What a library audit is for
Most of the site's YouTube pages deal with one video at a time. The titles, descriptions and chapters guide covers the text a video carries and ends with a quarterly pass over the library; the thumbnails and click-through guide covers the image and how to test it; the diagnosis for a video not showing in YouTube search starts from one video that is missing from the results. The video SEO hub names the library pass in a sentence. This tutorial is that pass written out, and it covers more than search: whether each video still has a next step, whether two videos are competing for the same query, and whether the video is still accurate.
The audit is not an account scorecard. The social media presence audit rates whole accounts across platforms and decides which deserve effort; this page assumes the channel deserves effort and looks inside it. It is also not a set of house rules for corrections. The guide to updating and correcting content sets those for web pages, and the same thinking applies to a video description that has gone out of date.
Two limits shape the method. YouTube documents the broad signals its search and recommendation systems use, but not their weights, and it does not document whether editing an old video changes how widely it is shown. And a change to a live video is a before-and-after comparison, not a controlled test, so the audit changes little at a time and reads the result slowly. Where the channel is eligible for YouTube's own thumbnail test, which the thumbnails guide describes, that is the cleaner way to compare thumbnails on a single video.
Before you start
- Access to YouTube Studio with analytics, including Advanced mode, which the traffic sources guide uses to break views out by source.
- The channel strategy in writing: its promise, pillars and formats from the channel strategy hub, so each video can be judged against the job the channel gave it. A channel without one can write it first with the YouTube channel strategy document template.
- A spreadsheet with one row per video, which will become the audit record.
- The channel's query register, if one exists: the list of primary queries assigned to search-led videos. The search diagnosis's prevention section explains how to keep one.
- A change log, shared with anyone who edits the channel. The measurement change log tutorial shows the habit for tracking changes; here it records which video changed, what changed, when and why.
- A list of products, prices, offers and facts that have changed since the oldest videos were made, from whoever owns the product range.
- Time set aside in proportion to the library. The time shown for this tutorial covers setting up and running the audit on a small library. A channel with dozens of long-form videos, such as the 84 in the example below, should plan several sessions: the export, the sorting and the checks first, then the changes in logged rounds.
The steps
Export the library with its traffic sources
In Advanced mode, export every long-form video with its publish date, views, impressionsOne display of an ad or search listing, as counted by the platform., impressions click-through rate (CTRShare of impressions that result in a click.), Average View DurationTotal watch time of a video divided by its number of views. and subscribers gained for a recent period, then add the share of views from each main traffic source: YouTube search, suggested videos, browse features, external and playlists. Export Shorts as a separate sheet, because the two formats are read on different terms and pooled figures describe neither. Videos with no impressions in the period go to the bottom of the sheet; nothing done to their packaging can be measured.
Give each video its job
Add a column for the job each video does for the channel: answering a search, being suggested beside related videos, serving returning viewers in a series, supporting a product page, or something else the strategy names. Most videos announce their job in their traffic source mix: a video led by YouTube search is a search video whatever it was meant to be. Where the intended job and the actual mix disagree, note both. The job decides which checks matter, which the table after the steps sets out.
Set the channel's own baseline by age and format
Judge videos against the channel, not against figures from elsewhere. For each format and job, take the channel's recent videos and read impressions click-through and retention at a fixed age, as the low click-through diagnosis does for a single video. Where only a handful of videos share a format, treat the baseline as a rough range, as the guide to reading small numbers advises, and look only for large gaps.
Find search videos whose packaging underperforms
Sort the search videos by search impressions. For each one that is shown often in search but clicked less than its baseline, open its search terms list and compare it with the title. If the video is found for words its title does not use, the title is the change to make; if the words match but the click still lags, the title and thumbnail pair is the suspect, and the thumbnails guide covers how to judge it alongside retention. Mark one proposed change per video. Search videos are often Evergreen ContentContent that stays relevant and useful long after it is published, rather than being tied to a date or a news moment., which is why their packaging repays attention years after publishing.
Check chapters and descriptions against the documented requirements
For every long-form video, check that chapters exist and meet the requirements YouTube documents on its chapters page, which the titles guide summarises: the list sits in the description, the first timestamp is 0:00, there are at least three timestamps in ascending order, and each chapter runs for at least 10 seconds. These rules can change, so check YouTube's current chapters documentation before treating a video as failing. Then read the first lines of the description as a viewer would see them. Do they say what the video covers in the words viewers search for, and does the one link that matters sit there rather than below the fold? Note captions that were never reviewed; the guide to AI for alt text, captions and transcripts sets out the review a caption file needs.
Check the next step at the end of every video
Watch the last moments of each video that earns views and note what a viewer is offered: an end screen, a playlist, a named next video or nothing. A search answer that ends on the answer is the commonest gap, and the diagnosis for views without subscribers explains why it costs the channel returning viewers. Check that end screens and cards still point to videos that exist and still fit, because old end screens often point to videos the channel has since replaced.
