Troubleshooting

Find your symptom in the headings below. Each entry quotes the message you see on screen, says what caused it, and tells you what to do.

Getting in

"New signups are paused"

The sign-up form has been replaced by a panel reading "New signups are paused" and "We have temporarily paused new account registration. Please check back shortly - existing accounts are unaffected."

This is an operator switch, not something your newsroom can change. Existing accounts sign in normally through the "Sign in to an existing account" link on the same panel. Try again later, or ask the person who invited you.

"Sign-in is temporarily unavailable"

You see a page headed "Sign-in is temporarily unavailable" with "We could not render the authentication surface. Retry, or come back shortly." and buttons "Try again" and "Go to the homepage".

Use "Try again" first. If the sign-up page shows this but sign-in still works, the outage is on the ErmisAI side and is being seen by everyone — sign-in does not depend on the same data. If it persists, contact us at ermisai.com/contact and include the "Reference:" line if one is shown.

Every screen sends me back to newsroom setup

You finished the wizard, but every part of the app bounces you back to Set up your newsroom.

Setup counts as complete only when all of these exist: a publication name, a desk or team name, a section under Coverage, at least three selected sources, and — the one people miss — "Your title" on the Identity step. Without a title, the save is refused with "Add your title before finishing onboarding" and the app keeps returning you to the wizard.

Open Identity, type a title (any text up to 80 characters; the suggestions are just suggestions), and press "Go to press". See Set up your newsroom.

I was invited to a newsroom, but my setup changes did nothing

Invited members walk the same four-step wizard as the founder, but on a newsroom that is already configured, only two things you enter are actually kept: "Your title" and the fact that you finished. Publication name, desk name, coverage section, and source selections are discarded server-side, and nothing on screen tells you so.

That is by design — those are workspace-wide settings owned by the newsroom owner and admins. Ask an owner or admin to change them in Settings. See Team, roles and permissions.

My newsroom is empty after I changed workspace type

If you picked "A desk of one" while a team workspace was active, finishing onboarding switched your session out of the team and into your personal workspace. Your stories, sources, billing, and drafts all belong to the team workspace, not to you — they still exist, but you are no longer looking at them.

Use the workspace switcher to return to the team workspace. Nothing was deleted. There is no way to move content between a personal workspace and a team one.

The feed

"No stories match your current filters." — but I have no filters set

This is the only empty-state text the feed has. It appears whether you have five filters applied or none at all, including on a brand-new newsroom that simply has no stories yet.

Check the filter bar for the "{n} active" badge. If it is not there, no filters are set and the newsroom genuinely has nothing to show yet — give ingestion time, or press "Refresh sources".

The feed says it is warming up and stays that way

A muted strip with a rotating line — "Scanning the wires…", "Sifting through the sources…" and similar — plus "Fetching, clustering, and writing the newsroom articles can take up to a minute. Please be patient - and stay hydrated."

The first build for a new newsroom fetches every selected source, groups the articles, and writes the synthesised stories in one pass. It is slow exactly once. Wait; the page re-checks itself automatically. If it is still empty several minutes later, confirm on Choosing and adding sources that your sources are actually returning items.

"Stories feed unavailable" or "Failed to load stories"

A red strip above the list.

"Stories feed unavailable" means the first build ran past its time limit — roughly a minute and a half — and was cut off. It is almost always a cold newsroom with many sources. Press "Refresh sources" and try again.

"Failed to load stories" is the generic fallback when the feed could not be read at all. Reload the page. If it repeats for more than a few minutes, contact us.

Search returns nothing for a headline I can see on the card

The headline printed on a feed card is the source headline as it arrived from the outlet. Search matches the article headline — the one your newsroom wrote during compose. On any story that has been composed, those two are usually different, so searching for text you can see returns nothing.

Search for words you expect in your own article, or clear the search box and use the category, pipeline, or editorial filters instead.

