| name | youtube-shorts |
| description | Write YouTube Shorts as searchable answers — a query-shaped title, a beat table, and a description written for YouTube search rather than for the feed. Use when the user says 'YouTube Short,' 'write a Short,' 'Shorts script,' 'Shorts title,' 'Shorts description,' 'youtube channel setup,' 'make this Short searchable,' or names YouTube Shorts as the channel for a product. Shorts only — sub-3-minute, square or taller. NOT for a long-form YouTube script, video chapters, or a video meant to earn AI citations: those belong to upstream video and social, and the citable written answer belongs in geo-answer-page. Covers why Shorts are a search surface rather than a scroll surface, the title-as-query rule, why Shorts are a search asset and not an AI-citation asset, pinned comments, transcripts, and channel setup before the first upload. Writing only — to publish see youtube-shorts-publish. For TikTok see tiktok-script, for Reels see instagram-post. |
| metadata | {"version":"1.1.0"} |
YouTube Shorts
Shorts share an aspect ratio with TikTok and Reels and almost nothing else. The difference is where the viewer comes from:
- TikTok: a cold scroll. No intent.
- Reels: a feed of things adjacent to what you already follow.
- Shorts: increasingly, a query. Someone typed a question, or YouTube matched their question to your video.
That changes the writing at every level. A TikTok hook interrupts; a Short answers. Because of the intent, a Short can be more expository and less performative without losing the viewer — they came for the answer, so give it in the first sentence.
Two separate distribution systems reach the same upload, and they rank on different things. YouTube's own guidance is unusually clear here:
- Feed surfaces (Shorts tab, Home) rank on performance and relevance to that viewer — percentage who chose to view, average view duration, average percentage viewed, plus likes and post-watch surveys. Metadata is not named.
- Search ranks on, in YouTube's words, "how well the metadata matches the viewers' search."
So the honest answer to "do titles and descriptions matter for Shorts" is: not for the feed, decisively for search. Since January 2026 there is a dedicated Shorts option under the search Type filter, which makes that lane genuinely addressable — a viewer selecting it is choosing short-form on purpose. (It cuts both ways: Shorts can now be filtered out by anyone wanting long-form.)
What Shorts are not: an AI-citation asset
This is worth stating plainly because a lot of current advice gets it backwards. Two independent 2026 datasets, different methodologies, agree on the thing that matters:
- Indexly (May 2026, 3.03M citation URLs across 95K prompt responses): YouTube was ~0.75% of all citations, and of those, 95.7% were long-form and 3.0% were Shorts. Their explanation is the right one — AI extracts from transcripts, and Shorts barely have any.
- Otterly.AI (March 2026): Shorts are 5.7% of YouTube citations. The high-value mechanism they identified is chapter-level extraction — 78% of timestamped citations hit 2–5 different chapters of the same video, turning one upload into several citable units.
Chapters are structurally impossible on a sub-three-minute Short. If AI citation is the goal, Shorts is the wrong asset class and no amount of description-writing fixes it — the citations land on long-form with a clean transcript, real chapters, and a substantial description.
(The two studies disagree wildly on per-engine splits — Indexly puts Gemini at 34.9% of YouTube citations, Otterly at 0.2%. Treat engine-level breakdowns from either as unreliable; carry only the claim both support.)
What Shorts are good for is YouTube search, and as the top of a funnel into the long-form piece that can be cited. Write them for that, and put the citation ambition on the canonical written answer (see geo-answer-page) and on the longer video.
Read first
If .agents/brand-context.md exists, read it — the claim ledger, prohibited-language list and
channel CTA policy. If it does not, do not stop: derive what you need from the codebase as you
draft, and run claim-check on the finished draft, which builds the ledger properly.
Channel setup, before the first upload
Do not upload into an unfinished channel. It costs nothing to fix first and is expensive to fix later:
- Channel name and handle — the product or the person, matching how you sign everywhere else.
- About text — what the product does and, explicitly, what it does not. This text is retrieved and quoted.
- Recovery and 2FA — before the channel has value, not after.
- Upload defaults — description template, category, language, comment settings, and the boundary statement pre-filled so nobody ships a Short without it.
- Description template — the four-block structure below, saved so it is never rebuilt from memory.
Verify claims in the About text the same way as any other copy. "Free readings, no card required" in an About box is a claim, and if guests actually get exactly one, ever, that sentence is the same defect the channel might be apologising for elsewhere.
The title is a query
Write the title as the question a person types, then a short clarifying tail.
What Does Whale Eye Mean in Dogs? The "Give Me Space" Signal
Does a Wagging Tail Mean a Dog Is Happy? Not On Its Own
Rules:
- Question first, answer-flavoured tail second. The tail is what makes it a click rather than a maybe.
- Under ~70 characters stay visible before truncation; the hard cap is 100.
- Use the words people search, not your internal vocabulary. If your product calls it a "cue trace" and searchers call it "body language", the title says body language.
- No clickbait gap. If the title asks a question, the video answers it in the first ten seconds. A Short that withholds its answer gets closed, and Shorts are ruthless about the first few seconds.
- One question per Short. Two questions is two Shorts, and two Shorts is two queries you can rank for.
The beat table
| Time | On screen | Voice |
|---|---|---|
| 0:00–0:04 | `WHALE EYE` over the schematic, whites highlighted | "Whale eye is when you can see the whites at the side of a dog's eye." |
| 0:04–0:10 | Split diagram: *head still* / *eyes moved* | "It happens because the head holds still and only the eyes move away." |
| 0:10–0:18 | Big text: `I want this to stop.` | "It generally means the dog wants the situation to end…" |
| 0:18–0:26 | Four labels: `ears · eyes · tail · posture` | "One cue is never the whole read…" |
| 0:26–0:31 | End card + `brand.com` | "We build X. Visible cues, not a diagnosis." |
Structure that suits the search intent:
- Define the thing (0–5s) — the literal answer to the title.
