ChatGPT Search: How It Cites Sources & How to Get Cited

ChatGPT Search is the largest AI search surface by reach. With 900 million weekly active users, it drives a meaningful share of the AI search referral traffic that grew 527% year-over-year in 2025. Getting cited by ChatGPT Search is now a top-tier GEO goal.

ChatGPT Search works like other RAG-based engines: it retrieves passages from its index, re-ranks them, and synthesizes an answer with inline citations. The crawler that builds that index is OAI-SearchBot, and it is distinct from GPTBot (which is used for model training). If OAI-SearchBot cannot crawl your site, you cannot be cited.

How ChatGPT Search cites sources

When ChatGPT Search generates an answer, it places inline numbered citation markers next to claims that came from a specific source. Below the answer, it displays source cards — typically 3–8 per answer — with the page title, domain, favicon, and a clickable link back to the original page.

The model decides citations in the generation stage of the RAG pipeline. Five factors drive the decision: factual density, source authority, information uniqueness, content structure, and semantic consistency. ChatGPT tends to cite sources that contribute specific, verifiable facts (numbers, named entities, dates) rather than generic commentary.

"AI search referral traffic grew 527% year-over-year in 2025, with ChatGPT Search as the largest single contributor. Publishers who optimize for AI citations now are building a durable advantage as traditional search traffic declines."

OAI-SearchBot: the crawler you must allow

OpenAI operates two main crawlers: GPTBot (used for model training) and OAI-SearchBot (used to build the ChatGPT Search index). They are separate, and allowing one does not allow the other. Many sites that explicitly block GPTBot for training concerns also accidentally block OAI-SearchBot, eliminating themselves from ChatGPT Search citations.

The minimum robots.txt configuration to be eligible for ChatGPT Search citations is:

User-agent: OAI-SearchBot
Allow: /

# Optional: keep GPTBot blocked if you do not want training
# User-agent: GPTBot
# Disallow: /

Verify your robots.txt actually serves this — Cloudflare, WordPress security plugins, and custom middleware sometimes inject blocks that override your robots.txt. Use OpenAI's documentation or a crawler simulator to confirm OAI-SearchBot can fetch your pages.

5 ways to get cited by ChatGPT Search

Based on the Princeton GEO study (KDD 2024) findings and observed ChatGPT Search behavior, these are the five highest-leverage optimizations:

1. 1.
   Allow OAI-SearchBot and verify crawl access

   Without this, nothing else matters. Add the Allow directive to robots.txt and verify with a log review that OAI-SearchBot is fetching your pages.

2. 2.
   Add specific statistics with named sources

   Statistics addition boosts AI visibility by +33% in the Princeton study. Replace "many users" with "900M weekly active users (OpenAI, 2025)". Cite the source inline so the AI can verify the claim.

3. 3.
   Include expert quotations with full attribution

   Expert quotations give the largest single visibility lift at +41%. Use blockquotes for direct quotes, name the speaker, and identify the occasion (paper, conference, interview).

4. 4.
   Structure content for passage extraction

   One idea per paragraph. Clear H2/H3 headings. FAQ blocks. Numbered steps. Tables. ChatGPT pulls passages, not pages — discrete, well-structured units are easier to extract and cite.

5. 5.
   Implement Schema.org structured data

   Article, FAQ, HowTo, and Organization schema help ChatGPT parse your content. Use JSON-LD (not microdata) and validate with Schema.org's validator before deploying.

What ChatGPT Search rewards and penalizes

Source: Aggarwal et al., "GEO: Generative Engine Optimization," arXiv:2311.09735, KDD 2024. Visibility measured by position-adjusted word count on GEO-bench.

Common ChatGPT Search citation mistakes

• ▸ Blocking OAI-SearchBot while allowing GPTBot (or vice versa) — they are separate crawlers and must be configured separately.
• ▸ Vague claims with no sources — ChatGPT rarely cites "many experts believe" statements. Replace with named experts and specific numbers.
• ▸ Walls of text with no structure — long paragraphs without headings get fragmented badly during retrieval. Break them up.
• ▸ Keyword repetition — stuffing "GEO, GEO, GEO" reduces visibility by 8% and reads as low-quality to the re-ranker.
• ▸ No Schema.org — without structured data, ChatGPT has to infer content type, which reduces citation probability.

Frequently asked questions