Separate eligibility from selection

OpenAI's publisher guidance says public websites can appear in ChatGPT search. To help content be discovered, surfaced, cited, and linked, the company tells publishers not to block OAI-SearchBot. Its ChatGPT search help page adds two important limits: the host or content delivery network must allow traffic from the published IP ranges, and ranking uses multiple factors intended to find reliable, relevant information. It also says there is no way to guarantee top placement. These statements support a practical goal—make good pages eligible and useful—not a guaranteed-placement service.

What you can controlWhat you cannot promise
Crawler accessRobots rules, server responses, firewall rules, and CDN access for OAI-SearchBotWhen the crawler returns or how quickly systems process a change
Published factsClear names, services, locations, policies, evidence, dates, and source linksWhich facts a generated answer chooses to include
Page qualityOriginal usefulness, maintenance, accessibility, navigation, and accurate claimsA citation for a particular prompt, user, date, or follow-up question
TrackingServer logs, referrals, UTM-tagged visits, conversions, and lead notesA complete record of every time a model considered or mentioned the brand

Configure the right OpenAI crawler

OpenAI user agents have different jobs

User agentOfficial purposePublisher decision
OAI-SearchBotAutomatic crawling used to surface websites in ChatGPT search featuresAllow it on pages you want eligible for search answers and allow OpenAI's published IP ranges
GPTBotCrawling content that may be used to improve and train OpenAI generative foundation modelsChoose separately whether to allow potential training use; this choice is independent from search access
ChatGPT-UserVisits that can happen because a person asks ChatGPT or a custom GPT to retrieve a pageDo not treat it as the automatic search crawler; OpenAI directs publishers to OAI-SearchBot for Search controls

A publisher may allow OAI-SearchBot and disallow GPTBot. OpenAI documents these controls as independent, so opting out of potential training does not require opting out of ChatGPT search. Do not copy a crawler name or IP list from an old marketing article. Use OpenAI's current crawler documentation because user-agent versions and published ranges can change. After changing robots.txt, check the live file, review server or CDN logs, and allow processing time; OpenAI notes that search-system adjustments after a robots change can take about twenty-four hours.

A safe search-access check

01

Choose eligible pages

Decide which public services, articles, product information, policies, and other pages should be available in ChatGPT search. Keep private, account, test, and sensitive areas out of scope.

02

Review robots.txt

Confirm OAI-SearchBot is not disallowed on intended pages. Keep the decision for GPTBot separate instead of using one broad rule for every OpenAI user agent.

03

Review infrastructure

Check web application firewalls, bot protection, hosting, and CDN logs. An allowed robots rule cannot overcome a network block, challenge page, or repeated server error.

04

Verify public rendering

Make the main facts available without a login and in accessible page content. Avoid requiring a fragile interaction before the core answer appears.

05

Publish a sitemap and internal links

Give public pages stable URLs and meaningful links. A sitemap can support discovery, while navigation explains the relationships between services, evidence, and guides.

06

Monitor rather than assume

Watch logs for successful crawler requests, test key URLs, and record configuration changes. A successful crawl confirms access, not indexing, ranking, or citation.

Publish information that is worth citing

OpenAI does not publish a recipe that guarantees citation, so claims about exact “AI ranking factors” should be treated skeptically. A defensible editorial approach is to make each important page a strong source for its subject. State what the company does, who it serves, where it operates, what an offer includes, and what the limits are. For informational pages, answer the question directly, show how the conclusion was reached, link to primary evidence, and distinguish observations from estimates. This improves usefulness to people and creates clearer material for any search system to evaluate.

A source-ready business pageNo layer guarantees selection. Together, the layers make the page easier to retrieve, interpret, verify, and use.
01Stable identityConsistent business name, ownership, locations, contacts, and official profile relationships
02Clear subjectOne understandable service, question, product, policy, case, or dataset per page
03Original evidenceFirst-party process, examples, research, pricing assumptions, images, or documented outcomes
04Verifiable supportPrimary source links, dates, methods, authorship, and explicit limitations
05Useful pathRelated services, proof, comparisons, and next steps connected with descriptive links
  • Create a complete official about page with the company's real name, role, ownership context, and contact path
  • Describe each core service on a dedicated page with scope, audience, process, boundaries, and a relevant action
  • Publish original examples or cases with permission, context, dates, and limits on what can be inferred
  • Use current primary sources for technical, legal, medical, government, or product facts rather than citation chains
  • Show authorship and review responsibility when expertise or accountability matters to the topic
  • Keep location, hours, availability, pricing assumptions, and policies synchronized across owned pages
  • Correct outdated statements and preserve useful URLs where possible instead of constantly republishing the same idea

Third-party corroboration can help people evaluate a business, but do not turn that common-sense observation into a claimed ChatGPT ranking factor. Earn accurate coverage through real work: professional listings, government licenses where applicable, customer reviews on appropriate platforms, partner pages, event programs, reputable trade publications, and independent reporting. Do not buy fake mentions, publish undisclosed advertorials as editorial proof, or create a network of sites that repeat the same claims. The goal is a truthful public record, not a manufactured citation count.

