Quality flags are automated warnings about potential issues in generated output. Smelt checks every output and flags problems for your review.
The 11 Quality Flags
Length Flags
| Flag | Description | Severity |
|---|
too_short | Output below 15 characters | ⚠️ Warning |
over_char_limit | Exceeds template’s character limit | ⚠️ Warning |
over_word_limit | Exceeds template’s word limit | ⚠️ Warning |
Content Flags
| Flag | Description | Severity |
|---|
forbidden_word | Contains a word from template’s forbidden list | 🚫 Error |
has_preamble | Starts with “Here’s a hook:”, “Sure!”, etc. | ℹ️ Info |
has_postamble | Ends with “Let me know!”, “Hope this helps!“ | ℹ️ Info |
has_quotes | Output wrapped in quotation marks | ℹ️ Info |
has_placeholder | Contains [Name], , or | 🚫 Error |
has_ai_speak | Contains “As an AI…”, “I cannot…” | 🚫 Error |
has_cliche | Contains “hope this finds you well”, “circle back” | ⚠️ Warning |
generic | Not personalized enough based on input | ⚠️ Warning |
Auto-Fix Feature
Smelt automatically fixes certain issues before you see them!
| Issue | Auto-Fix |
|---|
| Preambles | ”Here’s a hook:”, “Sure!”, “Certainly!” → Removed |
| Postambles | ”Let me know!”, “Hope this helps!” → Removed |
| Quote wrappers | Surrounding quotation marks → Removed |
Flags shown in results are issues that couldn’t be auto-fixed.
Flag Details
too_short
What it means: Output is under 15 characters.
Common causes:
- AI misunderstood the prompt
- Input data was insufficient
- Template constraints too restrictive
How to fix:
- Edit inline to expand
- Re-run with revised prompt
- Check input data quality
over_char_limit / over_word_limit
What it means: Output exceeded your template’s length constraints.
Common causes:
- AI didn’t fully respect limits
- Limits set too low for the task
How to fix:
- Edit to shorten
- Adjust template limits
- Make limit clearer in prompt (“MUST be under 100 characters”)
forbidden_word
What it means: Output contains a word you banned in the template.
Common causes:
- AI used the word despite instructions
- Word appears in a different form
How to fix:
- Edit to remove the word
- Make prohibition clearer in prompt
- Consider if the word is truly necessary to ban
Forbidden word detection uses word boundaries—“loan” won’t flag “loans” or “alone”.
has_preamble
What it means: Output starts with AI-style introductions.
Examples:
- “Here’s a hook for you:”
- “Sure, here’s what I came up with:”
- “Certainly!”
Why it’s flagged: These should have been auto-fixed. If you see this flag, the preamble pattern wasn’t recognized.
How to fix: Edit to remove the preamble manually.
has_postamble
What it means: Output ends with AI-style sign-offs.
Examples:
- “Let me know if you need changes!”
- “Hope this helps!”
- “Feel free to ask for more!”
Why it’s flagged: These should have been auto-fixed. If you see this flag, the postamble pattern wasn’t recognized.
How to fix: Edit to remove the postamble manually.
has_quotes
What it means: Output is wrapped in quotation marks.
Example:
"Scaling your SaaS team in Austin is no small feat—"
has_placeholder
What it means: Output contains unfilled placeholders.
Examples:
[Name]{Company}{{variable}}[INSERT CITY HERE]
Common causes:
- AI left placeholders instead of using data
- Variable wasn’t found in CSV
- AI misunderstood the task
How to fix:
- Check that CSV has the expected columns
- Edit to fill in the placeholder
- Re-run after fixing the prompt
This is a serious flag—placeholders will look terrible if sent to prospects!
has_ai_speak
What it means: Output contains language revealing it’s AI-generated.
Examples:
- “As an AI language model…”
- “I cannot provide…”
- “I don’t have access to…”
Common causes:
- AI broke character
- Prompt triggered safety responses
- Unusual input data
How to fix:
- Edit to remove AI language
- Revise prompt to prevent this
- Check input data for issues
has_cliche
What it means: Output contains overused phrases.
Examples:
- “I hope this email finds you well”
- “Circle back”
- “Touch base”
- “Low-hanging fruit”
- “Synergy”
Why it matters: These phrases are so common they reduce impact and feel generic.
How to fix:
- Edit to replace with more original phrasing
- Add clichés to forbidden words list
generic
What it means: Output doesn’t seem personalized to the specific lead.
Detection: Checks if the output could apply to almost any lead rather than being tailored.
Common causes:
- Not enough data in the CSV
- Prompt doesn’t reference enough variables
- Template is too general
How to fix:
- Use more variables in your prompt
- Add more specific instructions
- Ensure CSV has relevant data
Filtering by Quality
In the Results view, filter to show:
| Filter | Shows |
|---|
| All results | Everything |
| Has flags | Only outputs with quality issues |
| No flags | Only clean outputs |
Review flagged outputs first, then bulk-approve clean ones.
Quality Flag Strategy
Filter to flagged only
Focus on outputs that need attention
Review and edit
Fix issues inline or decide to re-run
Bulk approve clean outputs
Select all clean outputs and approve
Re-run problematic rows
If many rows have issues, revise template and re-run
