Labels

Comment and post lists used to carry model judgments: labels, relevance, label_share and labels_evidence, with parameters to tune them. Schema 2, which every key gets from October 4, 2026, removed them on every endpoint. Responses now carry the platform's data only. What changed is on Schema 2.

The parameters now return 400

These parameters asked for analysis. Under schema 2 they answer 400 UNSUPPORTED_PARAMETER, with error.param naming the parameter, and nothing is charged:

label, label_evidence, relevance, relevance_threshold, relevant_to, judgments, fit, goal, fit_tokens, seen, brand, brand_description, offer, exclude (except on /v1/pinterest/trends, where it filters terms), brief, use, moments, and format=stationery.

Two more are refused on specific endpoints only, where they ask for analysis; elsewhere they are ordinary parameters:

{
  "success": false,
  "error": {
    "type": "UNSUPPORTED_PARAMETER",
    "message": "`relevance` is not supported: it asks for analysis, which is not part of schema 2. Remove it and retry. Nothing was charged.",
    "param": "relevance"
  },
  "request_id": "req_80d197121a1e",
  "credits_used": 0,
  "credits_remaining": null
}

Remove the parameter and send the call again.

Filtering results yourself

A search returns what the platform returns for your query. To keep only the rows about your topic, filter on the text in your own code. A keyword filter covers most cases:

KEYWORDS = ["insightsocial", "insight social"]

def mentions(text):
    t = (text or "").lower()
    return any(k in t for k in KEYWORDS)

posts = [item["post"] for item in body["data"]["items"]]
on_topic = [p for p in posts if mentions(p["content"]["text"])]

For tone, topic or paid-promotion checks, run your own classifier or model on post.content.text or comment.text. To keep one language, filter on post.language or comment.language, the platform's own tag.

A full monitoring example is in Brand mention monitoring.