Write an AI visibility update people can use.

Put the scope beside the headline.

Name the project, reporting period, and platform selection before quoting a result. A report for ChatGPT in the last seven days describes a different sample from an all-platform report covering 30 days. Include the generation time when sharing a snapshot that may be read later.

Use the saved-answer report for browser-collected and manually saved responses. Keep its results distinct from API tracking measurements. Different collection methods can produce different answers, even when the buyer question sounds similar.

Say what the sample represents: selected buyer questions asked through the configured collection workflow. It does not measure every conversation people have with AI, and a report is not a count of all potential buyers who saw a recommendation.

Show counts alongside percentages.

Mention rate uses answers with a recorded assessment of the relevant brand. If a brand appears in six of 12 assessed answers, its mention rate is 50%. If four additional answers are saved but unassessed, they remain unknown; they do not become four missing mentions.

Report the saved-answer count and the assessed-answer count separately. If a competitor has a different assessment count, do not assume that its rate describes the same response set. Inspect shared answers before making a direct competitive claim.

Keep recommendation rank separate from mention rate. Average rank uses answers with an explicit recorded order. An unranked mention is still a mention, and a missing rank does not assign the brand the last position.

Explain what is still missing.

Review platform coverage and the questions included in the report. Ten saved answers for one question are different from one answer for each of ten questions. A provider without saved answers has missing observations, not a confirmed zero mention rate.

Check Tracking for queued, running, or failed collection and inspect the answer library for pending assessments. Keep those gaps outside claims about brand absence. If a round is incomplete, share the partial evidence with that limitation stated.

When comparing periods, check that question wording, provider mix, brand identities, and collection conditions are reasonably consistent. A change in the sample can explain a change in the aggregate result.

Choose a small amount of supporting evidence.

Pick a question that explains the main observation. Open the full saved answer and check the mention assessment, recommendation order, and supporting quote. If sources are relevant, inspect the saved reference links too.

Keep the observation separate from your interpretation. An observation might be that a competitor appears in an explicit shortlist while your brand is absent from that saved answer. A hypothesis might be that the answer emphasizes a capability your website describes poorly.

End with a focused next step, such as reviewing the cited page or collecting a comparable follow-up sample. A changed answer after a website edit does not, by itself, establish that the edit caused the change.

Choose the export for the person reading it.

Use Copy report or Download report for a Markdown summary that can go into a team update. Preview report text shows the same Markdown before you share it. If clipboard access is unavailable, the preview provides text you can select and copy manually.

Use CSV when someone needs the underlying platform, brand, and question summary rows in a spreadsheet. The export includes the selected platform scope, reporting period, and generation timestamp so the rows can be interpreted later.

The on-screen question review shows a short shortlist, while downloads include every question in the selected report scope. Check the scope shown above the export controls rather than assuming the visible shortlist is the complete export.

• Team update: Markdown with scope, counts, and a short interpretation.
• Further analysis: CSV with the exported summary rows and denominators.
• Evidence review: open the saved answer behind the relevant brand or question.

Keep the update short enough to act on.

Start with what you observed, then add the sample and the next action. For example: In our selected ChatGPT sample, the brand appeared in six of 12 assessed answers. Four saved answers still need assessment. We will inspect the three discovery questions with missing mentions before choosing a content change.

Add one qualification when it materially affects the conclusion: an incomplete provider round, changed question wording, or a small sample. Keep the complete report attached so someone can inspect the details without making the headline carry every row.

At the next review, return to the same scope and inspect the new evidence. A useful reporting routine creates a traceable series of observations and decisions, rather than a sequence of disconnected percentages.