Section 1 — QA Settings

Open QA & Compliance configuration for your connection, then open QA Settings.

  • Peer review toggle (on/off): If on, self-review submission moves the run to peer review before final approval. If off, self-review can complete the run without a second reviewer.
  • AI Checks toggle (on/off): If on, AI-powered checks run normally. If off, all AI checks show as skipped.
  • Custom categories: Create organization-specific categories to organize checks in a way that matches your process.

Section 2 — The Check Catalog

The Check Catalog is a library of available checks for your platform. Each check includes a name, description, type, and default configuration.

  • Browse checks by category.
  • The catalog is platform-specific. Marketo checks do not appear for Eloqua connections.
  • The app catalog is the source of truth for available checks and descriptions.

The catalog shows all available checks with descriptions. Browse it in the app to see what is available for your platform.

[SCREENSHOT: Check catalog browsing view showing categories and check entries]

Section 3 — Enabling and Configuring Checks

  1. Open the Check Catalog.
  2. Select a check.
  3. Add it to the connection.
  4. Configure severity, required status, and parameters (if available).
  5. Save.

Severity

  • Error: A critical problem likely to cause a visible issue or compliance violation if sent.
  • Warning: A quality concern that should be reviewed but may not block sending.
  • Info: Informational awareness without implying a defect.

Required

When a check is marked required, the QA run cannot be approved if that check fails. Optional checks still run and show results, but failures do not block approval.

Example: You might configure unsubscribe presence as Error + Required (blocking), while subject line quality is Warning + Optional (non-blocking).

Parameters

Some checks accept parameters that control behavior. For example, a check might allow a maximum character count, a required value list, or a regex pattern. Parameters appear in the configuration panel when you select a check.

Disabling a check

You can turn off a check without deleting it. This pauses enforcement while preserving its configuration.

[SCREENSHOT: Check configuration panel showing severity, required, and parameters]

Section 4 — Multi-Instance Checks

Some checks can be added multiple times with different configurations. For example, you can add one naming convention rule for nurture emails and another for event emails.

  • Look for the allow multiple indicator in the catalog.
  • Each instance requires a unique name within the same check type.
  • Each instance runs independently and returns its own result.

[SCREENSHOT: Multiple instances of the same check type with different names]

Section 5 — Custom Manual Checks

Custom manual checks capture steps specific to your organization that cannot be automated.

  1. Add a custom manual check.
  2. Enter name and description.
  3. Write clear step-by-step instructions.
  4. Assign a category.
  5. Set severity and required.
  6. Save.

Limit: up to 50 custom manual checks per connection.

These appear as manual checks in QA runs. Reviewers follow the instructions externally and then mark completion with notes.

Section 6 — Custom AI Checks

Custom AI checks let admins define organization-specific quality or compliance rules in plain language. You choose the data, define criteria, and decide how pass/fail/warning should be interpreted.

Prerequisite: AI Checks must be enabled in QA Settings.
Limit: Up to 10 custom AI checks per connection.

Step-by-step authoring workflow

  1. Go to Check Catalog and find AI-Powered Check.
  2. Click Add to Connection.
  3. Enter a display name (for example, Brand Voice Compliance).
  4. Select variables from the dropdown:
    • Program Metadata: Program name, description, type, and tags
    • Email Content: Subject, preheader, sender details, HTML, links, unsubscribe status
    • Smart Lists: Filter rules and logic
    • Tokens: Program-level token values
  5. Write your prompt:
    • Use placeholders like {{variable_name}}.
    • At runtime, placeholders are replaced with real values from the snapshot.
    • Define what pass/fail/warning means.
    • Ask for evidence by requesting exact values in the explanation.
  6. Set severity and required.
  7. Click Test This Prompt, choose a recently scanned program, review the result, then refine and re-test.
  8. Save.

Prompt writing tips

  • Be specific: "Check for placeholder text like TBD, [INSERT], or Lorem ipsum" is better than "Check content quality."
  • Define decision criteria: Explain exactly what should pass, fail, or warn.
  • Ask for evidence: Require exact text, link, or values in the explanation.
  • Keep checks focused: One check should evaluate one objective.

Starter prompt templates

Brand Voice Compliance

Review the email subject line '{{email_subject}}' and body content for brand voice compliance. Check that the tone is professional yet approachable. Flag if the language is overly casual, uses slang, or contains aggressive sales language like 'Act now!', 'Limited time!', or 'Don't miss out!'. Reference the specific phrases you found.

Regulatory Language Check

Review the email HTML content for any language that could be interpreted as making a medical claim or health guarantee. Flag phrases like 'cure', 'guaranteed results', 'clinically proven' that appear without proper disclaimers. Also verify that opt-out language meets CAN-SPAM requirements. Reference the exact text you found.

Content-Language Match

The program tags indicate this email targets the {{program_tags}} market. Review the email subject '{{email_subject}}' and body content to verify all copy is written in the correct language for that market. Flag any content that appears to be in a different language or contains untranslated placeholder text.

CTA Quality

Review the email HTML content and check if the call-to-action is clear, prominent, and actionable. The CTA button text should be specific (e.g., 'Register Now', 'Download the Guide') rather than generic ('Click Here', 'Learn More'). Flag if the CTA is missing, vague, or buried below the fold.

Data privacy note

When an AI check runs, selected data (for example, subject lines, email HTML, and tokens) is sent to Anthropic's API for evaluation. Only selected variables are sent. Data is not stored by the AI provider.

[SCREENSHOT: AI-Powered Check entry in catalog]
[SCREENSHOT: AI check authoring form — name, prompt textarea, variable dropdown]
[SCREENSHOT: Variable dropdown expanded showing categories and descriptions]
[SCREENSHOT: Test Prompt result showing pass/fail with explanation]