Flag near-identical openings in your last 20 Short scripts

YouTube's monetization page, read on October 5, 2026, describes AI-generated content made with generic or unoriginal templates, giving the impression of mass production. It gives no similarity threshold, and 9to5Google reports that the October 2 recommendation update draws no clear line either. You can still measure your own sameness. The easiest signal is the opening line.

The idea

Keep each Short's script as a text file. Take the first sentence of each, compare every pair with difflib's SequenceMatcher, and print the pairs above a ratio. The threshold is yours; 0.6 is a reasonable start for short sentences because swapping one noun in a seven-word line already gives about 0.85.

This is a heuristic. It cannot tell you whether the idea is original. It catches the failure you can fix in minutes: the same sentence with a new topic.

The script

Put the scripts in a folder called scripts, one .txt per Short, and run it with Python 3.

import glob, re
from difflib import SequenceMatcher

LIMIT = 0.6

def opening(text):
    text = " ".join(text.split())
    return re.split(r"(?<=[.!?])\s", text, maxsplit=1)[0].lower()

openings = {}
for path in sorted(glob.glob("scripts/*.txt"))[-20:]:
    with open(path, encoding="utf-8") as f:
        openings[path] = opening(f.read())

names = list(openings)
flagged = 0
for i in range(len(names)):
    for j in range(i + 1, len(names)):
        a, b = openings[names[i]], openings[names[j]]
        score = SequenceMatcher(None, a, b).ratio()
        if score >= LIMIT:
            flagged += 1
            print(round(score, 2), names[i], "|", names[j])
            print("  ", a)
            print("  ", b)
print(flagged, "pairs at or above", LIMIT)

What to do with a flagged pair

Rewrite the later opening, not the earlier one, so a published Short is not edited after the fact. Change the structure of the line, not just a word: swap a statement for a question, a number for a scene, a claim for a quoted objection.

Then extend the idea. The same loop works on the last sentence, which is where templated calls to action repeat. Add a second check on word count per script, since near-equal lengths are another sign of one frame reused.

Limits of the check

Two Shorts can open differently and still tell the same story. Two can open alike on purpose, as a series signature, which the monetization page tolerates for intros and outros. Use the output as a prompt for a human decision.

Run it on a schedule

Add the check to the step that writes the script, not the step that publishes. If a new opening scores above the limit against any of the last 20, the writer, human or model, gets the flagged pair back and tries again. That makes the check part of drafting, which is cheaper than a failed review later.

Keep the threshold in one place and log the scores. After a month you will see whether your channel's range of openings is widening or collapsing, which is more useful than any single score.