23SEOGEO · Weekly focus ·

ChatGPT brand mentions: audit names and citations separately

Audit ChatGPT brand mentions, aliases, citations and competitors using fixed prompts and original answers. Separate web observations from model answers and missing sources.

the 23SEOGEO team · Updated

Concept illustration separating a named brand from a linked web source
Concept illustration separating a named brand from a linked web source

The short answer

To check whether a brand appears in ChatGPT answers, fix buyer prompts and brand aliases, then retain the date, web-search state, full answer and citation URLs. A brand name is a mention; a traceable source link is a citation. A missing source list is unknown rather than no citations. Retest the same prompts before comparing the results with this website’s visits and enquiries.

From question to reviewable evidence

Set a baseline

Build fixed buyer questions for one brand.

  • Confirm brand aliases
  • Retain date, engine and web-search state

Delivery evidence

Original prompt, full answer and sources.

Open AI Citation Tracker
Full action and acceptance checklist
  1. Set a baseline

    Build fixed buyer questions for one brand. Confirm brand aliases · Retain date, engine and web-search state

    Original prompt, full answer and sources.

    Open AI Citation Tracker
  2. Review sources

    Match mentions and citations to source passages. Exclude namesake businesses · Unknown sources are not zero; invent no evidence

    Mention passage, URL and material for review.

    Organize owned facts and sources
  3. Retest the same prompts

    Retest unchanged prompts instead of creating growth by changing the sample. Summarize only verifiable web records · Separate visits, enquiries and sales

    Comparable samples and the site’s referral window.

    Review AI traffic and conversions

Fix prompts without steering the answer

Collect questions from support, sales and verified Search Console queries across brand, category and buying scenarios. Brand prompts test factual accuracy; category prompts show whether the brand appears without being named in advance. Do not instruct the model to recommend you and count the response as an organic hit. Keep language, market and engine comparable and record the interface or call method. Accounts and timing can still change answers.

Review aliases, products and citations separately

Record owner-approved Chinese and English names and product aliases to avoid matching a different company. Locate the mention in the retained full answer, then inspect whether the source URL belongs to your site or an actual third-party reference. A spelling match does not prove correct citation, and a listed URL does not mean the source page was read. Missing Sources, malformed lists and restricted pages retain unknown or failure states rather than zeros.

Monitor answers; improve the cited-page gap

When a competitor appears for the same prompt, inspect what its cited page answers: product conditions, regions, audience, sources and dates. Translate controllable gaps into improvements on existing pages using real owned facts, not invented cases, credentials or endorsements. Google AI features do not require special files or special schema. Do not portray a shared practice as another engine’s official guarantee.

Only verifiable web records enter measured observations

A plain model answer can reveal a brand-description problem but does not automatically qualify as a verified web observation. Summarize only completed, non-simulated engine-web records with verifiable retrieval evidence. Keep failures, pending runs and unknown sources separate. One screenshot is not a trend; recurring fixed-prompt tests support comparison. Site ownership is required for referral collection, enquiries require business review, and events on the same date do not prove a citation caused a sale.

Separate four records from an answer

Separate four records from an answer
RecordEvidenceBoundary
Brand mentionApproved name or alias in the answerNot a linked citation
Linked citationRetained, reviewable source URLMissing list stays unknown
Verifiable web observationComplete, non-simulated, retrieval evidencePlain model answers do not upgrade automatically
AI referral visitAuthorized site collectionNot proof of a human or a sale

Related questions

  • How to track brand mentions in ChatGPT?
  • How does ChatGPT brand monitoring work?
  • What evidence should ChatGPT visibility tracking retain?

Common questions

Does no source list mean no citations?

No. Missing or malformed sources are unknown, not evidence of a complete check.

Do English brand aliases count?

Yes, when it is an owner-approved alias for the same brand. Mentions and citations remain separate.

What if the brand is absent?

Check the prompt, sources and readable pages; improve one real content gap and retest the same question.

Sources

Topics and tools