Contents 10
Identify what happened in the answer
A general provider question checks the options selected for a stated problem. A branded question checks the company description. If you supplied the brand in the question, its appearance is not independent evidence of a recommendation for the broader topic. Keep these groups separate even when they concern the same service.
For an inaccurate response, isolate the statement: a wrong market, an old fee or a service the company does not provide. A judgment such as not the best without a verifiable fact needs different treatment. Do not combine a brand mention, a website citation and buyer suitability into one score.
| Observation | What it establishes | Next check |
|---|---|---|
| Brand absent | Absence in this response under these conditions | Buyer scenario, repeat observations and access |
| Mention without a link | Company named, source not established | Description accuracy without guessing which website was used |
| Inaccurate fact | Conflict with substantiated current information | Source, fact owner and correction route |
| Citation does not support the claim | Claim-to-source mismatch | Exact page text and the unresolved question |
Preserve the test conditions
The log needs the exact question, platform, available mode or model, language, market, date and complete response. Record search use, conversation history and known personalization conditions. Leave unknown fields open. Do not call two incomparable answers a trend merely because they were obtained a week apart.
An API observation and a consumer-interface response are different environments. Repetition helps reveal variability; it does not establish the whole market's questions about the brand. Keep answers without mentions in the report. Record retrieval failures and unavailable modes separately so they are not presented as confirmed company absence.
- Exact query and the test purpose.
- Platform, mode, language, market and date.
- Complete response and cited sources.
- Fact status, verification owner and next action.
Check the facts and cited sources
Open the cited link and locate the passage relevant to the error. Check for a different page, outdated material or another company with a similar name. If the source does not substantiate the statement, record the mismatch. The presence of a citation does not establish the answer's accuracy on its own.
Then reconcile current information across the website and controlled profiles: company identity, service scope, markets, fees and enquiry paths. A hypothetical conflict is an old page saying support is included while the current offer has different terms. Check this with the service owner. Do not reinterpret a historical case using current rules if it accurately describes its own period.
| Log field | What to record |
|---|---|
| Prompt and conditions | Exact wording, platform, search mode, language, region and date; mark unknown conditions explicitly |
| AI statement | The specific verifiable fact rather than the entire brand judgment |
| Response source | Cited URL and supporting passage or lack of substantiation |
| Current fact | Approved public page and verification date |
| Fact owner | Who confirms current information and approves a correction |
| Status | Confirmed error, accurate fact or unresolved question |
Which owned and external pages to correct
Correct substantiated errors on your own website, check the relevant translations and reconcile visible copy with metadata. Preserve the previous version and change date. If a new page is needed, identify the question it answers and the source supporting that answer. Adding ChatGPT to a heading does not itself correct a fact.
For an external article, send the author or editor a precise correction request: what is outdated, which fact is substantiated and where to verify it. The source owner decides whether to change the page. Do not disguise advertising as independent coverage or create fabricated reviews. If no source is identifiable, do not assert that a particular page caused the response.
Source: Google Search spam policies (EN) ↗
When a technical check is useful
When the company is absent, inspect the relevant page's accessibility, indexing, canonical URL and infrastructure restrictions. Google has a separate Search generative AI control. OAI-SearchBot is used for ChatGPT Search. GPTBot collects content for model training, while ChatGPT-User visits pages for user-initiated actions. Keep search permission separate from model-training permission.
These checks can uncover an actionable problem, but they cannot explain every absent brand or guarantee the next recommendation. The question may not fit the company's actual offer. Before a technical change, identify the confirmed fault, owner and verification method. Do not change access policy automatically because of one unsuccessful query.
Source: OpenAI Docs crawler overview (EN) ↗
Source: Google Search generative AI control (EN) ↗
How to verify corrections without overstating the outcome
First establish that the approved fact is correctly published and accessible in the required language. Then repeat the observation under recorded conditions and compare the specific disputed statement. A new accurate response can be useful evidence, but it does not prove that every user sees it or that the correction produced additional sales.
Assess referrals and enquiries separately from the response log where their source can be identified. Native Google AI impressions, GA4 and CRM records answer different questions. For a complicated or recurring error, bring the log, sources and fact owners into a GEO audit. This creates a verifiable task rather than a demand to force the AI to name the company.
Source: Google Generative AI performance report (EN) ↗
Where to start
If ChatGPT misses the company or describes it incorrectly, start with a log of the query, conditions, statement, source and current fact. Correct substantiated problems and verify the outcome separately from referrals and sales. Bring these records to a UPLIFY SEO/GEO audit when you need joint diagnosis.
Questions about brand mentions in ChatGPT
Does an absent company mean the website has a penalty?
No. Absence in a particular response does not establish a penalty or technical fault. You need the exact scenario, conditions, repeated observations and an accessibility check. The business may not match the buyer's problem or may be incompletely described in available material. State only a substantiated cause; where it is unknown, keep the question open and define the next check rather than inventing a diagnosis.
Is changing the homepage enough?
It depends on the confirmed error. If a current homepage fact is inaccurate, correct and approve it. The response may nevertheless cite another page or lack an identifiable source. Check important translations, controlled profiles and the actual citation path. Accept the published information correction without promising that it automatically changes every ChatGPT response for each reader and their own selection scenario.
Can we test recommendations with our brand in the question?
That question is appropriate for testing a company description, but not as an independent general-recommendation test. Supplying the name changes the meaning of its appearance in the response. Separate branded questions, general provider selection and specific fact checks. Keep comparable conditions within each group and report what the observation establishes, as well as what it does not establish.
Sources and verification date
Documentation and copy checked on October 12, 2026.
- Google Search generative AI control (EN)
- Google Generative AI performance report (EN)
- Google Search spam policies (EN)
- OpenAI Docs crawler overview (EN)