- Explain the mechanism (5–12s) — why it happens. This is what a search viewer stays for and a scroll viewer does not.
- Give the nuance (12–22s) — the qualification, the counter-case, the thing one cue cannot settle.
- Widen (22–28s) — how it fits the larger method.
- End card (last 3–5s) — who you are and the boundary. Brief and unapologetic; the viewer got their answer, they can tolerate five seconds of who you are.
On runtime: there is no verified sweet spot, and anyone quoting one is guessing. I found 15–24s, 15–45s, 20–60s, 30–60s and 45–55s all asserted as the 2026 optimum, each with confident retention figures and none with a primary source; the widely-cited studies behind them are from 2023 and predate three-minute Shorts entirely. YouTube publishes no recommended length.
The cap is 3 minutes (180s) for square-or-taller uploads. Beyond that it uploads as long-form and leaves the Shorts feed. Shorts Audio Library music is capped at 90 seconds within a 3-minute Short, so long Shorts with library music are constrained.
Take the length from the content's natural rhythm, then measure average percentage viewed — which is a signal YouTube actually names — and let your own numbers set your norm. Do not pad to hit a target, and never truncate an answer to fit one: an incomplete answer defeats the entire reason to be on this channel.
The description does the search work
Write it as prose that stands on its own, because it will be read by people who never watch — and because it is the metadata YouTube says search matches against.
Four blocks, in order:
[1] The answer, 60–120 words, standalone.
Opens with a direct definition or answer. No "in this video". Assume the reader
has not watched and never will.
[2] The nuance and the method.
The qualification, and how this fits the broader thing you teach.
[3] Links.
Primary: the product or the canonical guide, with full tracked URL.
Secondary: the written long-form answer this video is based on.
[4] Who you are and the boundary.
One short paragraph: what the product does, what it does not do, and where
someone should go instead when it matters.
#three #relevant #hashtags
Write it in complete sentences — fragments and bullet-stacks match queries poorly. Repeat the question's phrasing at least once naturally, because that is the string being matched. The first ~100 characters show before "Show more", so make them the answer.
Block 1 is not wasted effort even though Shorts rarely earn citations: it is the block that gets reused verbatim in the long-form video's description, where it can.
Pinned comment
A second, smaller citation surface. Use it for the link to the fuller written answer, with one sentence of why someone would want it:
The whole four-cue read — ears, eyes, tail, posture — is written out here: — including the readings that stop being about mood and become a vet visit.
Pin it immediately on publish. An unpinned comment is invisible within an hour.
Captions, transcripts and chapters
- Upload a corrected transcript. Auto-captions mis-transcribe exactly the domain vocabulary you are trying to rank for — the proper nouns and technical terms. Fixing them is a ten-minute job that decides whether the video is findable by its own subject.
- Burn captions into the video too. Most viewing is silent; the platform's own captions are not always on.
- Chapters apply to the longer companion piece rather than Shorts, and each chapter title is another query-shaped string.
The longer companion video is worth planning alongside the Shorts: a 4–6 minute explainer that the Shorts each point into. It is where retention, chapters and a full transcript compound, and it is a far better citation target than any single Short.
Links and attribution
Unlike TikTok and Instagram, YouTube descriptions carry working links. Use full tracked URLs, and give the primary and the secondary link distinct content IDs so you can tell "watched and wanted the product" from "watched and wanted to read more" — those are different people and they behave differently later.
2026 changes with production consequences
- Views now count from the first frame (August 2026, completing a change Shorts made in March 2025). The old definition was renamed "engaged view" — watching past the first frame, or clicking to watch. YouTube is explicit that recommendations and Partner Program qualification did not change. For a small channel this means the public view count is now a reach number and nothing else. Report engaged views internally or you will hallucinate traction, and be careful with year-over-year comparisons that straddle the change — they compare two different counters with no marker in the data.
- The dislike button was retired on Shorts and the thumbs-up became a heart (June 2026), with YouTube leaning on "Not interested" and "Don't recommend this channel" instead. Since likes are a named ranking signal and "not interested" is a named negative one, the shape of the feedback loop changed.
- Custom thumbnails came to Shorts (July 2026, Partner Program creators first, desktop-only upload path). Note there is no A/B testing for Shorts thumbnails — long-form got title-and-thumbnail testing in December 2025 and Shorts sits outside it. A Shorts thumbnail is a one-shot decision with no measured alternative, so decide it deliberately.
- Shorts monetisation is effectively closed to small channels. From February 2027, Shorts ad and subscription revenue share requires 10M qualified Shorts views over 90 days, and Partner Program entry via Shorts rose to 20M qualified views per 90 days. Treat Shorts as a distribution channel here, not a revenue one.
Platform facts in this file were verified 2026-08-26. This channel changed four times in 2026 alone; re-check before betting on any of it.
What not to do
- A title that is a statement. Statements do not match queries. Save those for TikTok.
- "In this video I'm going to…" — the answer starts now, in both the video and the description.
- A description that summarises the video. It should contain the answer, not describe a video that contains it.
- Leaving auto-captions uncorrected.
- Uploading a TikTok with the watermark on. Suppressed, and it announces that this channel is an afterthought.
- One Short answering three questions. Split it; each becomes a query you can own.
Related skills
- geo-answer-page — the canonical written answer this Short should point into. Write it first where possible; the Short is a derivative of a reviewed answer, not a substitute for one.
- youtube-shorts-publish — the publishing step, once the script is approved and the asset exists.
- tiktok-script / instagram-post — the same idea, authored for scroll and feed.
- claim-check — run on title, on-screen text, VO, description and pinned comment together.
