Claude Research Taking an Hour: What to Check
Check for evidence of progress, protect the useful work, and decide whether to continue or restart a smaller research task.
Updated 5 min read
If Claude Research is taking an hour, check for new activity, an error message, and a wider service incident before restarting it. Elapsed time alone does not prove the run is still working or permanently stuck. Tools like fenbs keep the research question, checked findings, and next steps outside the run so recovery does not mean starting the whole assignment again.
What should you check before waiting any longer?
- Save the original research question and note when the run began. Record when you last saw a meaningful change, such as a new source or a different research step.
- Read the screen for an error, a usage notice, or a request that needs your attention. Do not treat every long wait as the same failure.
- Check the official service status and compare the conversation on another supported surface if available. Preserve unsent text before navigating.
- Decide what minimum result you need now. If you have a deadline, a smaller verified answer may be more useful than waiting for a comprehensive report.
Keep your observations literal: “No visible change since 2:15 p.m.” is evidence; “the server crashed” is a guess. That distinction helps both your next attempt and any support report. Our guide to Claude AI explains the product more broadly if you are unsure which app or feature you are using.
Does an hour mean the research has failed?
There is no reliable diagnosis based only on the clock. Scope, source availability, and the work you requested all matter. Use Anthropic's Research help (opens in a new tab) for current behavior and availability rather than treating a time quoted in an older announcement as a guaranteed completion deadline.
A changing research step is a reason to investigate further, not a promise that a useful report is close. A stationary display could reflect a stalled session or a display problem; neither can be confirmed from an unchanged indicator alone. Compare actual observations over a reasonable interval instead of refreshing repeatedly.
| What you see | What it suggests | Next action |
|---|---|---|
| New activity keeps appearing | The run appears to be advancing. | Keep the original open while setting a deadline for your own next decision. |
| A specific error or limit notice | The message is more useful than elapsed time. | Follow the stated recovery or reset instructions. |
| No change across supported surfaces | A session problem needs investigation. | Preserve the record, check status, then try a small separate test. |
| A report exists but misses the question | The problem is output quality or scope. | Review the findings and request a targeted follow-up. |
How do you rule out a service or access problem?
Open the official Claude status page (opens in a new tab) and read the affected service and incident details. An API incident is not automatically an explanation for every web conversation. Equally, a clear status page does not show that your individual session is healthy. Our Claude outage guide covers that distinction.
If you can start a separate conversation, try one small, public-source question. This is a diagnostic test, not a second copy of the full assignment. Success shows that a basic request can work; it does not prove every connector, long conversation, or source in the original run is accessible.
For a visible usage or length warning, follow Anthropic's error-message guidance (opens in a new tab). Copy the exact message before changing anything. A fresh chat should not be treated as a way around an account limit. Avoid granting broader access just because one research job is slow.
How do you preserve findings before changing anything?
Save what you can actually see: the research question, completed output, accessible source links, open questions, and the last visible step. Do not describe a search preview as a verified finding. If the interface does not expose intermediate work, say that in the handoff instead of inventing a partial summary.
- Goal: the decision or document this research must support.
- Checked evidence: the claim, its source link, and what you personally verified.
- Unfinished work: the questions still unanswered and sources not yet opened.
- Recovery state: what appears to be running, what failed, and what the next person should check before retrying.
Suppose you need a vendor comparison for an internal meeting. Keep two checked vendor documents even if the broader report never arrives. Do not label one vendor the winner until the same criteria have been checked for the others. fenbs Where we left off gives the next assistant that distinction between evidence collected and a decision still pending.
When should you restart with a smaller question?
Restart when you have preserved what is available, checked obvious errors, and decided the remaining wait no longer serves your deadline. Keep the original conversation accessible. If you choose to stop it, use the controls available in your app; do not assume that closing a browser tab cancels remote work.
Replace “research the entire market and recommend a strategy” with one answerable request: “Compare the documented export formats for these three products using their official help pages. Return each claim with its source and mark anything not verified.” Then review that output before commissioning the next stage.
Do not launch several identical long runs in the hope that one wins. Splitting the assignment should reduce duplication and clarify what remains. For an unresponsive workflow rather than elapsed-time triage, see Claude research stuck. If you later automate the work, the Claude API guide covers a separate development path.
Resume the unanswered part
Keep the research goal, checked sources, and remaining questions in a fenbs project. Save Where we left off before switching chats or supported AI apps, including whether the original run might still be active. The next assistant can continue from an explicit gap instead of repeating every search.
Create a free accountFree for one person · No credit card · Works with Claude, ChatGPT, Cursor and more
How to do this with fenbs
- Create a free account and add a project named for the research outcome you need.
- Add tasks for the unanswered questions. Put the recovery approach and the evidence each answer needs in their plans.
- Save checked reference links and a context note distinguishing completed findings from unverified leads. Record what a person chose as a decision.
- Connect a supported assistant using its setup guide, ask it to read the project context, and save progress with the next check before the session ends.
fenbs does not restart Claude Research or extract invisible intermediate results. It is useful when the research question and handoff must survive a stalled session; a one-off question with no follow-up may only need a short note. History records changes to the saved work, while the AI at work hub has more ways to organize ongoing assistant work.
What should a useful support report include?
Include the approximate start time and time zone, app and browser version, exact visible message, last observed progress, and whether a small separate request succeeded. Describe whether the issue recurs with a public-source question. Use a shortened, nonconfidential example where possible, and do not send credentials or private research material merely to demonstrate the delay.
Follow Anthropic's support instructions (opens in a new tab) for your account. A precise report makes the symptom reviewable; “research is slow” leaves too many possibilities. Keep the support outcome with your recovery notes so the next attempt begins with what you learned.