Build a fantasy football AI from podcast transcripts

The shows

The main fantasy football podcasts all have transcripts:

• FantasyPros Fantasy Football Podcast
• Fantasy Football Today
• Fantasy Football Today Beyond the Box Score
• Fantasy Focus Football
• The Late-Round Fantasy Football Podcast
• Fantasy Football Happy Hour with Matthew Berry
• Dynasty Nerds Podcast
• The Ringer Fantasy Football Show
• Establish The Run
• Fantasy Footballers

Others are a free search away: dynasty, DFS and single-team shows included.

No code: ask Claude or ChatGPT

Connect Spoken to Claude or ChatGPT (how to connect), then ask in plain words: "Get this week's FantasyPros and Fantasy Football Today episodes and tell me what they said about Puka Nacua." The assistant finds the episodes and reads the transcripts itself. Each new episode costs one credit, the same as through the API.

That works for a handful of episodes at a time. For every episode of six shows across a season, use the script below and keep the files.

How the script fits together

Step 1: collect the last three weeks

The show ids come from /search or from the show links above. Files already on disk are skipped, so the same script is also the daily update.

import pathlib, time, requests
from datetime import datetime, timedelta, timezone

API = "https://spoken.md"
HEADERS = {"x-api-key": "YOUR_KEY"}
SHOWS = {
    "1138942145": "FantasyPros",
    "261735167": "Fantasy Football Today",
    "260537420": "Fantasy Focus Football",
    "1224965828": "Late-Round",
    "1523722173": "The Ringer",
    "1473055758": "Establish The Run",
}
since = datetime.now(timezone.utc) - timedelta(days=21)
root = pathlib.Path("ff")

for show_id in SHOWS:
    episodes = requests.get(f"{API}/podcasts/{show_id}/episodes", headers=HEADERS).json()["episodes"]
    for ep in episodes:
        if datetime.fromisoformat(ep["date"].replace("Z", "+00:00")) < since:
            continue
        path = root / show_id / f"{ep['date'][:10]}-{ep['id']}.md"
        if path.exists():
            continue
        r = requests.get(f"{API}/transcripts/{ep['id']}", headers=HEADERS)
        if r.status_code == 402:       # out of credits: top up, then re-run
            raise SystemExit("out of credits")
        if not r.ok:                    # no transcript, or upstream error: skip
            continue
        path.parent.mkdir(parents=True, exist_ok=True)
        path.write_text(r.text, encoding="utf-8")
        time.sleep(0.34)                # about 3 requests a second

For those six shows, three weeks in season is about 150 episodes, so 150 credits. Change days=21 to reach further back; before the draft, last season and the offseason episodes are the useful part.

Step 2: ask about a player

Every speaker turn starts with a bold name and a timestamp, so a transcript splits cleanly into turns. Keep the turns that mention the player, label each with show and date, and hand only those to the model. Three weeks of six shows is far too much text for one prompt; the turns about one player are a few thousand words.

import re
from anthropic import Anthropic

def mentions(*names):
    found = []
    for path in sorted(root.glob("*/*.md")):
        show, date = SHOWS[path.parent.name], path.stem[:10]
        for turn in re.split(r"\n\n(?=\*\*)", path.read_text(encoding="utf-8")):
            if any(n.lower() in turn.lower() for n in names):
                found.append(f"[{show}, {date}]\n{turn}")
    return found

excerpts = "\n\n".join(mentions("Puka Nacua", "Nacua"))

msg = Anthropic().messages.create(
    model="claude-sonnet-5-5",
    max_tokens=1500,
    messages=[{"role": "user", "content":
        "These are excerpts from fantasy football podcasts from the last three weeks, "
        "each labelled with show and date. What do the analysts say about Puka Nacua: "
        "role, usage, injury news, outlook? Where do they disagree? Cite show, date and "
        f"analyst for each point.\n\n{excerpts}"}],
)
print(msg.content[0].text)

Pass the surname as well as the full name: hosts say "Nacua" far more often than the whole name, and a less common name can be spelled more than one way in a transcript.

Other questions the same files answer

• Waivers: "Which players were recommended as pickups this week, by how many shows, and at what FAAB bid?"
• Start/sit: "For these two players, collect every start/sit call from this week's episodes."
• Usage: "Who do the analysts say is gaining snap share or target share, and on what evidence?"
• Trades: "Which players are being called buy-low or sell-high this week, and by whom?"
• Handcuffs: "If this running back misses time, who do the shows say takes over?"
• Trends: "Which players are mentioned much more this week than in the previous two?" (count mentions per week first, then ask)
• Draft prep: "Summarise the sleepers, busts and ADP risers each show named in August."

Keeping it current

Once your key has fetched from a show, GET /new lists that show's episodes released since the newest one you hold, with a transcript URL for each, for free. Run step 1 every morning, or poll /new and fetch what it lists.

curl -H "x-api-key: YOUR_KEY" https://spoken.md/new

What it costs

One credit per episode, charged on the first fetch only. Searching, listing and polling /new are free.

Credits don't expire, so what is left in February carries over to next August. Prices exclude VAT.

FAQ

Thousands of transcripts fetched by developers building summarizers, RAG pipelines, and podcast tools