The honest summary: the MVP is an automated OpenAI Codex snapshot with web search. It is designed to identify useful visibility signals, but it is not a replay of every individual ChatGPT experience.
1. Monitoring brief
You submit one domain, one to five services, at least one accepted brand name, one primary city or market and one language. Website inspection can suggest these values, but you confirm or replace them. A run keeps a snapshot of the confirmed brief so its findings can be traced to the inputs used at that time.
2. Prompt execution
The system creates a repeatable set of service-and-location questions and executes them through OpenAI Codex with web search enabled. The workflow asks for useful recommendations and source evidence, then stores a structured representation of the output.
OpenAI systems, models and search sources may change. Availability and depth can also depend on third-party and subscription limits.
3. Signals shown in the report
- Brand mention: the submitted brand or an accepted alias appears in the answer.
- Direct recommendation: the business is explicitly presented as a relevant option, rather than merely appearing in background text.
- Site citation: a source or citation resolves to the monitored hostname or an accepted canonical hostname.
- Competitor: another identifiable organisation is recommended or meaningfully surfaced for the same service context.
- Source: a web page or domain cited in the generated answer.
4. Totals and scores
Report totals are counts across the checks completed in that run. For example, “brand mentions” is the number of service checks where the brand was detected, not an estimate of the percentage of all ChatGPT users who would see it. If a composite visibility score is shown, the report identifies it as a derived summary rather than an OpenAI metric.
5. Variability and comparison over time
Generative answers are non-deterministic. Two prompts with the same wording can produce different recommendations or sources. Location, product version, account context and model updates may also affect results. A single run should be treated as a snapshot. Repeated runs are most useful for spotting persistent patterns and material changes, not minute ranking movements.
6. Relationship to ChatGPT
Codex and ChatGPT are OpenAI products, but their execution context and displayed answers are not guaranteed to be identical. Therefore the MVP reports its surface as OpenAI Codex with web search. It does not claim that every ChatGPT website or app user received the same response.
7. Review before action
The system can misclassify a similarly named brand, miss an indirect citation, or reproduce inaccurate source material. Check the recorded answer and sources before making SEO, advertising, medical, legal or commercial decisions. Contact Rotgar if a run appears technically invalid.
8. Planned evolution
Future method versions may add more providers, sampling, prompt families and stronger calibration. A method change should receive a new version so results produced under materially different methods are not presented as perfectly comparable.