{
  "schema_version": 1,
  "dataset_id": "chatgpt-medical-provider-shortlists-v1.0",
  "title": "ChatGPT Named Most Final-Shortlist Medical Providers in Its First Search Queries",
  "description": "A descriptive US benchmark of medical-provider names in the first returned web-search action and the final four-provider answer from one chat-latest Responses API configuration.",
  "version": "1.0",
  "date_published": "2026-08-18",
  "fieldwork": {
    "start": "2026-08-18T13:35:58Z",
    "end": "2026-08-18T14:10:01Z"
  },
  "spatial_coverage": [
    "New York, New York",
    "Chicago, Illinois",
    "Houston, Texas"
  ],
  "language": "en-US",
  "creator": {
    "name": "Evgeniy Yudin",
    "organization": "Rotgar Research"
  },
  "methodology_reviewer": {
    "name": "Boris Teplyakov",
    "role": "SEO Lead",
    "review_scope": "Methodology only; not a clinical subject-matter review or an assessment of providers named in the responses"
  },
  "publisher": {
    "name": "Rotgar Research",
    "url": "https://rotgar.com/"
  },
  "canonical_url": "https://rotgar.com/medical/resources/chatgpt-medical-provider-shortlists",
  "license": {
    "name": "Creative Commons Attribution 4.0 International",
    "identifier": "CC-BY-4.0",
    "url": "https://creativecommons.org/licenses/by/4.0/"
  },
  "counts": {
    "responses": 180,
    "prompt_cells": 72,
    "service_lines": 6,
    "metros": 3,
    "final_provider_positions": 720,
    "first_action_query_strings": 720,
    "scenario_summary_rows": 3,
    "segment_rows": 42,
    "response_metric_rows": 180,
    "stability_rows": 72,
    "stability_pair_rows": 144,
    "prompt_manifest_rows": 72
  },
  "collection_configuration": {
    "surface": "OpenAI Responses API",
    "model_alias_requested": "chat-latest",
    "model_label_returned": "chat-latest",
    "web_search": "required",
    "search_context_size": "low",
    "store": false,
    "conversation_state": "fresh for every response",
    "approximate_user_location": "fixed to the prompt metro",
    "maximum_output_tokens": 1200
  },
  "measurement": {
    "final_positions_already_named_rate": "Final provider positions occupied by organizations present in first-action query strings divided by all final provider positions.",
    "first_action_to_final_retention_rate": "First-action provider occurrences also present in the final answer divided by all first-action provider occurrences.",
    "final_absent_from_first_action_rate": "Final provider positions occupied by organizations absent from first-action query strings divided by all final provider positions.",
    "stability": "Mean pairwise Jaccard similarity within exact repeated-prompt cells, then an unweighted mean across cells in the reported scope.",
    "confidence_intervals": "95% percentile intervals from 10,000 bootstrap samples clustered by exact prompt cell; fixed scenario-specific seeds."
  },
  "bootstrap": {
    "interval_type": "95% percentile",
    "repetitions": 10000,
    "cluster_unit": "exact prompt cell",
    "scenario_order": [
      "general-comparison",
      "appointment-access",
      "cost-insurance"
    ],
    "retention_seeds": {
      "general-comparison": 2026081810,
      "appointment-access": 2026081811,
      "cost-insurance": 2026081812
    },
    "final_absent_seeds": {
      "general-comparison": 2026081820,
      "appointment-access": 2026081821,
      "cost-insurance": 2026081822
    }
  },
  "interpretation": "A dated descriptive benchmark of observable query strings and final provider lists from one API configuration. It does not identify hidden model reasoning or prove that a provider name came from training data.",
  "public_data_boundary": "The package contains prompts, anonymous response-level counts, aggregates, charts, and documentation. It excludes raw API payloads, full answers, returned query text, response IDs, provider-level mention records, credentials, and internal collection identifiers.",
  "distributions": [
    "chatgpt-medical-provider-shortlists-v1.0.csv",
    "chatgpt-medical-provider-shortlists-v1.0.json",
    "chatgpt-medical-provider-shortlists-v1.0.xlsx",
    "scenario-metrics.csv",
    "segment-metrics.csv",
    "response-metrics.csv",
    "cell-stability.csv",
    "stability-pairs.csv",
    "prompt-manifest.csv",
    "data-dictionary.csv"
  ]
}