When someone asks ChatGPT a question that needs current information, the assistant quietly runs a web search, reads the pages it finds, and folds the result into its reply with a handful of linked citations. That retrieval step, bolted onto the most-used chatbot in the world, is why ChatGPT matters more to publishers than any other AI surface: it is where most people now go first, and it reads publisher content at a scale no rival approaches. The complication is that the reading and the answering both happen on OpenAI's surface, so the work your content does to resolve the question rarely converts into a visit to your own site.
What is ChatGPT Search?
ChatGPT Search is the feature that lets ChatGPT pull live information from the open web and cite it, rather than answering only from the static knowledge baked into its model during training. OpenAI introduced it as a standalone capability at the end of October 2024, then folded it into the default ChatGPT experience so that the assistant now decides for itself when a question needs a fresh web lookup. When it does, it retrieves relevant pages, synthesises them into a direct answer, and attaches a small set of source links the user can expand.
The reason this is a publisher story rather than a product footnote is reach. OpenAI reported ChatGPT at around 900 million weekly active users in February 2026, and the assistant crossed a billion monthly users by June 2026. No other AI surface combines that audience with live retrieval of third-party content. So while citation-first rivals such as Perplexity lean harder on sources as a design principle, ChatGPT is the surface that actually reads the most publisher content, simply because so many more questions pass through it.
How does ChatGPT Search work?
ChatGPT Search runs a retrieval-augmented pipeline. When the model judges that a query needs current or factual grounding, it issues a web search through third-party search providers, ranks the returned pages, reads the most relevant passages, and a language model writes the answer with citations drawn from those passages. Content from OpenAI's licensed publisher partners can also be surfaced directly. The citation is therefore generated from material the system has just read, not from a fixed link directory, which means your page has to be retrievable in that moment to appear at all.
It helps to know that OpenAI does not crawl with a single bot. OAI-SearchBot builds the search index that powers ChatGPT Search, ChatGPT-User fetches a specific page live when a user's question triggers it, and GPTBot gathers data for model training. These are separable in robots.txt, which matters because a publisher can allow the search and live-answer bots while restricting the training crawler, or the reverse. Blocking OAI-SearchBot removes you from the index ChatGPT Search draws on, so the page can no longer be cited.
How does ChatGPT decide which sources to cite?
Independent analysis of ChatGPT's retrieval behaviour finds a consistent bias toward what looks authoritative: established publications, well-structured pages that answer the specific question directly, and recent content. A peer-reviewed study of around 11,000 real queries found the system systematically favours earned media in trusted outlets over brand-owned pages, and separate network-traffic analysis shows Wikipedia is heavily overrepresented, appearing in roughly one in six cited conversations as a baseline knowledge layer. There is also evidence of a licensed tier, with researchers identifying an internal source tag that appears reserved for publishers holding content deals with OpenAI. One practical quirk: a conversation's opening question is far more likely to trigger a fresh web search and citations than later follow-ups, so the first turn is where most publisher visibility is won.
The takeaway for a publisher is the same discipline that wins featured snippets in classic search, applied to a system that quotes you rather than just links you. Clear question-led headings, a self-contained answer near the top of the page, factual precision, freshness, and clean server-rendered HTML all raise the odds of being retrieved cleanly. Content buried behind heavy client-side rendering or vague preamble is far less likely to make the cited set.
How much traffic does ChatGPT actually send publishers?
ChatGPT dominates the referral traffic that AI assistants do send. Across 2026, measurements have repeatedly put ChatGPT at the great majority of all AI-to-web referrals, with some datasets attributing around 82 percent of measurable AI referral traffic to it; an October 2025 report had it as high as 89 percent before rivals such as Gemini and Claude began taking measurable share. Reading these figures, the methodology matters: a separate analysis of business-to-business referrals across March and April 2026 put ChatGPT's share lower, near 63 percent, as the field diversified. The direction of travel is a slowly shrinking but still commanding lead.
Dominating AI referrals, though, is not the same as sending meaningful traffic. The volume of clicks that flows back is small relative to how much content the assistant consumes, a gap captured by the crawl-to-refer ratio, where OpenAI has been measured reading on the order of several hundred pages for every visitor it returns. The Columbia Journalism Review's Tow Center has separately documented how poorly AI search engines, ChatGPT included, attribute and link the news they summarise. The structural result is the one publishers keep running into: your content answers the question inside ChatGPT, your brand may even appear in a citation, and you still receive no page view, no ad impression, and no first-party data from the exchange.