Connect owned content by meaning. The SEO, GEO, and AI search hub gathers related buyer questions without claiming that one format wins AI citations. The guide to building entity signals for a service business focuses on keeping identity and evidence clear across the site.

When crawl access, architecture, content evidence, and reporting need coordinated work, SEO services can improve the foundation while keeping the outcome language honest. The work should still be measured as a system, not presented as guaranteed placement inside a particular AI answer.

VISUAL CHECKPOINT · SearchA source-ready business page

No layer guarantees selection. Together, the layers make the page easier to retrieve, interpret, verify, and use.

Do not confuse schema with a citation request

Schema.org vocabulary can label organizations, people, articles, products, services, locations, and other visible facts. It can reduce ambiguity when implemented accurately, but OpenAI does not document schema as a guaranteed path into ChatGPT answers. There is also no universal “AI citation schema.” Use JSON-LD that fits the page, keep it consistent with visible content, and validate the code. Never insert fake awards, reviews, sameAs profiles, credentials, addresses, or service areas simply to make the entity look more established.

Track the traffic you can observe

OpenAI says referral URLs from ChatGPT search include the UTM parameter utm_source=chatgpt.com, which publishers can use in analytics. Build a report for those sessions and connect it to meaningful on-site events: service-page views, downloads, qualified forms, bookings, calls, or purchases. Also review server logs for OAI-SearchBot access. Referral reports show visits, while crawler logs show retrieval. Neither one is a full impression report, and neither proves why a generated response cited or omitted a page.

An honest ChatGPT visibility dashboard

SignalUseful interpretationLimit
OAI-SearchBot requestsThe search crawler reached particular URLs at recorded times and response codesDoes not prove indexing, selection, or a user-visible citation
chatgpt.com referralsA user clicked from a ChatGPT experience to the siteDoes not show every brand mention, view, or answer that did not receive a click
Landing-page behaviorVisitors continued to useful content or completed defined actionsSmall samples and consent settings can make comparisons noisy
Qualified leads or revenueObserved visits contributed to a business result under the chosen attribution modelAttribution may miss cross-device, later, assisted, or offline influence
Manual prompt checksA snapshot of what one query returned at one momentResults can vary and repeated self-testing is not a reliable market-share measure

Use the report to find page improvements, not to chase mentions. If referrals reach a pricing article and leave before seeing scope assumptions, make those assumptions easier to find. If OAI-SearchBot repeatedly receives errors, fix the access problem. If an old address appears in public sources, correct the owned site and request changes from the source that controls the listing. Do not rewrite a strong page every week because a manual prompt test changed; generated results can vary, and constant URL or message churn can make the public record less stable.

Avoid five common AI-citation traps

  1. Do not treat GPTBot as the control for ChatGPT search; OpenAI identifies OAI-SearchBot for that purpose
  2. Do not allow a crawler through robots.txt while a CDN challenge or firewall still blocks its published traffic
  3. Do not create dozens of shallow question pages that repeat the same answer in hopes of matching more prompts
  4. Do not buy citations, fake reviews, or invented expert quotes and describe them as authority building
  5. Do not report a visibility score, successful crawl, or one manual test as a guaranteed ranking or business outcome
Can I submit my website directly to ChatGPT for a citation?

OpenAI's public publisher guidance focuses on allowing OAI-SearchBot and maintaining accessible public content. It does not offer a general submission that guarantees inclusion, placement, or citation for ordinary business pages.

Should I allow GPTBot to appear in ChatGPT search?

GPTBot and OAI-SearchBot have different documented purposes. OAI-SearchBot is for ChatGPT search discovery; GPTBot concerns content that may be used for foundation-model training. A publisher can make those choices independently.

Does llms.txt guarantee ChatGPT can cite my site?

No. OpenAI's official search guidance tells publishers to manage OAI-SearchBot and published IP access. No file, markup format, or citation service can guarantee that a page will be selected for an answer.

How do I know whether ChatGPT sent visitors?

OpenAI says ChatGPT search referral links include utm_source=chatgpt.com. Use that value in analytics, review landing-page outcomes, and compare it with server logs. This measures observable clicks and crawls, not every answer impression or mention.