Availability
Open Conversations to search the sources available to your account. Complaint analysis is an optional dashboard pilot and is shown only when enabled for your account. If you see “Complaint analysis is not enabled for this account yet”, you can still read and export retrieved mentions; this guide does not grant pilot access.
The pilot’s analysis features described below do not form a public API-key service. See For API and MCP users for that distinction.
Which tool should I use?
| Your task | Start here |
|---|---|
| Find pages, articles or other search results. | Search pages & articles |
| Read social mentions and, where enabled, review likely complaints. | Conversations |
| Create a saved research briefing with cited claims. | Research reports |
Run a search
- Enter a subject. Start with a name such as “Spotify”. For an ambiguous name, use Add context to describe what you mean, such as “music streaming”. A description or comma-separated related terms guides analysis; posts do not need to contain those exact words. This does not guarantee that every retrieved mention concerns the right subject.
- Choose up to three sources. Open More platforms for other options. Coverage varies: some sources need a handle or profile URL instead of a keyword. Source notices explain unavailable or unsupported requests.
- Choose a time window. Options include Any time, Last 7 days, Last month, Last 6 months and Last year. With a window, displayed mentions must have a verifiable publication date inside it. Older and undated results are excluded and counted in the notice. Mentions are displayed newest first; source coverage can still be incomplete.
- Choose Quick search or Deeper search. Quick makes a smaller collection. Deeper can make up to three collection rounds within the displayed maximum cost, stopping early when sources run out or a round adds no new mentions. Neither option promises a fixed result count or every online mention.
- Check the maximum credit cost and select Find conversations. Retrieval uses your credits. Where enabled, relevance checks and complaint analysis are included during the pilot, subject to its limits. Deeper retrieval and analysing more saved text are different actions.
How the first analysis batch is chosen
Where analysis is enabled, Conversations checks which returned texts discuss your subject, then selects up to 20 for analysis. Meaningful experience, evaluation, questions, workarounds, praise and criticism take priority over passing mentions. Sources take turns among similarly relevant candidates; a weak mention does not get priority just to fill a source slot.
The mention list stays newest first even when the analysis order differs. Open How mentions were selected or an item’s Analysis details to inspect the selection. Original records are kept, including lower-priority material.
A batch can take up to 90 seconds. Mentions without body text or with more than 4,000 characters of combined title and body are skipped. An eligible mention can remain queued when the batch, relevance-check or time limit is reached. The displayed breakdown separates queued items, text exclusions and failed checks.
Read the labels and themes
What needs attention groups likely complaints into suggested themes. Worth reviewing keeps possible problems separate. A theme can contain several mentions, so the number of cards need not equal the complaint count.
| Label | Meaning |
|---|---|
| Likely complaints | The returned text supports a complaint about the intended subject. Read the evidence before acting; neither the claim nor the author’s customer status is verified. |
| Possible complaints | A possible problem about the subject needs closer review, including who or what caused it. |
| Uncertain | The available text does not support a confident conclusion. This is not a negative or positive verdict. |
| No complaint identified | No supported complaint was identified in this text. It may contain praise, neutral discussion or a question. |
| Unrelated | The analysis identified another subject or a namesake. The original mention remains inspectable. |
| Not analysed | No completed label: the mention may be queued, lack body text, exceed the text limit or have a failed check. Read the displayed reason. |
Appears promotional describes marketing wording, not verified sponsorship. A promotional post can still contain a supported complaint. These findings describe the returned, analysed sample; they do not measure overall customer satisfaction or prove an allegation. No complaint themes can also mean too little usable evidence, rather than no problems with the product.
Inspect the original evidence
Cards show either a Selected passage from the returned text or Full returned text when no precise passage was selected. Long evidence starts collapsed and is explicitly marked as a preview. Use Read full passage or Read full returned text to see all wording and qualifications, then collapse it again.
Use Read surrounding context when available and follow the original source link. Dates remain visible. These are excerpts from retrieved text, not replacement quotations generated by a model. Missing comments, full threads and video content are not analysed automatically.
Continue from saved results
- Open a saved run / Check saved labels: reads stored results without another search or model call.
- Analyse next mentions: analyses up to 20 more eligible saved texts without retrieving them again. Pilot analysis allowances still apply.
- Retry failed checks: retries eligible failed analysis or relevance checks against saved text. Missing body text is not fetched by this action.
- Update saved analysis: explicitly applies a revised analysis policy to the same saved pool. A ranking-only update recomputes relevance and reuses compatible successful complaint answers. Earlier analysis snapshots are preserved; simply opening history never starts an update.
Completed labels can accumulate beyond 20 across batches and updates. “Analysed” counts those completed labels; “Not analysed” covers the remaining displayed mentions. Relevance ranking by itself does not count as completed complaint analysis. If an allowance or service failure stops analysis, original mentions and exports remain available.
Filter and export
Use the label tabs and text filter to narrow the loaded mentions. CSV with labels exports the filtered result set with available analysis labels, topics, selection details and evidence fields. Without analysis, the button is simply CSV. Reading, filtering and exporting loaded results does not start a paid search.
Original JSON preserves the search response. It does not gain the dashboard’s analysis labels or lose items excluded from the displayed time window. Use labelled CSV when you want those annotations; use original JSON when you need the underlying retrieval response.
For API and MCP users
Public search endpoints retrieve data. The Conversations dashboard adds its own saved-run workflow, publication-date checks, selection, complaint analysis and evidence display. Do not assume those dashboard steps or labelled CSV fields are part of a search API response.
For retrieval, start with search/all or a platform-specific search in Which endpoint?. Read the current endpoint reference for supported parameters, date filters, pagination, price and availability. Each additional API page or retrieval request follows that endpoint’s metering.
There is currently no public /v1 API-key endpoint or dedicated MCP tool for this dashboard’s complaint labels, analysis batches, saved-analysis retries or evidence passages. Dashboard sign-in is required for those actions where enabled. Do not build an API-key integration against private dashboard routes. You can retrieve mentions through the documented API and analyse that returned data in your own application.