--- title: "GPTBot vs OAI-SearchBot vs ChatGPT-User vs ChatGPT-Referral" description: "Learn what GPTBot, OAI-SearchBot, ChatGPT-User, and ChatGPT-Referral actually do, how they work, and what each one means for your GEO strategy." url: https://justinha.info.vn/geo/openai-bots/ --- # GPTBot vs OAI-SearchBot vs ChatGPT-User vs ChatGPT-Referral # GPTBot vs OAI-SearchBot vs ChatGPT-User vs ChatGPT-Referral: Full Guide for *GEO* Written by [Justin Hà](https://justinha.info.vn/about/) · March 28, 2026 · On This Page ![GPTBot vs OAI-SearchBot vs ChatGPT-User vs ChatGPT-Referral: Full Guide for GEO](https://justinha.info.vn/wp-content/uploads/2026/03/gptbot-vs-oai-searchbot-vs-chatgpt-user-chatgpt-referral.jpg) Take this article with you Copy as Markdown Copy link Ask an AI to summarise it Key Takeaways - OpenAI runs 4 separate bots, GPTBot (trains AI models), OAI-SearchBot (powers ChatGPT's live search), ChatGPT-User (fetches a page only when a real user shares a link), and ChatGPT-Referral (a traffic source, not a bot, recorded when someone clicks a ChatGPT citation). - Each bot is controlled independently in robots.txt, blocking GPTBot does not affect OAI-SearchBot or ChatGPT search visibility, and vice versa. - None of the 4 execute JavaScript, content loaded dynamically after the initial HTML is invisible to all of them. - ChatGPT-User visits are the strongest positive signal in a log file, they only happen when a real human brings a URL into a ChatGPT conversation. - ChatGPT-Referral visitors convert around 15.9%, and AI referral traffic overall is growing roughly 165x faster than organic search (WebFX, mid-2025). - After GPT-5 shipped in August 2025, OAI-SearchBot traffic rose 136%, ChatGPT-User 65%, and GPTBot 66%, confirming model releases trigger crawl spikes across all 3 bots at once (Akamai, late 2025). - A full crawl-to-citation-to-click journey requires all 4 steps to work: GPTBot trains, OAI-SearchBot indexes, ChatGPT-User fetches on demand, ChatGPT-Referral delivers the click. OpenAI runs four separate bots that interact with a website in completely different ways. Treating them as one single “ChatGPT bot” leads to bad decisions about robots.txt, content strategy, and GEO optimization. Each bot has a specific job, a specific trigger, and a specific impact on visibility in AI-generated answers. This guide explains exactly what each OpenAI bot does, how to spot it in an [SEO log file](https://justinha.info.vn/geo/what-is-seo-log-file/), and what its presence means for GEO strategy in 2026. ## Why OpenAI Uses Four Different Bots OpenAI built ChatGPT to do several different things at once: answer questions from training data, search the web in real time, fetch specific pages a user asks about, and send users to external websites through inline citations. Each task requires a different kind of web interaction, which is why OpenAI operates four separate bots instead of one. Treating all four as identical causes real problems. A site could block GPTBot to protect training data while accidentally cutting itself out of ChatGPT search results, or allow OAI-SearchBot while blocking ChatGPT-User and missing the highest-value signal in the log file. ## The Four OpenAI Bots at a Glance Bot Purpose Trigger Impact on GEO GPTBot AI model training Automatic, scheduled Shapes what ChatGPT knows about a brand long-term OAI-SearchBot Real-time search indexing Automatic, periodic Determines if a site appears in ChatGPT Search results ChatGPT-User On-demand page fetch Real user action Strongest signal that real users are finding content via ChatGPT ChatGPT-Referral User click on a cited link Real user click Means ChatGPT cited a page and a user clicked through to read it ## GPTBot: The AI Training Crawler GPTBot is OpenAI’s primary web crawler. Its job is to collect content from across the internet and use it to train OpenAI’s large language models, including GPT-4 and future versions of ChatGPT. When GPTBot visits a site, it reads the text, stores it, and feeds it into the training process. This is an asynchronous, offline process. GPTBot is not fetching pages to answer a user’s question right now, it is building the knowledge base ChatGPT will draw from weeks or months later. **User agent string:** ``` Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; GPTBot/1.3; +https://openai.com/gptbot) ``` GPTBot crawls on a scheduled basis with long revisit intervals. It does not crawl as frequently as Googlebot because it does not need fresh content for search results, its goal is broad coverage of quality content, not speed. It follows links page to page, reads publicly accessible text, and respects robots.txt. > **GPTBot does not execute JavaScript.