Part III · The problem
A wrong policy answer
is a liability.
Five rule sets, each with its own deadlines, each revised periodically. A creator who guesses wrong loses revenue, and so does YouTube.
A monetising creator cannot get a fast, correct, citable answer to a policy question, so they guess. That guess is expensive three times over: the creator loses revenue, YouTube loses its 45% share of the same revenue, and a support agent spends twenty minutes re-reading a policy they already know.The policy surface is five separate rule sets — eligibility, advertiser-friendly ratings, claims versus strikes, strikes and appeals, and payments. All five are reproduced in Part V.
Why not just ask a general model?
Two reasons, and the second is the one that matters. Policy changes: thresholds get revised, and a model trained last year answers with last year's rule. And a confidently outdated answer is worse than no answer — told “4,000 hours” after that rule has moved, the creator plans around a number that is no longer true.
Which is why every answer here carries the clause it came from. Not so the assistant looks rigorous, but so the creator can check it.
Why it needs a tool, not just retrieval
Retrieval answers what does the rule say. It cannot answer do I qualify — that is arithmetic over the creator's own numbers, and a language model doing arithmetic is guessing. So the assistant carries one tool, and the comparison runs in Python.The tool is deterministic, auditable and tested 9/9, including the case where the question carries no numbers and it refuses to compute rather than invent a verdict.
What is cited
| Figure | Value | Source |
|---|---|---|
| Channels in the Partner Programme | 3M+ | Neal Mohan, YouTube community letter, Feb 2024 |
| Paid to creators over four years | $100B+ | YouTube blog, Sep 2025 |
| YouTube ad revenue, FY2024 | $36.1B | Alphabet FY2024 results |
| Long-form ad revenue split | 55/45 | YouTube Partner Programme terms |
What is estimated
Two arguments: support deflection, which is cost, and prevented demonetisation, which is revenue. The second is the stronger one, because money never lost beats money saved. Every input that has not been published is swept across a band rather than asserted.Support contact volume has never been published by YouTube. Nor has the rate of preventable demonetisation. Those are assumptions and are marked wherever they are used.
| Band | Support deflection | Prevented demonetisation | Total / year |
|---|---|---|---|
| low | $3.1M | $3.5M | $6.6M |
| central | $26.2M | $17.5M | $43.7M |
| high | $105.0M | $35.0M | $140.0M |
assumption Contacts per channel, handling time, deflectable share and the preventable-demonetisation rate are all unpublished. The bands above exist because of them.