Stories vanished from the list and "Queued" went up

Your newsroom hit its daily delivery cap. Approved stories beyond the cap are removed from the list entirely — not greyed out, not at the bottom — and counted in the "Queued" metric card, which turns amber. An amber banner also appears above the filters reading "{n} queued for delivery (daily cap resets at midnight UTC)" with a "Manage subscription" link.

The cap resets at midnight UTC, not local midnight. Stories still in progress are never capped — only approved ones. Caps by plan are in Plan and quota reference.

Opening a story shows "Story unavailable" and "Story is currently queued."

Same cause as above: that specific story is over today's delivery cap. It is not lost. It becomes readable again after the cap resets at midnight UTC, or immediately if the newsroom moves to a plan with a higher cap.

I cleaned the feed and the stories came back

"Clean feed" hides drafts behind a watermark rather than deleting them. Pressing "Refresh sources" — including the automatic refresh that runs right after cleaning — ignores that watermark and re-shows everything. The next ordinary page load hides them again.

Two other things to know before using it: "Clean feed" is tenant-wide and clears every member's personal queue, not just yours, and it does not give back any of the day's delivery count. The server allows both owners and admins to use it, but the button itself is only shown to the workspace owner and to anyone in a personal workspace — an admin in a team newsroom does not see it, even though the action would be permitted.

Filtering by "Collecting" or "Clustering" always returns nothing

Those two pipeline phases exist as filter options but effectively never appear on a real card. An uncomposed story reads "Ready to compose"; once composed and submitted it reads "Review", and once approved, "Published". Filter by those instead. See Pipeline phases and editorial states.

"Editor personal queue is not available."

The personal queue is a feature that operators can switch off. When it is off, the toast appears once and the queue buttons disappear for the rest of your session. There is nothing to fix on your side.

Writing a story

The draft is placeholder text — "Draft in composition"

Every story arrives as a placeholder. The card reads "Needs first article in {language}" and "The source story is available, but no article in {language} has been saved yet.", and the pipeline badge reads "Ready to compose".

Nothing composes it for you and there is no background job that will fill it in. Open the story, switch "Workspace mode" to "Compose", and press "Generate article" — or use the "Compose the first article" button on the empty article panel. See Your first story, end to end.

Chat refuses to answer until I compose

The refinement chat is blocked on a placeholder draft with "Generate the first article draft before using story chat." Only the compose action can produce the first article — chat cannot, no matter how you phrase the request.

Compose once, then chat works normally.

"Regeneration is locked because the saved article has already entered editorial workflow."

You can only regenerate an article while it is still in Draft state. Once it has been submitted for review, approved, or rejected and re-submitted, the regenerate button is locked.

To regenerate, get it back to Draft: a reviewer presses "Send back to compose" on the review desk. A rejected article is editable again but cannot be regenerated from scratch — use chat and "Replace article" instead.

"Someone else updated this draft. The latest version was loaded - review and retry."

Two people saved the same story at once, or you had the story open while a teammate added a note. The screen has already reloaded with the newer version; your unsaved edits were not applied.

Re-apply your change to the version now on screen and save again. In the story workspace the same situation reads "This story was updated by someone else. Refresh and try again."

"Generate the first article draft before submitting for review."

You pressed "Submit for review" on a story that is still a placeholder. Compose first, then submit.

"Draft cannot be updated in its current state"

You are trying to change the headline or body of a story that is in review or already published.

• In review — the text is locked on the author side. Notes still save. Either wait for the decision, or have a reviewer edit it directly on the review desk, or have them press "Send back to compose".
• Published — everything is locked. Published is final.

Rejected stories stay fully editable so you can revise and resubmit.

Appending a long assistant reply fails with "Invalid draft payload"

The article body has a hard ceiling of 25,000 characters, and nothing warns you before you hit it. Appending one more assistant reply to a long article can push it over, and the save fails with that bare message.

Trim the article, or use "Replace article" instead of "Append to article".

"Members can only compose drafts they own."

You have the member role and the story already belongs to someone else. Members can work on drafts they own and on drafts nobody has claimed — the first member to touch an unclaimed draft becomes its owner.

