over those thirteen weeks. Reddit is still one of its two most-cited domains — but the space it occupies was divided by six.

Semrush, October 2025

55 → 20%

the same collapse for Wikipedia, over the same period and on the same engine.

Semrush, October 2025

0

comparable change on Google AI Mode, where Reddit and Wikipedia stayed stable. The engines do not move together.

Semrush, October 2025

Source: Semrush, The Most-Cited Domains in AI, a thirteen-week study published in 2025.

Why a tactic pinned to one domain does not hold

The advice heard everywhere — "be present on Reddit, ChatGPT cites it all the time" — was true in the summer of 2025 and much less so three months later. The advice was not wrong: it was dated, and nobody re-dated it.

That is the deep trap of this subject. Every month brings a list of "most-cited domains" that agencies turn into an action plan, on an engine whose source mix has already moved by the time the plan is presented. A strategy depending on a third-party site depends on a decision you do not make.

What does not move: a page that fully answers a precise question, says where its figures come from, and can be read by the crawler. Less spectacular than a trick, and it survives the updates.

Seven moves that improve your odds

In this order: the first three condition all the others.

01

Open the door to the right crawler

Check that OAI-SearchBot is not disallowed in robots.txt, and decide consciously what happens to GPTBot — blocking it is a legitimate choice, as long as it is a choice.

02

Serve the page without JavaScript

Content that only appears after a script runs may not be seen. The test takes one command: request the page and look for the text in the response.

03

Write one page per question, not one page per keyword

Assistants pick up the passage that answers on its own. A page covering six subjects offers no complete passage to lift; six pages covering one each offer six.

04

Put the answer at the top

A complete answer in two or three sentences, before the detail. That is what a model lifts, and incidentally what a hurried reader reads.

05

Date and attribute every figure

"68% of searches" is worth nothing; "68.01% of US searches between January and April 2026, according to SparkToro" can be quoted as it stands. That is exactly the lever the research measured.

06

Make your facts consistent everywhere else

A model cross-reads your site, your company listing, industry directories and the press. A price or a location that differs from one source to the next produces a hesitant answer, or a wrong one.

07

Measure, then start again

Put your questions to the engines, several times, and look at who gets cited. Without that step the other six stay opinions.

What does not work, and what nobody knows

Four widespread ideas. The first is measured; the other three are grey areas better announced as such.

Keyword stuffing
Measured, and ineffective: it is the one lever tested by the GEO study that brings nothing. Hardly surprising — a language model does not count occurrences, it judges whether the passage answers.
The llms.txt file
A proposed standard that appeared in 2024, offering models a clean version of the site. None of the major vendors has publicly confirmed reading it. Publishing one costs an hour and cannot hurt; presenting it as a solution would be dishonest.
Mass-produced AI content
No public study supports the claim that it increases citations. What we do know: the levers that work are the ones carrying verifiable information — which content produced without sources does not carry.
"Most-cited domains" rankings
Useful for understanding the terrain, dangerous as an action plan: they describe a past state of an unstable mix, and they say nothing about your market.

Frequently asked questions

Should I block GPTBot?
That is a company decision, not a visibility decision: GPTBot serves model training, not citations. Blocking it does not damage your presence in ChatGPT Search. What you must not do is block OAI-SearchBot believing you are blocking training.
How long before being cited?
OpenAI documents no delay. In practice the ChatGPT Search index refreshes continuously, but what a model knows from memory only changes at its next training. Hence the value of separating the two modes when measuring.
Does Google ranking help with ChatGPT citations?
Partly, and the exact share is disputed: depending on the study, the share of citations coming from the top ten organic results ranges from 16.7% to 52%. Ranking well helps, is not enough, and is not required.
Can I pay to appear?
There is no purchased placement inside ChatGPT citations to date. Any offer promising a position in the answers is selling something undocumented.
How do I know whether ChatGPT already cites me?
By asking your questions several times and recording the answers, by hand or with a tool. A single query proves nothing: an assistant's answer is not reproducible.
Is a mention without a link worth anything?
It brings no visit, but it builds awareness with the reader. The two cases are fixed differently, which is why they should be counted separately rather than inside a single rate.
  • The business law firms AI assistants cite most

    When a company asks ChatGPT or Google which law firm to call, which names come back? We put 26 questions to five assistants and counted the citations of seven…

  • Why my brand doesn't appear in ChatGPT

    You asked the question, and a competitor came out instead of you. Before rewriting anything, work out which of the four causes applies: they share neither the…

  • How much does a GEO audit cost

    Two of the three most-cited tools no longer show any price at all. Here are the ones still public, what the API calls behind them really cost, and why the gap…

  • AI prompt volume does not exist

    Every AI visibility tool ends up printing a number next to each question: "2,400 searches a month". Nobody measures that. Here is where the figure actually…

  • Getting into Google AI Overviews

    This is the generative surface most people see without ever looking for it: it sits above the results, in front of users who have never opened an assistant…

  • When AI gets you wrong

    A brand that is highly visible in assistant answers, with a wrong price, does not have a visibility problem. It is the one case where a good score is worse…

  • Building your question corpus

    It is the part of a GEO audit nobody wants to do, and it decides everything else. A badly written corpus produces impeccable figures about questions nobody…

  • GEO audit: what it measures, how, and what makes an audit wrong

    Public figures on generative search contradict each other: AI Overviews appear on "more than 20%" of searches by one measurement, on 75.24% by another. Both…

  • GEO or SEO: what really changes

    Three serious studies measured the same thing — the share of AI citations coming from pages that rank well on Google. They found 16.7%, 52% and a correlation…

  • AI visibility tracking tools in 2026

    Seven tools, read on the same day from their public pages. Two publish no price at all. And once you divide published prices by the number of answers actually…

  • robots.txt and llms.txt: letting AI read your site

    It is the one generative-visibility job that can be settled in an afternoon, and the one that makes every other job pointless if you get it wrong. Three files…

  • Etalo.ai review

    You are looking for reviews of Etalo and you cannot find any. That is normal: the tool is young and has no clients to parade. Rather than manufacture…

  • Best Peec.ai alternative

    Peec.ai does one thing well: it tracks your visibility in assistant answers, daily, in a dashboard that is easy to read. If that is what you need, it is a good…

  • Best Promptwatch alternative

    Promptwatch has one real strength few tools have: crawler logs, that is, what AI robots actually come and read on your site. If that is your subject, keep it…

Sources

OpenAI crawlers and what each one does: OpenAI documentation, read on 16 September 2026.

Measured levers: Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, Ameet Deshpande, GEO: Generative Engine Optimization, KDD 2024 — GEO-bench, 10,000 queries.

Cited domains and their instability: Semrush, 230,000 prompts and over 100 million citations recorded between 14 July and 12 October 2025.

Know where you stand first

A demonstration on your brand: what ChatGPT and the five other engines answer about your market, and what they can read on your site.