YouTube API with Python: Complete 2026 Tutorial

Part 2

How do you call the OutlierKit API from Python?

The OutlierKit API is a YouTube competitive-intelligence API that returns outlier scores, semantic channel search and similarity, video transcripts, comments, and keyword search volumes as structured JSON — one bearer-token key, 1 credit per call, on OutlierKit Pro ($49/month) and Max ($199/month) plans. The official YouTube Data API gives you raw metrics. What it can't tell you is which videos are outperforming their channel's baseline — the signal that actually matters for content research. Deriving it yourself means pulling every channel's history, computing rolling averages, and burning quota all the way down. The OutlierKit API ships that intelligence pre-computed: outlier scores, semantic search, keyword volumes, and cached transcripts. No SDK required — it's plain REST with bearer auth, so Python's requests library is all you need. Every call costs 1 credit.

1. Outlier video search

One POST replaces the search → stats → channel-baseline pipeline from Part 1. Results come back already scored:

import os
import requests

API_BASE = "https://outlierkit.com/api/v1"
HEADERS = {
    "Authorization": f"Bearer {os.environ['OUTLIERKIT_API_KEY']}",
    "Content-Type": "application/json",
}

resp = requests.post(
    f"{API_BASE}/outliers/search",
    headers=HEADERS,
    json={
        "query": "home workout",
        "limit": 10,
        "minOutlierScore": 2,   # only videos at 2x+ their channel baseline
    },
).json()

if not resp["success"]:
    raise RuntimeError(resp)

data = resp["data"]
print(f"Found {data['totalFound']} outliers "
      f"(credits left: {resp['credits']['remaining']})")

for video in data["results"]:
    metrics = video["outlierMetrics"]
    channel = video["channel"]
    print(f"{metrics['outlierMultiplier']:>5.1f}x  "
          f"{video['title'][:60]}  — {channel['title']} "
          f"({channel['subscribers']:,} subs)")

Useful knobs: limit (1–100, default 20), offset, threshold (semantic similarity, 0–1, default 0.35), and sortBy (similarity, recent, or views). Each result's outlierMetrics object carries outlierMultiplier, channelOutlierRatio, viewsVsChannelAvg, viewsVsRecentAverage, and highestValue.

2. Keyword research

Expand seed keywords into ranked, deduplicated YouTube search keywords with monthly volume and competition:

resp = requests.post(
    f"{API_BASE}/keywords/research",
    headers=HEADERS,
    json={
        "keywords": ["fitness", "home workout"],
        "minVolume": 5000,
        "limit": 25,
    },
).json()

if resp["success"]:
    for kw in resp["data"]["keywords"]:
        print(f"{kw['volume']:>9,}/mo  comp={kw['competition']}  {kw['keyword']}")
    print(f"Total found: {resp['data']['totalKeywordsFound']}, "
          f"after volume filter: {resp['data']['filteredKeywords']}")

3. Transcripts that work from servers

Unlike the unofficial scraper from Step 5, this is a plain authenticated GET — it works identically from your laptop, AWS, or a CI runner, and transcripts are cached on first fetch:

video_id = "dQw4w9WgXcQ"
resp = requests.get(
    f"{API_BASE}/videos/{video_id}/transcript",
    headers=HEADERS,
).json()

if resp["success"]:
    transcript = resp["data"]
    print(f"Source: {transcript['source']}, "
          f"duration: {transcript['totalDurationMs'] / 1000:.0f}s")
    for seg in transcript["segments"][:5]:
        print(f"[{seg['startTime']}] {seg['text']}")
    full_text = transcript["fullText"]  # ready for an LLM prompt

All endpoints share the same response envelope — {success, data, credits: {charged, remaining}, timing, requestId} — so one helper function can check success and track credit burn across your whole script. Full request and response schemas live in the live API docs.