There are two real paths. An owner or admin can compose the story themselves; or, once the story is in review, they can change the Owner select on the Editorial review screen ("Editorial routing") and press Save. In the story workspace the Owner select sits under Editorial handoff with no heading, and only takes effect when the story is submitted for review — so it is not a way to reassign an uncomposed draft.

The article came out in the wrong language

The article language is fixed onto a story the moment it is first composed, and chat and quick completion follow that fixed language for the rest of the story's life. Changing "Article language" in settings afterwards changes nothing about stories that already exist.

There is no per-story language switch. Change the newsroom setting and compose a new story. See Interface language and article language.

The quick completion result ignores my saved article

"Quick completion" is grounded in the source story, not in the article you have written and saved. The on-screen hint — "One instruction, one source-grounded continuation for the saved article." — reads as though it continues your text; it does not.

For changes that must build on what you have written, use the chat panel and then "Append to article" or "Replace article" on the reply.

When AI stops responding

All of these appear as a toast or an inline panel with the message text below. They are English-only regardless of interface language.

"AI generation is paused for this workspace"

Shown as a banner across the top of the story workspace, with "The subscription is not in good standing, so compose and refinement requests are blocked. Settle the open invoice to restore them immediately." and a link "Open billing settings".

Your subscription is past due or cancelled. While it is, the whole newsroom also drops to Free-plan limits everywhere — sources, monitoring rules, and the daily delivery cap. Settle the invoice in the billing portal; access returns immediately. Only the owner can open Billing.

Capacity and limit messages

Running out of capacity blocks compose, chat, and quick completion. It does not stop new stories arriving — the story pipeline is not charged to your capacity, so the feed keeps filling even while you cannot write. See Billing and AI capacity.

Article generation failed with a provider message

Messages such as "Article generation is temporarily rate limited by the AI provider. Please retry shortly.", "Article generation timed out while waiting for the AI provider. Please retry.", or "Article generation failed because the AI provider was temporarily unavailable. Please retry." come from the model provider, not from your plan.

Retry. If several attempts in a row fail, it is an upstream outage — wait a few minutes.

"Article generation was blocked by the AI safety system for this story context." means the model refused this particular material. Retrying identically will not help; the story needs to be written by hand.

Review and approval

I cannot open Editorial

The sidebar item is missing and the page shows "Editorial board access is restricted" with "Editorial review is limited to the editor-in-chief and managing editor."

The review board is open to the owner and admin roles only. Members can write and submit but never review, no matter what title they are given — titles carry no permissions. Ask an owner to change your role in Settings → Team.

Approve or reject failed with no useful message

You get "Failed to approve story" or "Failed to reject story".

The usual cause is that the story is no longer in review — a colleague decided it while your tab was open, or you have the page open from a stale link. Go back to the board with "Back to board" and check: if it is gone from the list, it has already been decided.

"Send back to compose" threw away my edits

"Approve" and "Reject" silently save any pending headline, body, category, owner, or priority changes before recording the decision. "Send back to compose" does not — it discards them.

Press "Save" first if you want your edits kept, then send it back.

A story is stuck in review and is not on the board

The board only lists stories that are in review and have a real composed article. A story that reached review while still a placeholder will never appear.

Open it from the feed instead and compose it, then submit again.

There is no way to filter, sort, or claim a review item

Correct — there is not. The board sorts urgent, then high, then standard, then by most recently updated, and offers only "Refresh" and the "Open review" link on each row. There is no multi-select, no bulk approve, no assignment lock, and no deadline or escalation. Two people can open the same item; only the first decision counts, and the second gets a failure toast.

Nothing prevents you approving your own article. A single-person newsroom sees the note "You're the only active member of this newsroom, so you review and approve your own drafts."

Alerts and notifications

"Monitoring rules are not available on the current plan."

The Rules tab and its "Save rule" button are fully rendered on every plan, but the Free plan is allowed zero alert rules — the refusal only appears as a toast after you press save.

Move to a paid plan to get rules. The allowance by plan is in Plan and quota reference.

"Monitoring rule limits are unavailable right now."