Does ChatGPT pay publishers?
Some publishers, yes; most, no. OpenAI has signed content licensing deals with a roster of large media organisations, reportedly including the Associated Press, Axel Springer, Condé Nast, Dotdash Meredith, the Financial Times, Hearst, News Corp, Le Monde, Reuters, The Atlantic, Time and Vox Media, granting access to their content for training and real-time retrieval in exchange for compensation and, in practice, more reliable citation. There is no public application process; OpenAI approaches publishers directly, and the deals have gone to organisations with large, recognised content libraries. For everyone outside that group, ChatGPT reads the open web for free.
The advertising dimension sharpens the point. OpenAI began testing ads inside ChatGPT in February 2026, showing clearly labelled ad units beneath answers for free-tier users in selected markets, and is targeting billions in ad revenue over the next few years. Crucially for this audience, OpenAI's vice president of media partnerships has confirmed the company has no plans to share that advertising revenue with publishers. The licensing deals cover content access, not a cut of downstream ad income. That is a deliberate contrast with Perplexity, which built an explicit citation-based revenue-share pool for publishers. ChatGPT's model is licensing for the few and unpaid reading for the many, with the emerging ad revenue staying entirely with OpenAI.
What should publishers do about ChatGPT Search?
There are two separate jobs. The first is comprehension and visibility: keep OAI-SearchBot allowed if you want to be eligible for citation, and structure content for retrieval with question-led headings, direct answer paragraphs, precise and current facts, and clean HTML. The second is compensation, and it is the harder one, because being cited by ChatGPT is not being paid by ChatGPT, and referral clicks will stay thin by design while ad revenue from those answers accrues to OpenAI alone.
This is the gap blankspace works in. Licensing deals reward a handful of national publishers and the ad model rewards the platform, but neither values the everyday read of the open web that powers most ChatGPT answers. blankspace's focus is that underlying read itself, detecting AI and Live Search Agent traffic at the CDN edge where the request actually lands and monetising it there, rather than waiting for a click that mostly will not arrive. The strategic shift for any media executive is to stop modelling ChatGPT as a search engine that owes you traffic and start treating it as a distribution surface that reads your content whether or not it sends a visit, then to make sure that read is measured and, where possible, monetised.
Frequently asked questions
Is ChatGPT Search a search engine?
Not in the traditional sense. A search engine returns a ranked page of links and leaves you to choose; ChatGPT Search reads the web on your behalf and returns a written answer with a short list of supporting citations. It is better described as an answer engine layered on top of a chatbot. It uses live web retrieval, so its replies are grounded in current sources it has just read, but the destination is its own answer, not a results page that routes you onward.
How is ChatGPT Search different from Google AI Mode for publishers?
Both read live web content and answer in place, but they sit on different front doors. Google AI Mode is a conversational surface inside Google Search, drawing on Google's index and reaching users who arrive through Google. ChatGPT Search lives inside ChatGPT, draws on third-party search providers plus OpenAI's licensed partners, and reaches the assistant's own very large user base. For a publisher the practical difference is which crawler and which optimisation each rewards, and the fact that the two surfaces frequently cite different sources for the same question, so visibility in one does not guarantee visibility in the other.
Does appearing in ChatGPT send traffic to my site?
Usually not in volume. ChatGPT resolves most questions inside its own answer, so the citation substitutes for the click far more often than it drives one, and the ratio of pages read to visitors returned is heavily skewed toward reading. Being cited delivers brand visibility and some authority signal, but it should not be modelled as a reliable source of referral visits or ad impressions.
Should I block OAI-SearchBot from crawling my site?
That is a commercial decision rather than a technical default. Blocking OAI-SearchBot removes you from the index ChatGPT Search uses, which protects against uncompensated reads but also removes any chance of being cited or surfaced to that billion-user audience. Because OpenAI separates its search, live-fetch and training crawlers, many publishers allow the search and answer bots while restricting the training crawler, then pursue compensation for the read through licensing or edge-level monetisation. The right call depends on how much you value ChatGPT visibility against the cost of the reads behind it.
How do I get my content cited by ChatGPT?
Write for retrieval. Lead each page with a clear, self-contained answer to a specific question, use question-format headings, keep facts precise and current, and serve clean, crawlable HTML rather than content locked behind heavy client-side rendering. Authority and freshness both help, so well-sourced pages from recognised publications are favoured, and the opening question in a conversation is where citations are most likely to be triggered. These are the same fundamentals that win citations across most AI answer engines, applied to the largest one.