Comet assistant not working: check tabs and permissions

If the Comet assistant is not working, first test whether it can read the intended webpage, then test the action separately. Reading a page and controlling it require different permissions in Perplexity’s Comet controls↗ (opens in a new tab). Keep the result in fenbs so another person or AI can continue the investigation without repeating every test.

What should you check first?

1. Write down the exact page and the task that failed. Separate “read this paragraph” from “fill this form.”
2. Open the intended webpage in Comet. Use the Ask Button for a small question about something visible on that page.
3. Compare the response with the page yourself. If it describes another page or gives a generic answer, do not proceed to an action.
4. If reading succeeds, ask for one harmless action you have authorized. Check the page afterward rather than relying on the assistant saying it finished.
5. For an organization-managed browser, ask the administrator to check the site’s effective permissions before changing anything else.

The getting-started guide↗ (opens in a new tab) documents the Ask Button and Summarize Button. Start with one of those documented entry points instead of looking for an imagined screen-sharing switch. Comet works with webpage context; that does not establish unrestricted access to your desktop or camera.

Use a public, non-sensitive page for the first test. For example, ask for the title and one heading of a page you can inspect yourself. A short, checkable question gives you much better evidence than asking the assistant to complete a long workflow again.

Is the assistant using the right tab?

Name the page you want it to use and state what it should extract. Avoid instructions such as “use that page” when you have several similar pages open. Ask for a detail that distinguishes the intended page from the others before trusting a summary.

Comet prompts can reference tabs with @ mentions in the search bar or sidecar, as described in Comet Shortcuts↗ (opens in a new tab). That makes the intended source more explicit. It does not cancel a permission restriction or prove that every open tab is readable.

If the answer is wrong, record the prompt, the expected detail, and the detail returned. Then try the same reading question with only the intended source named. Change one thing at a time; otherwise you will not know which change helped.

For the normal workflow, use the Comet assistant guide. This page is for recovering a failed session. A saved conversation is a different problem, covered in managing Perplexity history.

Can it read the site but not act on it?

The permission guide↗ (opens in a new tab) gives organization administrators three domain settings: Browser Control, Read Only, and No Access. Browser Control permits reading and actions; Read Only permits reading; No Access permits neither. Domain rules override global assistant settings, so check the particular domain rather than assuming the global switch settles the question.

For an Enterprise administrator, the documented location is organization settings, Comet Setup → Assistant Permissions, with site rules under Domain-Specific Assistant Permissions. Ask for the smallest change needed for the task. A blocked action can be the intended policy, and a successful reading test is not approval to submit a form.

How do you retry without making the problem worse?

Before repeating anything that could send, submit, book, or change a record, check whether the first attempt already took effect. Look at the actual destination and ask the responsible person if the result is uncertain. Repeating a request is not a substitute for finding out what happened.

• Reduce the task to one page and one action with a visible result.
• State what the assistant may do and where it must stop for your review.
• Keep the page address, your prompt, the observed result, and any refusal text together.
• Do not include passwords, session tokens, or private page contents in a troubleshooting record.

Say you are testing a contact-form workflow. First confirm that the assistant can identify the form fields. Then, if authorized, test filling harmless draft text while requiring it to stop before submission. Those are two distinct checks, and neither proves that a later submission has succeeded.

If you need a clearer instruction, use the patterns in prompts for Perplexity. Record a successful test and its limits in fenbs’s shared work record, so the next attempt starts with evidence.

What should you keep before ending the session?

Save the smallest explanation another person could follow: the goal, the affected domain, the reading-test result, the action-test result, and the next check. Mark an administrator’s review as still pending if it has not happened. Do not turn a suspected cause into a confirmed fix.

Keep separate notes for separate symptoms. A Comet action failure does not by itself explain a missing Space; Perplexity Spaces troubleshooting handles that problem. This distinction helps you avoid reporting every interruption as the same bug.

How to keep your troubleshooting progress with fenbs

1. Create a free account.
2. Save what you were trying to do, what you checked, and what happened. Leave out private page content and sign-in details.
3. Connect Perplexity to fenbs through its supported custom connector route if your administrator allows it; see connections.
4. Write the next small check before you stop. The next connected AI can read it so you can carry on from exactly where you left off, on any device or in another AI.

fenbs keeps the investigation and the next step. It cannot change Comet’s browser permissions or make a blocked website accessible. If you only need to ask one question about a public page, saving a lasting work record may be unnecessary.

For more everyday workflows, browse AI at work. Keep the finished fix short enough that you would actually read it before running the same task again.