Spam Detection

Detect spam comments in your discussions.

Every comment gets one of three outcomes:

Outcome	Response	What to do
Spam	is_spam is true	Delete it.
Suspected spam	is_spam is false, suspected_spam is true, suggestion is filled in	Send it back to the commenter with the suggestion, as you would a comment that needs revision. A real person can fix it and post again; a spammer won't bother.
Not spam	is_spam and suspected_spam are both false	Publish it as normal.

What it detects:

• Promotional/advertising content
• Bot-generated text patterns
• Link spam and SEO manipulation
• Repeated/duplicate comments
• Off-topic noise

Reference

See the Spam Detection API Reference for method signatures and parameters.

Response: SpamDetectionResult

By Example

Basic Usage

TypeScript

// With article context (more accurate)
const result = await client.checkSpam(
  "Buy cheap watches at example.com!",
  articleId
);

// Without context (standalone check)
const result2 = await client.checkSpam("Buy cheap watches at example.com!");

console.log(`Is spam: ${result.is_spam}`);
console.log(`Confidence: ${result.confidence.toFixed(2)}`);
console.log(`Reasoning: ${result.reasoning}`);

if (result.is_spam) {
  // Delete it
} else if (result.suspected_spam) {
  // Send it back to the commenter with result.suggestion
} else {
  // Publish it
}

Python

Blocking Client

# With article context (more accurate)
result = client.check_spam(
    "Buy cheap watches at example.com!",
    article_id
)

# Without context (standalone check)
result = client.check_spam("Buy cheap watches at example.com!")

print(f"Is spam: {result.is_spam}")
print(f"Confidence: {result.confidence:.2f}")
print(f"Reasoning: {result.reasoning}")

if result.is_spam:
    ...  # Delete it
elif result.suspected_spam:
    ...  # Send it back to the commenter with result.suggestion
else:
    ...  # Publish it

Async Client

# With article context (more accurate)
result = await client.check_spam(
    "Buy cheap watches at example.com!",
    article_id
)

# Without context (standalone check)
result = await client.check_spam("Buy cheap watches at example.com!")

print(f"Is spam: {result.is_spam}")
print(f"Confidence: {result.confidence:.2f}")
print(f"Reasoning: {result.reasoning}")

if result.is_spam:
    ...  # Delete it
elif result.suspected_spam:
    ...  # Send it back to the commenter with result.suggestion
else:
    ...  # Publish it

PHP

$client->checkSpam("Buy cheap watches at example.com!", $articleId)
    ->then(function ($result) {
        echo "Is spam: " . ($result->isSpam ? 'Yes' : 'No') . "\n";
        echo "Confidence: " . number_format($result->confidence, 2) . "\n";
        echo "Reasoning: " . $result->reasoning . "\n";

        if ($result->isSpam) {
            // Delete it
        } elseif ($result->suspectedSpam) {
            // Send it back to the commenter with $result->suggestion
        } else {
            // Publish it
        }
    });

$client->run();

REST API

Try it out live →

# With article context (more accurate)
curl -X POST https://app.respectify.ai/v0.2/antispam \
  -H "X-User-Email: your@email.com" \
  -H "X-API-Key: your-api-key" \
  -d "comment=Buy cheap watches at example.com!" \
  -d "article_context_id=550e8400-e29b-41d4-a716-446655440000"

# Without context (standalone check)
curl -X POST https://app.respectify.ai/v0.2/antispam \
  -H "X-User-Email: your@email.com" \
  -H "X-API-Key: your-api-key" \
  -d "comment=Buy cheap watches at example.com!"

Response for spam:

{
  "is_spam": true,
  "confidence": 0.95,
  "reasoning": "Classified as spam by first-pass screening.",
  "suspected_spam": false,
  "suggestion": ""
}

Response for suspected spam, here a comment with a link:

{
  "is_spam": false,
  "confidence": 0.3,
  "reasoning": "",
  "suspected_spam": true,
  "suggestion": "This looks like spam. How about removing the link?"
}

Response for a comment that is not spam:

{
  "is_spam": false,
  "confidence": 0.97,
  "reasoning": "",
  "suspected_spam": false,
  "suggestion": ""
}

First-Pass Screening

Spam is decided by a fast first-pass screen alone. No comment is sent for a fuller assessment of spam: spam needs a verdict, not an explanation. The screen gives each comment a probability of being spam, and that decides the outcome:

• 0.85 or above, spam: is_spam is true, confidence is the screen's probability, and reasoning is "Classified as spam by first-pass screening." The cutoff is set high so that real comments are not deleted.
• 0.50 up to 0.85, suspected spam: is_spam is false, suspected_spam is true, and suggestion holds a short message for the commenter. confidence is the confidence that it is not spam, and reasoning is empty.
• Below 0.50, not spam: is_spam and suspected_spam are false, confidence is the confidence that it is not spam, and reasoning and suggestion are empty, since there is nothing to explain.

The suggestion for a suspected-spam comment is one of two:

• If the comment contains a link: "This looks like spam. How about removing the link?"
• Otherwise: "This reads like promotion. How about saying what you think about the discussion?"

If screening is unavailable, the comment is let through rather than deleted without a verdict: is_spam is false, confidence is 0, and reasoning is "Not checked for spam: screening was unavailable." A check that was not screened is not charged.

In a megacall, a comment answered as spam ends the request: every other check is returned as null and only the spam check is charged. A suspected-spam comment is not spam, so the other checks still run and are returned as usual.

Best Practices

• Use article context when available for better accuracy.
• Act on the outcome, not the confidence. Respectify has already applied the thresholds: delete when is_spam is true, send the comment back when suspected_spam is true, and publish otherwise.
• Show the suggestion to the writer of a suspected-spam comment, alongside any other feedback for the comment (for example from comment health), so they can fix it and post again.
• Log false positives to understand patterns.