A temporary failure reading your plan entitlements. Retry in a minute. If it persists, contact us.

My rules stopped firing and nothing changed

Two causes, both silent:

• Your subscription went past due or was cancelled. The newsroom reverts to Free limits, which means zero alert rules, so evaluation stops entirely. Your rules are still stored and start working again once billing is settled.
• You downgraded to a plan with fewer rules. All your rules stay stored, but only the newest few active ones are ever evaluated — the rest are silently ignored. Nothing on screen marks which. Delete the rules you no longer want so the ones you care about are inside the allowance.

Alerts stopped after I added a monitored source

This is the single most common alert surprise. With no monitored sources, every story is eligible for every rule. The moment you add one monitored source, only stories from monitored sources can fire — and the product adds sources with "Breaking: On · All alerts: Off", so only breaking stories from those sources qualify.

There is no control anywhere in the product to turn on all-alerts for a monitored source. If you want broad coverage, remove monitored sources entirely and let the rules do the filtering. See Alert rules and notifications.

A rule matched a story it should not have

Keywords are matched as plain substrings, case-insensitively, anywhere in the story's headline, body, summary, angles, or noted conflicts. They are not matched as whole words — a keyword of art matches Sparta.

Use longer, more distinctive keywords.

A rule missed a story it should have caught

Check, in this order:

1. Was the story approved or high-signal? Only stories the pipeline marked published — a signal score of 85 or above and not flagged sensitive — or stories an editor approved are ever tested against rules.
2. Cooldown. Cooldown is per rule and per channel, not per story. One match silences the whole rule for the cooldown window, so a second matching story minutes later never fires.
3. Minimum sources. A rule with a minimum of 2 skips every single-source story.
4. Monitored sources, as described above.

The Alert history tab is blank

Alert history has no empty state. When nothing has fired you see only the heading and white space. That is what "no alerts yet" looks like — it is not a loading failure.

The only alert mode is "Instant"

Correct. The "Mode" dropdown has one usable option, and the field carries the note "Email digest delivery is not configured in this runtime." Email alert delivery does not exist. Rules deliver in-app, and to your webhook if one is configured.

Old rules created under a previous mode show an amber line: "This legacy rule depends on email digest delivery, which is not configured." Delete and recreate them.

I cannot edit or disable a rule

There is no edit form and no on/off switch. A rule card offers "Delete" only. To change a rule, delete it and create a replacement.

A monitored source shows "Last fetch failed" or "Feed returned no items"

"Last fetch failed" means the outlet's feed did not respond on the last build; a reason is appended when one is available. "Feed returned no items" means it responded with nothing. "Fetching OK · {n} articles in 24h" is healthy.

An occasional failure is normal and self-heals on the next build. A source that stays failed is usually a feed URL that has moved — remove it and add the current one. See Choosing and adding sources.

There are no notification settings

There is no mute, no per-type toggle, no digest schedule, and no way to send notifications anywhere but the in-app inbox. The inbox is at Settings → Notifications and via the bell in the header. Everyone in the newsroom can read it.

The inbox is shared per newsroom — anything that fires is visible to every member — but the read/unread state is yours alone. Marking something read does not mark it read for a colleague.

Delivery and integrations

The Integrations section is missing from Settings

Three separate conditions all have to hold:

• You are the workspace owner. Admins have no access at all — despite the in-product WordPress screen saying "Only owners and admins can connect WordPress sites.", which is wrong.
• Your plan includes managed integrations, which the Free plan does not.
• Your billing data was readable when the page loaded.

If billing could not be read, a warning appears instead: "Billing access check failed" with "Billing data is temporarily unavailable. Team and integration entitlements could not be resolved." Reload once the billing store recovers.

Refresh in Integrations fails with "API keys require Pro or Enterprise"

You are on Plus. The API Keys card loads on the initial page render, so it looks available — but pressing "Refresh" fails the whole panel, and the plan-gate message is surfaced as a workspace-level error. Creating or revoking a key fails the same way.

API keys need Pro, Business Pro, or Enterprise. On Plus, ignore the card; webhook delivery is the integration you have.

