Workspace troubleshooting

Most workspace problems come from a vague purpose, a vague request, or expecting the AI to know something it cannot. The publishing ones nearly always come down to page status.

Publish says the landing item must be Live

The publish dialog refuses with set the landing item to Live before publishing.

Open Configure page on the item readers should see first. Set Page status to Live and turn on Use as landing page. Both are required — a Live page that is not the landing page still leaves you unable to publish.

Publishing produced fewer pages than I expected

The publish dialog counts them for you: items to publish, then N draft items omitted.

Everything starts as Draft, and drafts are never published. Set each page you want readers to see to Live in Configure page.

A Live page is missing from the published copy

On the version tested, a workspace with two Live pages published both — the publish dialog counted them and gave each a published version — but the published contents list linked only the landing page, and the navigation group did not appear. Republishing did not change it.

Put anything readers must see inside the landing page until this is fixed.

The navigation organiser saves as you go and tells you so, but the panel is explicit: changes appear after the next publish.

Publish again.

I set a collection and nothing changed

A collection becomes a navigation group, and the grouping shows up in Organise navigation rather than on the item itself. Open the organiser to see it.

Like every navigation change, it reaches readers on the next publish.

The AI researched the wrong thing

Read the Question on the research item. It records what the AI thought you asked, and it is usually the difference between your request and the answer.

Re-ask, naming the specifics: the candidates, the criteria, and the units. “Compare search providers” produces a survey; naming three providers and three criteria produces a comparison.

The plan’s topics are not the ones that matter

Stop it. The registered topics appear before the expensive work happens, and they are the cheapest signal that the AI misread the task.

Stop this run takes effect after the current request finishes rather than immediately, so expect the running step to complete.

Research is thin or badly sourced

Set the workspace AI instructions rather than correcting each reply. Instructions apply to every future chat; a correction applies to one.

The three that reliably help: constrain source quality, force a comparable unit, and require unsourced claims to be flagged.

The summary sounds confident but I cannot verify it

Read the sources rather than the summary, and open the Sources used disclosure on the step that produced it.

A finding backed by one blog post reads exactly like one backed by three vendor documents. Only the sources tell you which you have.

The AI says there are evidence gaps

That is correct behaviour. A declared gap is more useful than a smoothed-over answer.

Treat a gap on a criterion you care about as a task to pick up, not a caveat to note and move past.

I asked it to create something and it linked to an existing item instead

By design. Asked for a scratchpad of open questions when one already existed, the AI said so and linked to it instead of creating a duplicate.

If you genuinely want a second item, say so in the request.

My edit was overwritten, or a change is waiting

Auto-apply creates new content automatically but stages edits to existing content until you approve them.

If changes seem to be pending, look for Approve and Dismiss on the step in the AI panel. Approval becomes available after the full run completes, so a change proposed early in a long plan waits for the plan to finish.

If you would rather review everything including new content, switch Apply AI suggestions automatically off.

The plan is still running after I left the page

By design — plans keep running when you navigate away, so a long research run does not need you to sit and watch it.

Come back to the workspace to see the result, or use Stop this run if it is no longer wanted.

I can only add one reference at a time

That is how the asset picker works: pick a card, confirm with Use asset, repeat. Load More pages through a long list.

An access level I want is greyed out

The share dialog says why: some options are limited by the service’s maximum access setting. That is a service-level decision, not a workspace one.

Anonymous visitors are a separate case — they cannot use internal access at any level and need an External Publisher URL.

Someone opened published content they should not have

Check whether a share link is active. Anyone holding an active link can open published content regardless of its access level, which is what share links are for, and why they take an expiry.

Something I said in the AI chat is visible to the team

Transcripts are visible to workspace members, and the create dialog says so.

That is deliberate, so the reasoning stays with the decision. Treat the chat as a shared document.

Content has no evidence behind it

Content written by hand has no provenance, and no Generated from N sources badge. Create from research is the path that carries evidence across from a research item.

For a short note that is fine. For anything that will be challenged, write it from research instead.

I cannot find research from another workspace

Evidence belongs to the workspace that gathered it, and there is no cross-workspace research view.

The route between workspaces runs through the Asset Library: research items are registered as assets, so they appear in another workspace’s asset picker under Workspace Research.