Social Media Post Verification

AI content and image analysis combined with crowd context review to verify social media posts and detect coordinated inauthentic behaviour.

The Social Media Post Verification pipeline processes posts from any platform — text, images, or both — and evaluates them for authenticity signals that single-model analysis frequently misses. AI stages assess the post's textual content for machine-generation patterns, claim consistency, and unusual formatting characteristics. If an image attachment is included, a separate image integrity check runs automatically. Crowd workers then review the full picture — post text, image (if present), and AI findings — to apply contextual human judgment that is particularly effective at identifying impersonation, coordinated inauthentic behaviour, and context manipulation (the practice of recirculating genuine content with a false framing). Suitable for trust & safety teams, newsrooms with rapid-turnaround verification needs, and DSA compliance workflows requiring documented human review.

Pipeline Details

Property	Value
Pipeline ID	`verify-social-post`
Tier	2
Estimated Duration	~40 min
Credit Cost	2,500 credits per run
Crowd Stages	Yes
Files Required	No — optional image attachment

Required Inputs

Files

A file upload is optional. If the post includes an image, submit it as a file alongside the text context keys for the most complete analysis. If no image is provided, the Image Analysis stage is skipped automatically and the pipeline proceeds with text and crowd review only. Accepted image formats: JPEG, PNG, WebP, HEIC, AVIF.

Context Keys

Key	Required	Description
`post_url`	Required if no file	URL of the social media post to verify
`post_text`	Required if no URL and no file	Full text of the post as a plain string
`platform`	Optional	Social media platform the post originates from (e.g. `"X"`, `"Facebook"`, `"Telegram"`, `"Instagram"`)
`claimed_date`	Optional	Date the post is claimed to have been published, ISO 8601 format
`claimed_author`	Optional	Username or identity of the post author (e.g. `"@newsaccount"`)

Providing platform and claimed_author allows the AI to calibrate its assessment against platform-specific behavioural norms and any known information about the claimed account. These keys are also shown to crowd workers who may be familiar with the account or platform context.

Stages

Content Analysis (AI)

The pipeline opens with an analysis of the post's textual content. The AI evaluates linguistic patterns associated with machine-generated text, assesses internal claim consistency, checks for unusual formatting (excessive capitalisation, irregular punctuation patterns, templated phrasing common in coordinated campaigns), and evaluates whether the claims made in the post are consistent with verifiable facts in its knowledge base. The output includes anomaly flags, a machine-generation probability score, and a structured summary of the post's primary assertions.

Image Analysis (AI, conditional)

If an image file was submitted alongside the post, this stage runs an integrity check on the image: EXIF anomaly detection, format-level manipulation signals, and a visual consistency assessment relative to the post's textual claims. If no image was provided, this stage is skipped automatically and the pipeline advances directly to crowd review. The conditional nature of this stage means the pipeline's overall duration is shorter when no image is included.

Crowd Context Review (Crowd)

Five crowd workers holding the general skill tag independently review the post content — including the image if present — alongside the AI findings. Workers are asked to evaluate the authenticity and plausibility of the post, whether the content appears consistent with the claimed platform and author, whether they recognise any coordinated inauthentic behaviour patterns, and whether the post's framing appears to accurately represent any events or sources it references. The stage gates on 60% consensus (three of five workers in agreement) before proceeding.

Verdict Synthesis (AI)

The AI integrates the text analysis findings, image integrity results (if present), and crowd worker assessments into a final weighted verdict. Divergent worker responses are noted explicitly. The output includes a final verdict (authentic, inauthentic, or inconclusive), a confidence score, a structured summary of the key evidence points, and a complete per-stage evidence log suitable for documentation in trust & safety or compliance workflows.

Starting a Run

curl -X POST "https://api.crowdee.ai/v2/projects/{projectId}/verification-runs" \
  -H "X-API-Key: crw_YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "pipeline_id": "verify-social-post",
    "file_ids": [],
    "context": {
      "post_url": "https://x.com/user/status/123456789",
      "platform": "X",
      "claimed_author": "@newsaccount"
    }
  }'

Example Response

{
  "id": "run_01j9x...",
  "status": "completed",
  "verdict": "authentic",
  "confidence": 81,
  "pipeline_id": "verify-social-post"
}

Pass the post image as a file upload alongside the text context keys for the most thorough analysis. If no image is available, the Image Analysis stage is skipped automatically and the pipeline completes with text and crowd review only. The estimated duration of ~40 minutes assumes a crowd stage — the AI-only path (no crowd) is not available for this pipeline.