​

Investigating a visibility change

Align comparison conditions, inspect the relevant answers, and separate observed changes from hypotheses to investigate.

Choose equal-length periods with the same prompts, platforms, and regions. Check configuration and sample-size changes first; separate observed movement from causal hypotheses.

Use this sequence once you know the task. Read the numbered steps below for the reasoning and checks.

Complete the CLI quickstart. These commands use CLI 0.5.0. Select the target workspace and use IDs returned by list commands.

A saved baseline and answers from both comparison periods.

Match the period length and measurement scope between your saved baseline and current reports. Choose a question, then use an ID returned by the answer list to read its text.

Check the returned period and dimensions. Choose a baseline with matching conditions and keep mismatched data separate.

Inspect sources cited in answers. Keep citations distinct from crawls and actual visits.

Filter by the target brand or prompt and check answer timestamps. Choose an answer ID returned by this list.

Replace YOUR_ANSWER_ID with an ID from the list. Keep the actual answer wording and citation URLs.

Check answer timestamps and use answers from the compared periods. Keep changed metrics, answer IDs, citation URLs, and hypotheses to investigate. For an update, also record the changed URLs and publication time. If history is insufficient, return to baseline collection. A before-and-after difference does not establish that the update caused it.

Compare platform changes first, then read answers for prompts with material movement. Inspect source retrievals separately from citations, including listicles under the same date conditions.

The losing and trending cohorts reflect retrieval observations. They do not by themselves explain brand visibility or citation changes. Have the agent match the same prompts’ answers and attach evidence, a possible explanation, and a page to investigate to each material movement.

If data is missing, check the workspace, collection period, filters, and pagination. Keep a failed request separate from an empty result. Use --help to inspect supported options.

All workflows · Global options

If a failure remains, report it and verify the fix, including this workflow and the failed step.

Compare platforms using the same questions, regions, and equal-length periods. Add prompt additions or pauses, model and measurement changes, owned-page publication, and competitor changes to the timeline. Separate movement in one platform, movement shared across platforms, and insufficient observations, then inspect the corresponding answers and citations.

Movement in one platform does not prove an engine update caused it. If conditions differ, report the mismatch and collect comparable observations. Keep observed changes, possible explanations, and the next evidence to inspect separate.