The failures that actually come up, grouped by where they happen.

Campaigns and sourcing

URL analysis fails when creating a campaign

The page must be publicly reachable. Login walls, paywalls, and heavy client-side rendering all mean the extractor sees something other than your content.

Fill the fields in manually and continue — nothing downstream depends on the auto-fill having run.

A scrape returns no leads

In order of likelihood:

The ICP is too narrow. An obscure language plus a specific city plus a niche topic can legitimately match nobody. Loosen one constraint at a time.
The ICP is too vague. The opposite failure: queries so broad that nothing scores above threshold.
The score threshold is too high. Leads may be sourcing correctly and failing the gate. Check whether any leads exist at sourced status.
Wrong source for the ICP. A design-focused ICP will find almost nothing on npm.

Leads have no email addresses

Expected for a portion of any list. Ozigi recovers addresses from public GitHub commit history where a profile hides them, but not every profile yields one.

Leads without an address can still be reached on LinkedIn if they were sourced from LinkedIn. Leads from GitHub or Dev.to generally have no LinkedIn URL attached.

Email sending

Emails are not sending

Is an inbox connected and active? See Email Setup.
Is the daily cap reached? Sends resume the next day from where they stopped.
Are any leads due? A step with a three-day delay has nobody due until day three.
Is the campaign paused?
Did leads clear the score threshold? Low-scoring leads are never contacted.

Authentication fails after months of working

App passwords are revoked when you change your account password or reconfigure 2FA. Generate a new one and reconnect.

For Gmail: delete the old entry in App Passwords and create a replacement.

SMTP connection test fails

Nearly always because you used your account password instead of an app-specific one. Outlook, Yahoo, and Zoho all require an app password once two-factor is enabled, and reject the account password outright.

Also worth checking: the port. 587 for STARTTLS, 465 for implicit TLS. Using the wrong one produces a timeout rather than a clear error.

Mail sends but lands in spam

A domain reputation problem, not a connection problem. In order:

SPF, DKIM, and DMARC on your sending domain. Missing records put you in spam regardless of everything else. These are configured with your DNS provider, not in Ozigi.
Sending history. A cold domain that starts at high volume gets classified as spam and stays that way. Start at twenty to thirty a day.
Bounce rate. A high rate signals a bad list and damages reputation across every future send.

Nothing sends on the LinkedIn channel

Extension installed and switched on? Check chrome://extensions.
LinkedIn tab open and logged in? No tab means the queue just waits.
Token current? If you rotated it, the extension still holds the old one until you paste the new one.
Do the leads have LinkedIn URLs? Only leads sourced from LinkedIn do.

Sends were working and stopped abruptly

Almost always the session or the tab, not the configuration. Check that you are still logged into LinkedIn and that a tab is open, then confirm the extension toggle is still on.

Connection requests send but nothing else happens

Expected. A message step only fires once the connection is accepted. Leads who have not accepted sit waiting at that step.

The exception is Open Profile leads, who can be messaged as first contact.

Content generation

Generated content sounds generic

This is a persona problem in almost every case.

Is a persona actually selected? Some surfaces default to none.
Is the prohibitions field empty? The most common cause. List ten phrases you would never use.
Does the persona describe a task rather than a person? If it contains the word "write", it is a prompt, not a persona.
Is it two sentences long? Too short to constrain anything.

See Personas.

A long-form draft is well written but wrong

Usually audience calibration. An article with no audience set is not aimed at anyone in particular.

Set the audience, regenerate, and read the calibration flags — they name the specific mismatch, such as undefined jargon in a beginner piece or a how-to with no code.

See Long-Form Articles.

A long-form draft came back with warnings attached

The lexicon repair pass was skipped because the request was close to its time budget. The draft is usable; the warnings point at the vocabulary that would have been repaired. Regenerating often produces a clean pass.

Code in a generated article does not run when pasted

Check the audit flags first — smart quotes and non-breaking spaces inside code blocks are caught and reported as errors precisely because they are invisible on the page and fail on paste.

If the audit flagged it, the fix is replacing the curly quotes with straight ones.

Integrations

CRM OAuth fails immediately

For Zoho CRM and Salesforce, you must create an OAuth 2.0 auth config in your Composio dashboard before the flow will work. HubSpot works without one.

Swipe One rejects the API key

The key is invalid or revoked. Ozigi tests keys against Swipe One before saving, so a rejection is genuine. Regenerate in Swipe One → Settings → API.

Discord or Slack posts are not appearing

Verify the webhook still exists on their side. Webhooks are deleted with their channel and can be revoked by any server admin.

LinkedIn publishing stopped working

The access token expired — they last about sixty days and cannot be refreshed silently. Reconnect LinkedIn in Settings. The post content is rarely the cause even when the error suggests it.

Leads are not syncing to the CRM

Sync fires on first contact, not at sourcing. A campaign that has scraped and scored but not sent will not have pushed anything yet.

Newsletters

The Schedule button does not complete

Two prerequisites: at least one active subscriber, and a sender name configured in Settings.

Check the From address before a large send, and set a Reply-To you actually monitor. Mail can send from a no-reply address depending on configuration, and replies to it are lost silently.

Still stuck

Email hello@ozigi.app with your account email and a description of what you were doing when it failed. Include the campaign or article name if it is specific to one.

Troubleshooting