I have to re-type the signing secret every time I save the webhook

Yes. Saving the webhook configuration requires all three fields — endpoint, secret, and the enable switch — and the secret box is cleared after every save. Leaving it blank blocks the save with "Secret is required".

Keep the secret somewhere you can paste it from. ErmisAI never generates or displays it; you supply it and your receiving system must use the same value.

The webhook config saved fine but every delivery fails

The saved configuration is not tested against the network. The endpoint is only fully checked at the moment a delivery is attempted, so a configuration can save cleanly and still fail on every send.

Read the code in "Recent webhook deliveries":

A 400 with a valid-looking public URL usually means split-horizon DNS or a load balancer whose addresses rotate — both are rejected by design.

A failed webhook was never retried

There is no retry queue, no backoff, and no replay. A delivery is attempted once.

Failed alert deliveries are re-attempted opportunistically the next time alert rules are evaluated, so they often recover on their own. Test webhooks sent with "Send test webhook" are never retried — press it again.

Only two kinds of webhook exist: the test, and a triggered alert. Approved stories are not pushed to your webhook.

"WordPress delivery is unavailable" or "WordPress sync is not enabled on this plan"

WordPress needs Pro, Business Pro, or Enterprise, and the check runs against your plan and your billing standing. A past-due Pro newsroom still sees the WordPress card but every WordPress request is refused.

I revoked an API key and a WordPress site went offline

Revoking a key deletes the WordPress connection bound to it, and disconnecting a site revokes its key. Nothing warns you before either. Reconnecting the same site also silently revokes the previous key.

Reconnect from WordPress — the handshake always starts there, in the connector's own settings screen, never from ErmisAI. See API keys, webhooks and WordPress.

A CMS integration is eating our daily delivery allowance

An external system pulling approved stories draws from the same daily delivery cap as the in-app feed. A polling loop that runs every few minutes will consume the newsroom's quota on its own, and stories that are already queued over the cap are invisible to the pull as well.

Poll less often, or move to a plan with a higher cap.

The "EMAIL" tile in Usage is always zero

Email delivery is not implemented. The tile exists in the breakdown and will always read zero. See Getting approved stories out.

Team and billing

"Team management requires a Business or Enterprise plan."

Anyone can create a team workspace during onboarding, but only Business Plus, Business Pro, and Enterprise can put anyone in it. Inviting a colleague on a lower plan is refused with that message.

"Seat limit reached for the {plan} plan. Remove an active or invited member before sending another invitation."

Pending invitations count against seats, not just accepted members. Revoke an unaccepted invitation, or remove a member, or buy more seats using the seat spinner on the plan card in Billing.

The Team section is missing from Settings

Team requires all three: the owner or admin role, an actual team workspace (a personal workspace has no team), and a Business or Enterprise plan. A personal workspace can never show it.

The owner left the newsroom

Nothing to do. When an organisation is left without an owner, the longest-serving admin is promoted automatically and notified — or, if the newsroom has no admins, the longest-serving member.

You cannot change your own role or remove yourself — the attempt fails with "Change your own workspace access from another owner or admin account." To hand over, promote someone else to owner first.

I cannot cancel or change my payment method

Cancellation, payment methods, and downgrades all live in the hosted billing portal. Open Settings → Billing and press "Billing portal". There is no in-app cancel button, and only the owner can open Billing at all.

There is no trial in ErmisAI. The Free plan is permanent and free; there is no countdown and nothing expires.

An invoice that failed is shown in green

A known display bug: the status pill in "Invoice history" is styled the same colour for every state. Read the status text — "Paid", "Open", or "Failed" — not the colour.

"Billing data is currently unavailable."

The Billing section could not load. Press "Refresh". If it persists, contact us — do not assume your subscription changed.

Still stuck

Contact us at ermisai.com/contact. It helps to include:

• The exact message text, copied rather than described.
• The screen you were on and what you pressed.
• Your newsroom name and roughly when it happened.

Account deletion is handled from Settings → Account. See Your account and your data.