Check the technical access
Review the relevant crawler guidance, page responses, readable content, and canonical URLs. Keep search discovery and training preferences distinct.
1 / Verify the live page
Check that the preferred URL returns a successful response and that the main answer and links are readable. Review redirects, canonical tags, and accidental access restrictions.
2 / Review search crawler controls
Consult OpenAI’s current bot documentation and check controls relevant to OAI-SearchBot. Search discovery and model-training preferences are distinct: allowing one is not the same as allowing the other.
3 / Keep access claims precise
Crawler access supports eligibility for discovery. It does not promise indexing, retrieval, or citation in a particular answer. Do not expose private information merely to seek visibility.
Publish information worth referencing
Answer a clear question, explain what you know, cite attributable evidence, and make the page understandable without hiding its substance behind interface effects.
4 / Answer one clear question
Lead with a useful explanation, then provide the context, steps, or comparison needed to act on it. Avoid pages that simply repeat keywords.
5 / Add attributable expertise
Name the responsible organization or author where accurate. Use original observations or approved examples and clearly distinguish facts from interpretations.
6 / Support and maintain the answer
Link to sources that support the actual claim, state relevant dates, and revisit information that can change. A short factual page is preferable to unsupported length.
Keep the brand and sources consistent
Use clear names, accurate service descriptions, and credible references. Treat third-party mentions as evidence to earn, not proof you can manufacture.
7 / Use consistent names
Keep the brand, services, and preferred URLs consistent across pages. Help readers understand what your organization actually offers.
8 / Earn credible references
Seek relevant, factual third-party coverage through useful expertise. Do not fabricate reviews, endorsements, biographies, or customer results.
9 / Connect related explanations
Link to the pages that give the reader the next needed context. Descriptive anchors are more useful than repetitive keyword lists.
Measure without overstating the sample
Record the prompt, platform, date, observed answer, and cited URL. Keep this sample separate from overall market visibility and from real site referrals.
10 / Keep a dated observation log
Record the prompt, platform, search context, date, answer, and cited URL. Note that output may vary between sessions and that absence in one answer is not a full-platform measurement.
11 / Check actual referrals
Use available website analytics to distinguish a visible citation from a real visit. Do not infer traffic from a screenshot of an answer.
12 / Review the user journey
Check that a visitor can understand your service and complete the real next step. Report inquiry delivery only when an actual delivery system exists.
Check whether the candidate list changed the outcome
The LIVE-GEO preprint tested 80 fictional brands in 18 product families. In its ChatGPT test, the same method placed first 73.8% or 93.8% of the time under two opposite candidate URL orders and 82.5% under randomized order. These results describe an experiment that supplied URLs, which is a different task from measuring how often those URLs surface on the open web.
Follow ChatGPT referrals to the transaction record
MIM:AGENCY reports that an unnamed US electronics store had 879 transactions from AI referrals in January–May 2025. It attributes 773 to ChatGPT, 75 to Perplexity, 16 to Copilot, 14 to Gemini and 1 to Mistral. Those are referral and purchase records. Capture source URLs from a fixed buyer-question sample to measure citations separately.
Repeat observations before judging visibility
A 2026 Swiss-German preprint repeated eight prompts in each of four verticals across ChatGPT, Gemini, Google AI Mode and Perplexity. Across 45–46 observation days, cited-source sets overlapped 34–42% on consecutive days, and brand sets overlapped 45–59%. In that dataset, per-brand detection estimates reached a standard error below 0.10 after 10 days and below 0.05 after 24 days. Report the market and sample dates when comparing changes.
Questions & answers
Does allowing a crawler guarantee a citation?
No. Technical access is only one part of discovery. It does not guarantee indexing, inclusion, a citation, or a recommendation in a particular answer.
Are OAI-SearchBot and GPTBot the same?
OpenAI documents OAI-SearchBot and GPTBot separately. Check its current crawler guidance when setting search-discovery and training preferences.
How can a citation be recorded accurately?
Keep the prompt, platform, collection date, observed answer, and cited URL. This is a scoped observation, not a measurement of the entire market or all AI responses.