Video SEO: Titles, Descriptions, Chapters and Transcripts

Video metadata is a handful of fields, and most advice about them collapses into one instruction: put your keywords in. That produces titles that describe nothing, descriptions nobody reads, and chapter labels that cannot be used to navigate.

The fields do different jobs. Understanding which job each one has makes the decisions straightforward, and it produces metadata that serves both the viewer scanning a list of results and the systems deciding what to show them. This guide covers what each field is for, what makes a good one, and the two assets most people already have and never publish.

What Each Field Is Doing

The two-job problem in the title row is the whole difficulty. A title has to be accurate enough that the person who clicks is not disappointed, and interesting enough that they click. Those pull in different directions, and resolving them by abandoning either is why so much video metadata is bad.

Titles

Write the accurate version first. What is this video, in plain words, with no attempt to be interesting? That is the constraint, and everything else has to remain true to it.

Then make it worth choosing without breaking it. Specificity is the main tool available: a title naming a particular thing beats one naming a category, because it tells the reader whether it applies to them. Numbers and concrete nouns help for the same reason.

Front-load the identifying words. Titles are truncated in more places than they are shown in full, and the first few words are the only ones you can rely on being visible.

And keep the promise. A title that overstates produces exactly the behaviour you do not want — a click followed by an early exit, from someone who now knows not to trust the next one. The strongest hook is one the video actually pays off, which is the same principle that governs the opening seconds, discussed in what happens in the first three seconds of a short.

Descriptions

The description is really two fields with different purposes.

The first line or two is what appears before anyone chooses to expand. It is read by people who are still deciding, which makes it the most valuable text in the whole record. Use it to confirm what the title promised and add the detail the title could not fit — who this is for, what specifically it covers.

What it should not contain is a greeting, a channel introduction, or a subscribe request. Those occupy the one place a hesitant viewer is looking, with information that is irrelevant to their decision.

The body is where everything else goes: full context, links, credits, timestamps. It is read by relatively few viewers and by every system that indexes the page, which is a reason to write it as genuine prose rather than as a keyword list. Text that reads naturally describes the content accurately, and accurate description is what indexing systems are trying to extract.

Chapters

Chapters are navigation, and the test for a good one is whether a viewer could use it to get somewhere.

That means labels should name the content, not the theme. "Setting the exposure" is a destination; "Getting started" is not. A viewer scanning a chapter list is looking for the part that answers their question, and abstract labels give them nothing to aim at.

Two structural points. Chapters should follow the actual sections of the video rather than being imposed on it — if the video does not have distinct parts, it does not need chapters. And the first one should start at the beginning, since a gap before the first chapter usually breaks the feature.

The benefit is largely the navigation itself. A viewer who can jump to what they need is a viewer who stays rather than leaving to find a video that is easier to skim. Where the video is genuinely reference material, this matters more than any other item on this list.

The Two Assets Nobody Publishes

Most video producers already have both of these and use neither.

The transcript. If you captioned the video, you have a full text record of everything said in it. Published where the platform allows, it makes the content searchable at a level of detail no description can reach, and it serves viewers who prefer to read or who are scanning for a specific point.

The version worth publishing is the corrected one, not the raw automatic output. An uncorrected transcript full of misheard names is worse than none, because it associates your content with the wrong terms — the correction process is described in a systematic way to fix bad auto-captions.

The caption file itself. Distinct from the transcript, because it carries timing. Uploading it gives you accurate captions rather than platform-generated ones, and accurate captions serve a substantial part of your audience directly. That has its own set of requirements beyond search, covered in what accessibility standards ask captions to contain.

Both of these are produced as a byproduct of work you have already done. Publishing them is one upload each.

Thumbnails Are Metadata Too

A thumbnail is the field most often designed at the wrong size.

It will be seen small — often very small, in a crowded list, on a phone. A thumbnail with three elements and a line of text works there. One with a detailed scene, a subtitle and a logo does not, because none of it resolves.

Design at the delivery size and then view it at about the width of a thumb. Whatever is still readable is what the thumbnail communicates; everything else is decoration you paid for and nobody receives.

Consistency across a channel is worth more than variety. A recognisable treatment means a returning viewer identifies your video in a list before reading anything, which is a benefit no individual thumbnail can produce on its own.

What Metadata Cannot Do

A caution before the checklist, because a great deal of effort gets misdirected here.

Metadata determines whether the right people can find and choose your video. It does not determine whether they stay, and no title has ever rescued a video that did not deliver. When something underperforms, the metadata is the most visible thing to adjust and usually not the thing that was wrong.

The useful diagnostic is where people leave. If few people click, the title and thumbnail are candidates. If many click and leave quickly, the video did not match what they were promised — which is a metadata problem of a different kind, and the fix is to make the promise more accurate rather than more appealing.

An Order That Works

Next Steps

Take your most recent video and read only the first line of its description. If it is a greeting or a channel introduction, that is the highest-value text you have and it is currently spent on nothing.

Then check whether the caption file and transcript from that video were ever published. For most people the answer is no, and uploading them takes minutes for assets that already exist. Keeping this consistent across a schedule is its own discipline, covered in building a video content calendar you can actually keep, with related material under publishing and growth.