Coming Soon: Metadata That Updates Itself Based on Performance

Right now the pipeline generates metadata once, at publish time, from the video. It never finds out whether it was right.

That has always bothered me. The system produces a title, the title either works or it does not, and nothing about that outcome reaches the thing that wrote it. We are building the loop that closes it: metadata that reads how the video actually performed and revises itself.

It works in testing. It is not released.

What it does

After a video has been live long enough to have real numbers, the pipeline looks at how it converted and decides whether to try again — a different title angle, a different frame, a description reordered so a different thing is in the first two lines.

The important word is "long enough". Acting on a day of data would be acting on noise, which is the failure mode that makes most automated optimisation worse than leaving things alone. The post on which metrics mean anything works through why a CTR difference at low impressions is usually nothing, and that constraint is the hard part of building this rather than an afterthought.

What it will not do

It will not publish revisions without you. The review step applies to a revision exactly as it applies to the original, and versions remain rollback-able. An automatic system that quietly rewrites your channel while you sleep is a different product and not one I want to ship.

It will not test. There is still no A/B test against live traffic here — we are reading what happened, not running an experiment. That distinction matters and TubeBuddy remains the tool if a real test is what you want; the fuller comparison covers where each of us stops.

It will not fix a video nobody is being shown. Revision works on impressions with poor conversion. A video with no impressions has no signal to read.

The honest state of it

We have tested it and the statistics improved. I am not publishing a percentage, for the same reason the beta post does not: changing anything on a video is itself an event the algorithm reacts to, and separating that from the quality of the new title needs more data and more care than we currently have.

So treat "it improves stats" as our current internal read rather than a claim you should plan around. When it ships and we have measured it across real channels over a real window, that post will exist and it will include the cases where revision made things worse.

Why I am writing about it before it exists

Partly because pre-launch posts are how you find out you are building the wrong thing while it is still cheap to change.

The specific question I would like answered: would you want this running continuously on the whole channel, or as something you point at a set of videos deliberately? We have built it as the second, on the assumption that a channel-wide always-on rewrite is more than most people want. That assumption is doing a lot of work and I am not certain about it.

In the meantime

The manual version of this is available today and costs nothing: sort your videos by impressions, find the ones with weak click-through, and rewrite those titles yourself. The back-catalogue post covers which ones to pick and how to read the results without fooling yourself.

If doing that by hand for a dozen videos does not change anything for you, automating it will not either, and you will have found that out for free.