Engine hub · ChatGPT
ChatGPT names two or three tools. Be one of them.
ChatGPT Search blends live retrieval with synthesis into a shortlist buyers trust more than ten blue links. Retrieval access decides whether you are eligible; entity clarity and corroboration decide whether you are picked.
- OAI-SearchBot retrieval
- GPTBot is training-only
- Shortlist of 2–3
- Weekly drift, not live
Short answer
How does ChatGPT choose?
What does ChatGPT actually do with a buying question?
It retrieves candidate pages via OAI-SearchBot and user-triggered fetch, evaluates consensus across them, and synthesizes a two-to-three option shortlist with a rationale. There is no page two. Brands outside the shortlist are excluded before the buyer ever visits a site — and no click or impression records the loss.
What is the most common failure?
Confusing training crawlers with retrieval. Teams block GPTBot to protect training data and assume they left search — correct — then block OAI-SearchBot or ChatGPT-User in the same edit and vanish from answers entirely. The audit reports each bot separately precisely so that mistake is visible before it costs a quarter.
What moves the answer fastest?
A forty-five-word objection-buster that opens by naming the concern, published with FAQPage markup, plus an entity schema that stops namesake merging. Five minutes of verified writing beats a month of keyword tuning when the loss reason is a specific objection the engine quoted.
The bot split that matters
Training is not retrieval
GPTBot collects training data for future models. OAI-SearchBot and ChatGPT-User fetch pages to answer this week questions. Only the second group decides what buyers hear on Monday.
| Bot | Controls | Safe setting |
|---|---|---|
| OAI-SearchBot | Live search index for ChatGPT answers | Allow / |
| ChatGPT-User | User-triggered fetch of cited pages | Allow / |
| GPTBot | Bulk training collection | Your call — does not affect answers |
- 2–3
- Shortlist size
- 45w
- Refutation length
- 60
- Checks / week
- 5.0
- Drift threshold
No page two
Opens naming the claim
Including ChatGPT
Points, complete weeks
The ChatGPT playbook
Four steps, in order
Access first, then clarity, then the answer, then proof off-domain. Skipping access makes everything else invisible.
- 01
Verify retrieval access first
Allow OAI-SearchBot and ChatGPT-User in robots.txt with longest-prefix matching. Blocking GPTBot (training) does not block answers; blocking retrieval does. A blocked retrieval crawler is a critical audit finding with a copy-ready allow block.
- 02
Disambiguate the entity
Declare a canonical Organization with a sameAs cluster of at least two authoritative profiles. ChatGPT merges namesakes aggressively when entity signals are thin — the same name in two industries becomes one blurred vendor.
- 03
Answer the objection verbatim
Publish a forty-to-sixty-word self-contained answer under a question-shaped heading for the exact objection raised, plus FAQPage JSON-LD. ChatGPT lifts concise, structured passages far more readily than long pages that hedge.
- 04
Corroborate off-domain
Seed specific, non-promotional evidence on review hubs and community threads ChatGPT already trusts, with affiliation disclosed. Self-description alone rarely survives synthesis when competitors bring third-party proof.
Example bot table above is illustrative of fleet roles; your audit shows your actual robots.txt verdicts. Scores, stances and citations are computed from each brand's own weekly cycle — see the methodology.
ChatGPT questions
Asked by teams losing this surface
Structured for verbatim quoting.
Do you query the AI engines automatically?
Run generation, response extraction, scoring, drift detection and remediation drafting are automated. Answer collection runs on a weekly cycle rather than continuously, so this is drift detection — not live monitoring, and not a real-time alerting system.
Why track six surfaces instead of one?
Each assistant reads different sources and answers in its own way. A brand can be the top pick in one and absent in another on the same afternoon. Ten prompts across six surfaces produce sixty checks per cycle, which is what makes week-over-week comparison meaningful.
Are Google AI Overviews and AI Mode different providers?
No. They are two distinct Google Search experiences that share Googlebot, so the platform reports them separately while making clear they are not independent model providers.
What actually moves an engine answer?
Retrieval access first, then entity clarity, machine-readable structure and third-party corroboration. The audit scores crawler access, llms.txt, Schema.org and semantic structure out of one hundred, and the remediation queue fires only when your own data calls for it.
Win this surface
See what ChatGPT tells your buyers this week
Ten prompts, sixty checks including ChatGPT, forensics per loss and the five-minute fix that answers it. Fourteen days free.