Skip to main content
Use filters to select the answers you want to examine, groups to split the result, and a subject to choose the brands or sources returned.

​
Discover the available dimensions

Call list_dimension_options for the project. It returns options for platform, topic, prompt_type, intent, and user_profile, based on the latest run. An empty options list does not create a usable dimension. For an older run, confirm that the requested values exist in that run and inspect filter warnings. Use returned labels exactly. A topic filter matches the named topic; “running shoes” should not be assumed to select “trail running shoes.” In the product, prompt_type corresponds to the Category facet and intent to Customer Intent.
This is an argument fragment for tools that accept these fields. Replace the example values with your project’s options. Read data.filters_applied and its warnings before presenting the result; do not assume the requested filter was applied merely because the call succeeded. Some grouped responses identify platforms with codes. The worked example returned OPENAI for ChatGPT Search and VERTEX for Gemini. Preserve the returned code and its known display label in an export. Do not invent labels for an unknown code. Grouping support depends on the operation. Citation-score URL results do not apply group_by in this edition. Grounding-source platform grouping is available on the all-brand, unrestricted-sentiment path; narrower brand or sentiment requests can drop it. Check the operation’s parameter table and returned scope rather than assuming a dimension works on every tool.

​
Choose the subject for citation tools

​
Choose the subject for brand tools

Brand measurement tools use leaderboard, all, self, self_and_competitors, or specified. For specified, provide entities_or_bundles:
Optional selection flags include always_include_self, always_include_competitors, and include_non_brands where supported. Citation selection uses always_include_owned. Do not swap these two subject formats.

​
Check what was actually returned

Most measurement leaderboards default to 25 rows, with additional configured or owned rows where applicable. Inspect subject_summary.n_returned and n_available. Their difference can reflect your deliberate selection as well as truncation. These counts can include rows across groups, so do not label them “unique brands” without checking the grouping. unindexed_entities means the requested name was not measured in the run’s entity set. It does not prove the name never appeared in the answer text. unscored_entities means a requested entity lacks the sentiment result being requested. Read served_aggregation_level or requested_aggregation_level when returned. Mention coverage requested at bundle level can be served at entity level; adding alias coverage values would double-count responses. Sentiment also returns entity-level rows when bundle is requested in this edition. The grounding-source HTTP route does not apply subject; use its supported brand_subset selection. Next: Response fields and units.