What it does

authentic-sentiment takes a topic, brand, handle, or phrase and returns sentiment scored only from posts that pass two filters: the post does not read as AI-generated, and the author does not read as a bot. 

{
  "query": "agentutility.ai",
  "max_results": 15
}

returns a filtered read alongside the unfiltered one for comparison: 

{
  "authentic_sentiment": { "overall": "positive", "score": 0.42, "based_on_n_posts": 9 },
  "excluded": { "ai_generated": 2, "low_authenticity": 3, "too_short": 1 },
  "raw_sentiment_all_posts": { "overall": "positive", "score": 0.31, "based_on_n_posts": 15 },
  "top_authentic_posts": [
    { "text_snippet": "...", "handle": "some_real_user", "authenticity": 82, "sentiment": "positive" }
  ]
}

Why

mention-intel, x-kol-sentiment, and social-mindshare all aggregate sentiment across whatever a search turns up. That works until a topic gets astroturfed: a bot network or a batch of AI-generated posts can swing an aggregate sentiment score without a single real opinion behind it. authentic-sentiment is the composite built to catch that. It runs x-search for recent posts, then per post: ai-content-detector on the text and sentiment-analysis on the text, plus per distinct author (capped at 8): x-account-authenticity. Posts that score at or above 0.65 probability of being AI-generated, or whose author scores under 40 on authenticity, are excluded from the rollup before the sentiment average is computed. The response reports both numbers, so a buyer can see how much the filtering moved the needle.

Degradation

x-search is the required leg; if the post search fails the call fails. Per-post AI-detection, per-author authenticity, and per-post sentiment are all independent optional legs. A rate limit or LLM hiccup on any of them narrows what can be classified rather than failing the call: posts that couldn't be checked for AI-generation are marked too-short-to-assess and conservatively excluded from the authentic rollup, and posts whose author authenticity couldn't be scored are not excluded for that reason (only for what was actually verified).

Price: $0.06.