Title Oracle — score any YouTube title against any channel.

Length is decided by the surface, not by you.

The sweet spot most channels converge on lands around 50–65 characters, and that's not arbitrary — it's where browse, search, and suggested all render the full title without truncation on a typical phone. Mobile thumbnails start clipping somewhere around character 47–60 depending on the surface and device width, which means anything important you pack into the back half is rolling the dice on whether a viewer ever sees it. Desktop tolerates more, but desktop is no longer where the majority of watch time is decided.

The stronger move is to front-load. A specific, charged word inside the first ten characters catches scanning eyes faster than the same word buried at position 40, because viewers are pattern-matching titles in their peripheral vision while their thumb is already moving. “Killed”, “Broke”, a number, a name — whatever the payload is, it should hit before the eye has decided to scroll past. The Oracle's hook score weights early-position signals more heavily for exactly this reason: the first ten characters are doing roughly half the work of the entire title.

Patterns work because the algorithm has trained your audience.

YouTube's recommender doesn't score your title in isolation. It scores it against everything your subscribers have historically clicked on your channel and rejected elsewhere. Over months of impressions, the system builds a fairly tight profile of what your audience reliably engages with — the typical length range, the vocabulary, the hook structure, even the punctuation rhythm. When a new upload matches that profile, the algorithm serves it more confidently in browse, because its prediction model is more sure the impression will convert.

That's why a clean, technically excellent title can still under-deliver if it's wrong for the channel. The radar example: your viewers click “PrintNightmare killed every queue in 30s” because that's the pattern they've learned to engage with from you — sharp, specific, outcome-led, slightly conversational. They scroll past “PrintNightmare CVE technical breakdown” not because it's a bad title, but because it reads like somebody else's channel. The Oracle's pattern score is the distance between your candidate and the centre of that trained distribution.

Same principle applies to vlog and POV content — your viewers click “Riding the Alps at 4am, alone” not “Mountain road motorcycle riding scenic route review” because that's the pattern they've learned to engage with on your channel. The shape of the content doesn't change what the Oracle measures; it changes which distribution your titles are scored against.

Estimated CTR is a regression, not a guess.

The CTR estimate isn't vibes. It blends hook strength (front-loading, specificity, emotional markers) with pattern match (how cleanly the title sits inside the channel's learned distribution) and overlays a handful of proven structural patterns YouTube's data has rewarded for years: numbered-list openers, comparison framings like “X vs Y”, and time-bound concrete claims like “in 30 days” or “in 30 seconds”. Each of those structures has a measurable historical lift over a channel's baseline.

Read the score as roughly calibrated to the channel's own average. 50 out of 100 means you're landing around where the channel typically lands — not bad, not exciting. 80+ means the title is expected to over-index by roughly 1.6x the channel's baseline CTR, which is the difference between a quiet upload and one that pushes a creator into a new audience pocket. The estimate sharpens the more channel data the Oracle sees: with thirty recent titles to compare against, the regression is honest; with three, it's a rough sketch.