1. Define the research profile
Start with the website, topic, customer problem, audience, market, language, and known competitors. This keeps prompt ideas specific. For example, “Mac storage full” should lead to questions about freeing space, finding large files, System Data, and storage-cleanup tools—not generic productivity software.
2. Review prompts before searching
Prompt suggestions are drafts. Edit their wording, add or remove questions, choose the target platforms, and approve the final set. Keep discovery questions, comparisons, and branded questions distinct so their results remain interpretable.
3. Choose how answers are collected
Automated runs use configured provider APIs. Consumer ChatGPT, Gemini, or Claude apps can behave differently by plan, model, location, and browsing mode. Consumer-import observations therefore stay separate and record the plan, model, sources, and collection time. Missing credentials or provider failures are reported as unavailable rather than as a visibility score of zero.
4. Read each signal separately
- Mention: the answer names the brand.
- Citation: the answer links to or identifies the website as a source.
- Recommendation position: the order in that specific answer, not a universal AI rank.
- Google rank: an organic search observation from the configured keyword service, with its market and timestamp.
5. Inspect the website evidence
Site checks fetch robots rules, sitemaps, and relevant pages. They report what was actually observed: response status, crawl directives, page text, metadata, structured data, and other visible signals. A sitemap or schema fix can improve access and clarity, but it cannot guarantee an AI citation.
6. Turn findings into next actions
Use saved prompts, answers, citations, keywords, competitors, and page findings to make a focused change: repair crawlability, answer an uncovered question, improve a comparison page, or create a content brief. Exports preserve the evidence behind the recommendation.
Limits to remember
- AI output is variable and a sampled run can change later.
- API output is not proof of what every free-plan user sees.
- Keyword metrics depend on the connected service and selected market.
- No content or technical change can guarantee mentions, citations, or rankings.
Open the workspace to create and review a research run.