** It downloads .js files but does not run them. If the most important content loads by JavaScript after the initial HTML, GPTBot will likely miss it. Static HTML is what it reads best. Allowing GPTBot to crawl a site gives control over what ChatGPT learns about that brand. Blocking it means ChatGPT’s knowledge comes entirely from what other websites say, which could include outdated information or nothing at all. Blocking GPTBot does not block the other three OpenAI bots, each is controlled independently in robots.txt. On the justinha.info.vn log file covering February 28 to March 14, 2026, GPTBot made 93 requests. It spent most of its time on the sitemap and homepage before moving to specific content pages, showing a clear preference for GEO-related content, visiting pages like `/geo/what-is-generative-engine-optimization/` and `/geo/what-is-llms-txt-llms-full-txt-cats-txt/` multiple times. It also accessed the WordPress REST API directly via `/wp-json/wp/v2/posts/`. To allow GPTBot to train on content: ``` User-agent: GPTBot Allow: / ``` To block GPTBot from training while keeping other bots active: ``` User-agent: GPTBot Disallow: / ``` ## OAI-SearchBot: The ChatGPT Search Indexer OAI-SearchBot is OpenAI’s search crawler. Its job is to index the web for ChatGPT’s real-time search feature, sometimes called SearchGPT. When a ChatGPT user asks a question that requires current information, ChatGPT uses OAI-SearchBot’s index to find and surface relevant pages. It does not collect data for AI model training, its only job is powering live search results. **User agent string:** ``` Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36; compatible; OAI-SearchBot/1.3; +https://openai.com/searchbot ``` OAI-SearchBot crawls periodically, similar to Bingbot, updating its index regularly. Its crawl pattern is more targeted than GPTBot because it looks for pages that will answer specific user queries. Like GPTBot, it does not execute JavaScript. It is believed to work alongside Bing’s index through OpenAI’s data partnership with Microsoft, though it also independently discovers and indexes content. > **This is the bot that most directly affects ChatGPT search visibility.** OpenAI officially states that allowing OAI-SearchBot in robots.txt is required for pages to appear in ChatGPT search results. Blocking it is generally a bad idea for visibility. In the sample log, OAI-SearchBot made 66 requests. It checked robots.txt 45 times, showing it was actively checking crawl permissions. It focused on case study pages and GEO content, including `/geo-ai-branding/`, and also accessed WordPress REST API endpoints and oEmbed data. ``` User-agent: OAI-SearchBot Allow: / ``` ## ChatGPT-User: The On-Demand Page Fetcher ChatGPT-User is triggered by a real person. When a ChatGPT user shares a specific URL and asks ChatGPT to read, summarize, or answer questions about it, ChatGPT sends the ChatGPT-User bot to fetch that page in real time. It does not crawl the web automatically, it only acts when a human directs it to a specific page. This makes it fundamentally different from GPTBot and OAI-SearchBot, which run on schedules and crawl broad swaths of the internet. ChatGPT-User is reactive, it exists only because a real person found a URL and brought it into a conversation. **User agent string:** ``` Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; ChatGPT-User/1.0; +https://openai.com/bot ``` The content fetched by ChatGPT-User is used only for that specific conversation. OpenAI has stated it does not contribute to large-scale training datasets. Like the other OpenAI bots, it does not run JavaScript and it respects robots.txt. Blocking it means ChatGPT cannot fetch pages even when a real user specifically asks it to. > **ChatGPT-User visits are the strongest positive signal in a log file.