Find videos that are out of date or compete for one query
Go through the list of changed products, prices and facts and mark every video that mentions one. Then compare the primary queries of search videos: two videos built for the same search split its impressions, the YouTube version of Keyword CannibalisationTwo or more pages on one site competing for the same query and intent, splitting signals.. For each pair, decide which video should own the query and how the other should be retitled toward an adjacent one or point to the owner.
Prioritise the fixes
List every proposed change with the video's impressions beside it, and place each on the impact-effort matrix. Changes to the packaging of videos that are still shown widely, missing next steps on videos with steady views, and corrections to inaccurate videos sit high on impact; retitles are cheap, new thumbnails cost a design session, and replacement videos cost a production slot. Put corrections first regardless of effort, because a wrong video is a trust problem before it is a reach problem.
Change one round at a time and log it
Make the first round of changes on a manageable set of videos, one element per video, and record each in the change log with the date, the old version and the reason. Correction notes added below the description's first lines, pinned comments and end screens leave the packaging that viewers see in search and feeds untouched, so they can go onto many videos in one round; titles and thumbnails are the changes to ration. Add replacement videos to the production plan rather than leaving them as notes in the audit.
Read the results at a fixed age
Read each changed video at the same interval after the change, on the surface the change was meant to help, against the same interval before it and against the channel's baseline. Expect noise: distribution moves for reasons that have nothing to do with the edit. Keep changes that held, reverse those that clearly made things worse, and carry the lessons into the next round and into the briefs for new videos.
What to check by the video's job
Not every check applies to every video. The job given to each video earlier in the audit decides what deserves attention.
| Video's job | Read first | Usual fixes | Leave alone |
|---|---|---|---|
| Answering a search | Search impressions, click-through on search, search terms | Title against the words viewers use, the description's first lines, chapters | Thumbnail style that returning viewers recognise, unless click-through lags its baseline |
| Being suggested beside related videos | Click-through on suggested, retention through the opening | The thumbnail and title pair, the opening if viewers leave early | The search wording, which suggested viewers rarely see |
| Serving a series | Views from playlists and end screens, returning viewers | Playlist order, end screens to the next episode, series title marker | One-off packaging experiments that break the series pattern |
| Supporting a product or page | Clicks on description links, external views from the page | The link in the first lines, tagged links, accurate product details | Packaging built for search if the video is mainly watched from the page |
| Out of date | What has changed and how widely the video is still shown | A correction in the description and a pinned comment, an end screen to the newer video, a replacement on the plan | Deleting it before the links that point to it are moved |
A car-care brand audits its tutorials (illustrative)
Vostrilla Car Care audits its tutorial library
Vostrilla Car Care, an invented car-care products brand, has 84 long-form videos and 120 Shorts. Its export shows 31 long-form videos led by YouTube search, and its baseline for search videos at 28 days is a 4.0% click-through rate on search impressions. Nine search videos with steady impressions sit well below that, the lowest at 2.1%; their search terms show viewers typing "clay bar" while the titles say "decontamination". Two videos were both built for "how to wax a car", 23 videos have no chapters, 14 search videos end with no end screen, and 6 videos feature a polish the brand stopped selling. The first round retitles the five most-shown under-clicked videos, adds chapters to 12 videos, adds end screens to all 14, and puts a correction and a pinned comment on the six discontinued-product videos, pointing to a replacement tutorial added to the production plan. The second "how to wax a car" video is retitled toward hand-waxing a new car. Read 28 days after the change against the 28 days before, the five retitled videos average 3.6% on search against 2.3% before, on similar impressions, and views from end screens on the 14 videos rise from 310 to 1,140. The brand logs both, leaves the thumbnails for the second round, and schedules the next full pass for the following quarter.
The figures belong to the example. The shape is what transfers: every judgement is against the brand's own baseline, only one element changes per video in each round, corrections go first, and each result is read at the same age before anything else is touched.
Common mistakes
- Auditing against outside benchmarks. A click-through rate quoted for other channels says nothing about this audience; compare each video with the channel's own videos of the same format at the same age.
- Changing everything at once. A new title, thumbnail and description on the same day make it impossible to tell which change mattered. Change one element per video per round.
- Working on videos nobody is shown. Packaging changes on a video with no impressions cannot be read. Start with the videos that are still being offered.
- Reading results after a few days. Distribution moves on its own; read at a fixed interval and against the same interval before the change.
- Deleting dated videos first. Removing a video breaks links, playlists and end screens that point to it. Correct, redirect viewers with an end screen and plan a replacement before deciding anything more drastic.
- Leaving two videos on one query. Two search videos built for the same words compete for the same results. Decide which owns the query and retitle the other.
- Skipping the change log. Without the date and the old version, the next audit cannot tell which changes worked, and the channel relearns the same lessons.
