Detect AI Content

The "Detect AI Content" step allows you to analyze text to determine whether it was likely written by an AI or a human. This powerful tool uses Originality.ai's advanced AI detection technology to help you identify AI-generated content, ensuring content authenticity and compliance with content policies.

How to Configure the Detect AI Content Step

Configuring the step requires setting the parameters shown below:

[IMAGE: Screenshot of the Detect AI Content step configuration panel]

Text

Enter the text content you want to analyze for AI detection. This can be a snippet, a paragraph, or a full article.

You can use Liquid syntax to dynamically reference text from a previous step:

{{ step_1.output.content_to_analyze }}

Minimum Length (Optional)

The minimum text length required for reliable AI detection. If your text is shorter than this value, the step will return a warning about potential reduced accuracy.

Default: 100 characters

Output Format

The Detect AI Content step returns a JSON object containing detailed analysis results:

{
  "original_text": "The first few sentences of the analyzed text...",
  "ai_score": 0.92,
  "is_ai": true,
  "human_score": 0.08,
  "analysis": {
    "ai_probability": 92,
    "model_confidence": "high",
    "likely_models": [
      "GPT-4",
      "Claude",
      "Bard"
    ],
    "content_classification": {
      "original": 0.08,
      "ai_generated": 0.92
    }
  },
  "word_count": 325,
  "character_count": 1842,
  "language": "en"
}

[IMAGE: Screenshot showing an example of the Detect AI Content results in the AirOps interface]

Understanding the Output Metrics

• original_text: The first portion of the analyzed text (truncated for display purposes)

• ai_score: A score from 0 to 1 indicating the likelihood that the content was AI-generated (higher = more likely AI-generated)

• is_ai: A boolean value indicating if the system believes the content was AI-generated (true/false)

• human_score: A score from 0 to 1 indicating the likelihood that the content was human-written (higher = more likely human-written)

• analysis: Detailed information about the detection process

  • ai_probability: AI score expressed as a percentage

  • model_confidence: Level of confidence in the detection result

  • likely_models: Potential AI models that may have generated the content

  • content_classification: Breakdown of original vs. AI-generated probability

• word_count: Total number of words in the analyzed text

• character_count: Total number of characters in the analyzed text

• language: Detected language of the content

How to Use the Results

You can access the AI detection results in subsequent workflow steps using Liquid syntax. For example:

To access the AI score:

{{ step_2.output.ai_score }}

To check if content is likely AI-generated:

{{ step_2.output.is_ai }}

To use the results in a conditional statement:

{% if step_2.output.ai_score > 0.8 %}
  This content appears to be AI-generated and may need human review.
{% else %}
  This content appears to be human-written or well-edited AI content.
{% endif %}

Common Use Cases

• Verifying the originality of submitted content

• Ensuring compliance with content guidelines requiring disclosure of AI usage

• Filtering AI-generated content in user submissions

• Quality control for content creation workflows

• Validating content authenticity before publication

• Identifying potentially plagiarized content from AI tools

• Assessing content for search engine optimization concerns

How to Continue if the Detection Fails

By default, the Detect AI Content step will terminate the workflow if it fails. To continue the workflow if the step fails, click "Continue" at the bottom of the settings panel.

[IMAGE: Screenshot showing the "Continue" option in the step settings]

The step will return the following keys when it fails:

• output: this will be null

• error:

  • message: the message returned from the step

  • code: the error code representing the error

Common error causes include:

• Text too short for reliable analysis

• Unsupported language

• Service unavailability

Example Workflow

Here's a common pattern using the Detect AI Content step:

1. Start with content submitted for publication or review

2. Use the "Detect AI Content" step to analyze the content for AI generation

3. Use a Conditional step to branch the workflow based on the AI score:

   • If AI score is high (>0.8): Route to human review

   • If AI score is moderate (0.5-0.8): Add AI disclosure notice

   • If AI score is low (<0.5): Proceed with normal publication

4. Use an LLM step to provide customized feedback or add necessary disclosures

5. Output the content with appropriate handling based on the detection results

This workflow enables you to implement a content authenticity policy that accounts for AI-generated content, ensuring transparency and compliance with emerging content standards while maintaining quality control.