Guides · 2026-09-05 · By VSNARY | Emmanuel Orta

YouTube SEO case study: 81 videos rewritten, measured

A dormant channel had 81 titles and descriptions rewritten in two dated batches, with the baseline published first. Search views rose 19% against a flat year-earlier control; the flat and falling rows are printed beside it.

Terrarium Station is a reptile and invertebrate shop we operate. Its YouTube channel, @terrariumstation, had 86 uploads and had not published since 19 January 2025. We rewrote the titles and descriptions of 81 videos in two dated batches, on 1 and 5 September 2026, and published the baseline before the result came in. In the 28 days after, views from YouTube search rose 19.0%. Across the same calendar windows a year earlier, with nobody touching the channel, they rose 0.5%.

What changed after the rewrite? #

The season does not explain the gain: the same two windows a year earlier were flat. The channel’s own growth, 42% a year in search before any edit, is roughly 3% a month, which accounts for a few points of the 19, not most of it. The gain came mostly from being shown more (search impressions +11.3%), with click-through up 0.2 points. Long-form videos, all rewritten on 1 September, nearly doubled from a small base.

Flat and down, as promised before the result: subscribers gained fell from 10 to 4. Click-through on long-form fell from 7.6% to 7.4% and on Shorts from 5.5% to 5.4%. Watch time rose 11.2%, but it rose 23.1% across the same windows a year earlier, so it is not evidence of anything. In the after window 92 views fall outside both format rows; Studio counts them under other content types.

Why this channel was the test #

It was a compounding asset nobody was maintaining. In the nineteen months after the last upload its search views still grew 42% year on year, while the Shorts feed that once carried it collapsed. Search was about 85% of current views: a search channel that still looked like a Shorts channel.

What the audit found before any edit #

On 14 August, five descriptions were read by hand from the public watch pages. Three of the five linked to a marketplace storefront that returns HTTP 403, a page the shop left years ago. None linked to the shop’s own site. Each carried the dead auto-appended “Please watch” block. And the strongest query on the channel, millipede terrarium setup at 3,397 search views, sat as a tag on the video that ranks for it while its title read “Building A Bioactive Enclosure For Breeding Rare Flame Leg Millipedes”. The tags knew what people typed; the titles did not. Roughly 7,500 views a month were arriving with no path to anything the shop sells.

The channel tally, run on 2 September between the two batches, read all 86 uploads:

Every caption track was auto-generated, and fifty-one clips carried none at all, which counts as not measurable rather than failed. Search demand was also lopsided against the catalogue: of the top 100 queries, isopods took 35.2% of search views, millipedes 21.6%, springtails 17.4%, crested geckos 9.7% and mantises 9.6%. The largest single cluster, rubber-ducky isopods, had no product page when the audit ran. It has one now.

What exactly was rewritten? #

A rewrite, precisely: the title leads with the query the video already ranks for, with no channel-name suffix and no adjective in front of the animal. The description opens with one sentence that states the subject, then a link to the shop category that returned 200 when checked, a care-guide link where a real guide exists, the site root, the breeding-records registry and the community group. No claim the footage cannot support appears anywhere.

How the result was measured, and the one correction #

Before: 4 to 31 August, the last full days before any edit. After: 6 September to 3 October. 1 to 5 September holds both batches and belongs to neither. Every video was edited, so there is no untouched cohort; the control is the same calendar windows a year earlier, read the same way. The result is split by format rather than by batch, because Studio reports format directly and every long-form video was in the 1 September batch. For the same design on a website, see how to tell if an SEO fix worked.

Correction, 6 September 2026: this page first dated the first batch 29 August and treated an 8 August to 4 September window as a clean baseline. The batch landed on 1 September, so four of that window’s days followed the edit. The baseline was moved to 4 to 31 August before the after window produced any result.

What this case cannot show #

It shows that search impressions and views moved in the 28 days after the rewrite, against a seasonal control. It cannot show why. Click and watch-time signals live inside Google and no public surface exposes the ranking scores, so the check lists them as not measurable. One channel is one channel, not a rate. The same caution applies to AI visibility claims: see how to measure AI visibility yourself.

The channel today #

The subscriber figure is rounded by YouTube, not by us. Above 1,000 the public count is published to three significant figures, so it moves in steps of ten and will read lower than the exact number in the owner’s Studio. The view counters are exact. A subscriber change smaller than ten is invisible in this series by construction.

Read from the channel’s public counters on 2026-11-03. The series opens on 2026-09-06 and holds 53 daily points across 58 days. It updates itself; nothing on this page is retyped.

What this series does not contain: impressions, click-through rate, watch time and traffic source. Those live in YouTube Analytics behind the owner’s login, and no public surface exposes them — which is why the 3 October read still comes from a Studio export. The series is public as JSON.

Run it on your own channel #

The YouTube check is free for a single video and reads chapters, captions, description and language. The channel tally reads every upload and is included on paid plans from Watch upward. The site half answers whether your own pages carry your videos at all, and whether their VideoObject schema is complete. None of it changes what YouTube does; it tells you what YouTube can read.

Questions this post answers

Did rewriting YouTube titles and descriptions increase search views?

On this channel, views from YouTube search rose 19.0% in the 28 days after 81 videos were rewritten, against 0.5% across the same calendar windows a year earlier. Most of the gain came from more search impressions rather than a higher click-through rate. One channel is a case, not a rate.

What was the control in this YouTube case study?

The same two calendar windows a year earlier, when nobody was editing the channel. Every video was edited, so there was no untouched cohort; the year-earlier windows remove the seasonal effect, and the channel's pre-edit growth trend explains only a few points of the gain.

What did not improve?

Subscribers gained fell from 10 to 4, and click-through dipped slightly on both long-form videos and Shorts. Watch time rose, but it rose more across the same windows a year earlier, so it is not evidence of an effect.

What was wrong with the channel before the rewrite?

Titles did not use the queries the videos already ranked for, descriptions linked to a dead marketplace page and never to the shop, no video had authored chapters, every caption track was auto-generated and only three descriptions reached 150 words.

Why was the baseline published before the result?

Because a baseline chosen after the result can be moved to fit it. The before window and the edit dates were on record before the after window produced a result, and the one dating error found was corrected publicly before the result was read.

Can I run the same check on my own channel?

Yes. The single-video YouTube check is free and reads chapters, captions, description and language. The channel tally, which reads every upload, is included on paid plans.

Related findings

Comments

Comments are read before they appear. Nothing is published automatically, and no account is needed.

Writing about this? Facts, live figures and marks — every number on that page is dated and traceable to a scan.

All guides · The dataset · How the dataset works