** They confirm a real person found a URL valuable enough to bring into a ChatGPT conversation, not an automated crawl. In the sample log, ChatGPT-User made 28 requests across two weeks, a meaningful signal of real usage. ``` User-agent: ChatGPT-User Allow: / ``` Blocking ChatGPT-User is generally a bad idea. It prevents real users from getting help with content inside ChatGPT, and it does not protect training data or prevent search indexing, those are controlled by GPTBot and OAI-SearchBot respectively. ## ChatGPT-Referral: The Traffic Signal ChatGPT-Referral is not a bot at all. It is a referral traffic source that appears in Google Analytics or other analytics platforms. When ChatGPT includes a link to a page in a generated answer and a user clicks that link, the visit shows up as a referral from chatgpt.com or with a `utm_source=chatgpt.com` parameter. This is the clearest possible signal that ChatGPT is actively recommending content to its users. The bot crawled the site, ChatGPT decided the page was a good answer, it included a link, and a real person clicked through. Every step in that chain had to work correctly. ChatGPT-Referral traffic is widely reported as higher quality than average. Research shows ChatGPT referral visitors convert at around 15.9%, higher than most organic search traffic, since they arrive already having read a ChatGPT summary with clear intent. Metric Value ChatGPT-Referral conversion rate ~15.9% AI referral traffic share of total web traffic ~1% on average AI referral traffic growth vs. organic search ~165x faster Source: WebFX, mid-2025. The volume of ChatGPT-Referral traffic is still relatively small compared to Google organic traffic for most sites, but it converts better, engages longer, and is growing far faster than organic search. ### How to Track ChatGPT-Referral in GA4 1. **Open Traffic Acquisition.** In GA4, go to Reports, then Acquisition, then Traffic Acquisition. 2. **Filter by source.** Use the search box to filter by “chatgpt” or “openai.” 3. **Build an AI Traffic segment.** Create a custom segment where the session source matches a list of AI platforms, including chatgpt.com, perplexity.ai, and claude.ai, for a single view of all AI referral traffic. Some ChatGPT traffic shows up as direct, since users copy and paste URLs rather than clicking inline citations. Spikes in direct traffic alongside ChatGPT growth may be related. ## How the Four Bots Work Together Understanding each bot separately matters, but the bigger picture is how they work together as a system, the full journey from crawl to citation to click. 1. **Step 1, GPTBot.** Crawls content over time and adds it to ChatGPT’s training data, shaping ChatGPT’s base knowledge of a brand. 2. **Step 2, OAI-SearchBot.** Indexes pages for real-time search. When a user asks a question requiring current information, ChatGPT searches its index and may include a link to the page. 3. **Step 3, ChatGPT-User.** A user finds the URL somewhere, brings it into a ChatGPT conversation, and asks ChatGPT to read it, fetching the page for that conversation. 4. **Step 4, ChatGPT-Referral.** A user sees the page cited in a ChatGPT answer, clicks the link, and lands on the site, showing up as a referral from chatgpt.com. Each step depends on the previous one being allowed and working correctly. Block GPTBot and ChatGPT’s knowledge weakens. Block OAI-SearchBot and pages never appear in search results. Block ChatGPT-User and real people cannot get ChatGPT’s help understanding the content. ## The Relationship Between GPTBot and OAI-SearchBot Research from Akamai published in late 2025 found evidence that GPTBot and OAI-SearchBot activity is correlated, with GPTBot tending to follow OAI-SearchBot with a slight time offset. One explanation is that GPTBot uses OAI-SearchBot’s search data to find the highest-quality content to train on. > **Model releases trigger crawl spikes.