Find out why an AI answer is missing.

Check the view before changing the queue.

Confirm the current project, date range, question, and provider. A saved ChatGPT answer does not fill the Claude or Gemini result for the same question. A narrow date range or a mention filter can also hide an answer that was successfully saved.

Open the answer library and review its active filter chips. Widen the period or remove a filter deliberately, then look for the question again. Sorting changes where an answer appears; it does not change whether the answer exists.

• Is this the project where you added the question?
• Does the date range include the expected collection time?
• Are you viewing the intended provider?
• Could a brand, outcome, or analysis filter exclude the answer?

Read the question and provider state.

Tracking shows collection progress by question and provider. A queued request is waiting to be claimed. An opening or running request is being handled by the browser. A saved answer is available for review, while a failed request needs its recorded error inspected.

A question can have a saved answer from an earlier round and an active request in a later round. Use the timestamps and open the relevant saved answer rather than assuming one label describes all of its history.

Check whether the question is enabled and whether daily tracking is enabled for the intended providers. The daily check time is a scheduling time, not a promise that every provider will finish at that instant. Daily allowances and existing attempts can affect which requests enter the queue.

Check the browser connection and provider access.

Browser collection needs a connected runner and access to the intended provider in that browser. A working Echuno dashboard alone does not establish that the browser worker is online or that its provider session is ready.

Review the connection status in the browser connection view. In the runner browser, inspect the provider for a sign-in screen, rate limit, verification page, or another visible interruption. Resolve the actual interruption before deciding whether a new collection is needed.

Temporary Chat or Incognito is a provider conversation mode. It does not remove provider account requirements or guarantee that a provider will accept a request. A readiness failure is missing collection data, not a brand omission.

Separate queue wait from answer collection.

Queue wait runs from the time a request enters the queue until a worker starts it. Collection duration runs from that start until the answer is saved. The end-to-end time includes both stages.

For an illustrative request queued at 10:00, started at 10:06, and saved at 10:06:30, queue wait is six minutes, collection is 30 seconds, and end-to-end time is six minutes 30 seconds. It would be misleading to call that six minutes of AI generation.

Earlier requests, runner availability, and a deliberate maintenance pause can increase waiting time. Collection duration can also include browser navigation, readiness checks, and saving the response. Compare the recorded stages rather than attributing the whole delay to the provider.

A saved answer can still need assessment.

Collection and brand analysis are separate stages. An answer can be safely saved while its mention, position, sentiment, and evidence assessment is pending or needs attention. Open the full saved answer to establish what was collected.

Not assessed means there is no usable assessment for that brand in that answer. It does not mean the brand was absent. Likewise, a recorded mention with no explicit rank can be valid when the answer does not give a clear recommendation order.

Review the analysis state and supporting evidence before using the answer in a comparison. If analysis needs attention, investigate that stage; collecting the same question again creates a different observation and does not explain the first assessment gap.

Keep unresolved gaps visible in your review.

Record the affected question, provider, collection time, and observed state. Keep collection failures separate from assessed answers that do not mention the brand. This preserves the difference between missing evidence and negative evidence.

Before starting another run, check for an active request and inspect the existing error. A new answer may be useful after an interruption is resolved, but repeated submissions can create duplicate observations and extend the queue.

When sharing results, state which providers and questions have completed and which still need collection or assessment. A partial round can be useful to inspect, provided its limits remain clear.