How to show up in ChatGPT
Diagnose ChatGPT visibility per product surface and prompt. Inspect OAI-SearchBot policy for search discovery separately from the GPTBot training choice, compare page delivery and answer structure, and record the sources returned for a fixed prompt set. After a documented shipment, repeat the same configuration and report later answers as observations; the comparison cannot identify which change caused movement.
How does ChatGPT decide which brands to mention?
ChatGPT product surfaces and answer modes can differ, and a publisher cannot observe the provider's selection formula from an answer. OpenAI documents OAI-SearchBot for ChatGPT search discovery and GPTBot for potential model training; those roles and controls are not interchangeable.
For a sampled answer with web sources, preserve the exact prompt, surface, date, locale, and returned URLs. Use crawler policy, verified request evidence where available, and page inspection as separate inputs. None establishes indexing, retrieval, selection, or the reason a brand appeared.
Why isn't my brand showing up in ChatGPT?
Treat a missing brand as an unresolved diagnosis. Test search-crawler policy and delivery, the initial and rendered page, indexing evidence where available, answer coverage, entity clarity, and the first- or third-party sources returned for the prompt. These checks narrow hypotheses without revealing why the provider omitted the brand.
Step 1 — Let ChatGPT's crawlers read your site
Inspect the documented crawler role for the ChatGPT surface you care about. A robots.txt rule expresses policy, while edge or origin logs and provider verification can add request evidence. A User-Agent-only fetch is spoofable and shows only how the supplied header was handled; it does not prove what an authentic crawler received or whether the page was indexed or cited.
- Review GPTBot as a separate potential-training policy choice.
- Review OAI-SearchBot policy for ChatGPT search discovery.
- Compare the initial response and rendered page for the exact public URL.
- Keep intentionally public content outside login or paywall gates.
Step 2 — Structure pages so ChatGPT can extract them
A direct, self-contained answer, descriptive headings, and accurate structured data can make the page easier for readers and parsers to inspect. Keep the substantive answer in useful HTML and make markup match visible content. These are compatibility and clarity checks, not a citation formula.
Step 3 — Test third-party source coverage
Record whether sampled answers return reviews, roundups, comparison articles, community threads, or first-party pages. If a relevant third-party source recurs and omits material facts, accurate coverage there can be one hypothesis to test. Its presence does not disclose provider weighting or guarantee a citation.
Step 4 — Track matched prompt observations
Choose category, comparison, and alternative prompts, then record the surface, configuration, returned answer, owned citations, and competing sources. After a recorded shipment, repeat the same cells. Repetition exposes variation and creates a matched timeline; it does not separate the shipment from model, index, source, or run-to-run changes.
Frequently asked questions
Do not represent paid placement as an organic citation. For organic observations, inspect documented access, page content, and returned sources; none guarantees appearance.
When it browses the live web it shows citations you can click. When it answers purely from training it usually does not, which is why broad, consistent coverage across the web matters as much as any single page.
There is no supported universal timeframe. Record the shipment date and compare repeated observations for the same ChatGPT surface and prompt configuration without promising an appearance.
OpenAI documents GPTBot for potential model training and OAI-SearchBot for ChatGPT search discovery. Their controls are independent, and either policy observation remains separate from indexing or citation.
A sampled answer does not disclose why a competitor was selected. Compare the returned pages, source coverage, technical delivery, and repeated results, then test the smallest supported hypothesis while preserving provider variability.