** After GPT-5 shipped in August 2025, OAI-SearchBot traffic increased 136%, ChatGPT-User 65%, and GPTBot 66%. A sudden spike in any of these bots is worth checking against OpenAI’s announcement calendar. ## JavaScript Rendering: A Critical Technical Issue for All Four Bots None of the four OpenAI bots execute JavaScript. They download JavaScript files but do not run them, one of the most important technical facts for GEO. If a website renders content dynamically through JavaScript, that content is invisible to all four bots. This affects sites built with React, Vue, Angular, or other JavaScript frameworks where content loads after the initial HTML response, and WordPress sites using JavaScript-heavy page builders if critical content is injected rather than included in the initial HTML. The fix is server-side rendering or prerendering. For WordPress sites using standard themes, this is less of an issue since most content delivers as static HTML. Sites using Elementor, Divi, or similar builders should check whether key content appears in View Page Source. ## Recommended robots.txt Setup for GEO Visibility For maximum visibility across all ChatGPT features, allow all three crawlable bots: ``` User-agent: GPTBot Allow: / User-agent: OAI-SearchBot Allow: / User-agent: ChatGPT-User Allow: / ``` To allow ChatGPT search visibility without contributing to AI model training: ``` User-agent: GPTBot Disallow: / User-agent: OAI-SearchBot Allow: / User-agent: ChatGPT-User Allow: / ``` Changes to robots.txt take approximately 24 hours to affect ChatGPT search behavior. GPTBot changes may take longer to reflect in ChatGPT’s knowledge base, since training happens on a slower cycle. ## How to Find All Four Bots in Your SEO Log File A server log file is the only place to see GPTBot, OAI-SearchBot, and ChatGPT-User together in one view. - **GPTBot.** Look for `GPTBot/1.3` in the user agent string. Check which pages it visits and whether it wastes time on low-value URLs like parameter pages or pagination. - **OAI-SearchBot.** Look for `OAI-SearchBot/1.3`. Check the status codes it receives, any 404s are pages that could be ranking in ChatGPT search but aren’t. - **ChatGPT-User.** Look for `ChatGPT-User/1.0`. Every request represents a real human interaction, note which pages they visit most, these carry the highest current GEO value. - **ChatGPT-Referral.** This does not appear in server logs. Check the analytics platform for traffic from chatgpt.com or `utm_source=chatgpt.com`. > **Skip the manual grep.** The [AI Log File Analyzer](https://justinha.info.vn/ai-log-file-analyzer-tool/) requires no software installation and handles files up to 1GB, roughly 5 million rows. Upload a log file and instantly see which AI bots are crawling, which pages they visit, what status codes they receive, and [how crawl budget is being used across AI bots](https://justinha.info.vn/geo/ai-bot-crawl-budget-log-file/), including GPTBot, OAI-SearchBot, ClaudeBot, and Applebot. ## What GEO Practitioners Should Do With This Data - If GPTBot is crawling low-value pages, improve internal linking toward the most important content and add those pages to the sitemap - If OAI-SearchBot is hitting 404s, fix or redirect those pages, each one is a missed opportunity to appear in ChatGPT search - If ChatGPT-User visits are low, focus on content that answers specific, searchable questions people are likely to share - If ChatGPT-Referral traffic isn’t growing, focus on answer-first content structure, entity optimization, and presence on the platforms LLMs learn from, including Reddit, LinkedIn, and YouTube ## From Log File Analysis to Full GEO Strategy Knowing GPTBot visited a site 93 times is useful data. Knowing it spent most of those visits on GEO-related content, avoided the service pages, and never found the case studies, like [Holiday Inn Saigon Airport](https://justinha.info.vn/case-study/holiday-inn-saigon-airport/) or [Tsar Carpets](https://justinha.info.vn/case-study/tsar-carpets/), is actionable intelligence. That gap between what the bot found and what a brand wants it to read is exactly where GEO strategy begins, the same process covered start to finish in our [full SEO/GEO process](https://justinha.info.vn/geo/seo-geo-process/). ## Frequently Asked Questions ### Can I block GPTBot without affecting ChatGPT search results? Yes. GPTBot and OAI-SearchBot are controlled independently in robots.txt. Blocking GPTBot prevents content from being used in AI model training, it does not affect whether pages appear in ChatGPT search results, that is controlled by OAI-SearchBot. ### Does ChatGPT-User contribute to AI training? No. OpenAI has stated that content fetched by ChatGPT-User is used only for the specific conversation that triggered it. Blocking ChatGPT-User does not protect training data, it only prevents real users from getting ChatGPT’s help with the content. ### Why does OAI-SearchBot check robots.txt so many times? In the sample log, OAI-SearchBot