Status
Whether it is up, and what it cannot do
The check below runs when you open this page. Underneath is a list of the limits we know about, kept here so you find them before you spend credits rather than after.
Live check
Your browser calls the same probe our platform uses. It confirms the API is answering and can reach its database — it does not test every provider we depend on, and we would rather say that than show a green tick that means less than it looks like.
Known limitations
A video with no caption track cannot be fetched by URL
Paste a YouTube URL and we pull the captions — your own videos through your connected channel, and any public video that has a caption track. A video with no caption track at all has nothing for us to retrieve, and the import says so without charging you. You can still get the text in: the bookmarklet on the Transcripts page reads the transcript panel in your own browser and sends it here with its timestamps intact, or you can paste it, upload SRT/VTT, or upload the audio for transcription.
A pasted link gives Bench Optimize metadata, not the spoken content
Bench Optimize is built for the video you are publishing — a script, a transcript, or a video on a connected channel. It accepts any public link as a reference, but that path reads only the title, description and tags: chapters come back empty rather than invented, and nothing in the package reflects what was actually said. Import the transcript and optimize that instead.
A very short long-form video is counted as a Short
Outlier baselines are measured separately for Shorts and long-form, using YouTube's own 180-second definition. A genuinely short long-form video — a two-minute explainer, say — therefore gets compared with the channel's Shorts. That is the lesser of two errors, since pooling the formats skewed the median for every upload on the channel, but it does mean a factor on a very short video is worth a second look.
Uploads are capped at 9 MB
The request body is parsed before our code runs, and larger uploads fail without a useful error, so the cap is set below that point. Long recordings need compressing first — a spoken-word file at a low bitrate fits comfortably. Direct-to-storage uploads would lift this and are on the list.
An account is one person
There are no seats, invitations, shared workspaces or roles. Every plan is a single login, and the Studio plan raises limits rather than adding collaborators. If you need several people in one account today, you would be sharing credentials, which we do not recommend.
There is no public API
Every route authenticates with a session cookie; there is no token auth, so nothing here can be driven programmatically yet. The exports on each plan are file downloads — Markdown, plain text, SRT, CSV and JSON — not an API.
Research shares one YouTube quota
The YouTube Data API allows a fixed number of units a day across everyone using PublishBench, and a search spends a hundredth of it. Each plan therefore has its own daily research allowance, listed on the pricing page, and the last of the shared pool is held for paid plans. Your plan's allowance is a cap rather than a promise: on a busy day the shared pool can run out first. When it does, research says so and stops — or offers what earlier research already fetched, clearly labelled as not a live ranking. Everything resets at midnight Pacific time.
Incidents
We do not run a separate incident feed yet. If something is wrong right now, tell us on the contact page.