checked robots.txt 45 times. This is normal, crawlers re-read robots.txt frequently to check whether permissions have changed, confirming access before each session. ### How long does it take to appear in ChatGPT search after allowing OAI-SearchBot? OpenAI states it takes approximately 24 hours after a robots.txt update. Full indexing may take longer depending on site size and how frequently OAI-SearchBot revisits. ### Is ChatGPT-Referral traffic worth optimizing for? Yes, even though the volume is currently small. ChatGPT-Referral visitors convert at around 15.9% with higher engagement than average organic visitors, and the channel is growing roughly 165x faster than organic search. Getting into citations now, while competition is lower, builds a position that becomes harder to displace over time. ### How do I know if ChatGPT is citing my pages? Check GA4 for traffic from chatgpt.com or a `utm_source=chatgpt.com` parameter, and look for the ChatGPT-User bot in server logs, confirming real users are bringing URLs into ChatGPT conversations. If pages are being crawled but never cited, see the full breakdown in [why AI bots crawl a site but never cite it](https://justinha.info.vn/geo/ai-crawled-not-cited-fix/). ## Turn This Data Into a Full GEO Strategy Knowing which of the four bots is missing tells exactly where ChatGPT visibility stands and what to fix next. Our [GEO AI SEO Branding service](https://justinha.info.vn/geo-ai-branding/) turns log file insights into a structured AI visibility strategy, covering AI Visibility Audits across ChatGPT, Perplexity, and Google AI Overviews, llms.txt and schema setup, monthly prompt testing, and off-site authority building, available for [real estate](https://justinha.info.vn/geo-ai-branding/real-estate/) and [property management](https://justinha.info.vn/geo-ai-branding/property-management/) brands as well as hospitality. Ready to start? Upload a log file to the [AI Log File Analyzer](https://justinha.info.vn/ai-log-file-analyzer-tool/), completely free, no sign-up required, no file size limit up to 1GB. [Book a free consultation](https://justinha.info.vn/contact/) to turn the results into a strategy. --- ![Justin Hà](https://justinha.info.vn/wp-content/uploads/2025/11/justin-ha-seo-specialist-hcm.jpg) About the Author **Justin Hà** Senior Global SEO & GEO Specialist · Founder of Luminal Ho Chi Minh City, Vietnam Justin Hà is a Senior Global SEO & GEO Specialist with 5+ years of experience driving organic growth across 9 international markets. He is among Vietnam's first practitioners of Generative Engine Optimization (GEO) - optimizing brand visibility inside AI-generated answers from ChatGPT, Gemini, Perplexity, and other platforms. In 2026, he founded [Luminal](https://www.linkedin.com/company/luminal-geo/), a GEO agency helping brands get discovered, cited, and trusted inside AI-driven search. With 2+ years in GEO and 100+ projects led or consulted, Justin helps brands adapt their content strategies to the rapidly evolving AI search landscape. [Vietnam's First GEO Practitioner](https://justinha.info.vn/about/) 5+ Years in SEO 2+ Years in GEO 100+ Projects 10+ Active GEO [ ](https://www.linkedin.com/in/justinhaseo/) [ ](https://www.linkedin.com/company/luminal-geo/) [ ](https://www.facebook.com/justinhaseo) [ ](https://x.com/justinhaseo) [ ](https://www.instagram.com/justinhaseo/) ## Related Articles [ ![GPT-5.6 Listicle Citations: What Changed, What Didn’t](https://justinha.info.vn/wp-content/uploads/2026/09/listicle-format-citation-chatgpt-768x432.jpg) GEO GPT-5.6 Listicle Citations: What Changed, What Didn’t September 19, 2026 ](https://justinha.info.vn/geo/gpt-5-6-killed-listicle-citations/) [ ![ChatGPT Pulls In Google Maps Data for 75% of Answers, Here’s What That Means](https://justinha.info.vn/wp-content/uploads/2026/09/local-map-result-in-chatgpt-768x427.jpg) GEO ChatGPT Pulls In Google Maps Data for 75% of Answers, Here’s What That Means September 18, 2026 ](https://justinha.info.vn/geo/chatgpt-pulls-in-google-maps-data-of-answers/) [ ![ChatGPT’s Two Models Search Very Differently, But Cite the Same Winner](https://justinha.info.vn/wp-content/uploads/2026/09/site-search-in-chatgpt-768x410.png) GEO ChatGPT’s Two Models Search Very Differently, But Cite the Same Winner September 18, 2026 ](https://justinha.info.vn/geo/gpt-5-6-vs-gpt-5-6-mini/)