Quoting sources inside a Short

19 August 2026 measurement of 13 competitor homepages in this category found that the lowest server-side body was 11 words and the highest was 2910. That spread matters here because it shows how differently tools in this space treat verifiable text on a page, and the same question applies inside the video itself. This article explains how to quote sources in a vertical Short: what a credit line needs to contain, where the footage should come from, and where on screen the credit goes.

The short answer

A source citation in a Short is a visible, readable line that names three things: the archive or library the clip came from, the title or identifier of that clip, and the date or period it depicts. It appears as burned-in text over the clip itself, not only in the description, because most viewers never open the description. The citation is written once per clip, collected into a media credit file at export, and pasted into the video description alongside the upload package. If a tool cannot tell you where each clip came from, you cannot cite it, and the video is unverifiable by design.

Why the description alone is not enough

The common habit is to dump all credits into the YouTube description field. That fails for a structural reason: the viewer watching a vertical Short sees one full-screen frame at a time. The description sits behind a tap most people never make. A claim stated over an archival image with no on-screen attribution reads as assertion, not evidence, even if the description contains a perfect bibliography nobody opens.

The fix is to put a compact citation directly on the frame where the clip plays. Three fields are enough:

1. Source name: the archive, broadcaster, or map provider.
2. Clip identifier: reel number, broadcast date, or catalog title.
3. Depicted period: when the footage was recorded, which often differs from when it aired.

Keep each line under roughly two seconds of screen time per clip. The point is not that viewers read every word mid-scroll. The point is that the information exists in the artifact itself, so anyone who pauses, screenshots, or disputes the video can check it against the actual frame.

What a media credit file looks like

A credit file is the machine-readable half of citation practice. Every rendered clip gets one entry, written at render time rather than assembled afterward from memory. The minimum viable structure per clip has five fields:

Field	Example content	Purpose
clip_id	seq_04	Ties the entry to a position in the timeline
source_name	National Archives	Who holds the material
source_ref	Reel 7, 1943 newsreel	Which specific item
depicted_period	1943	What the footage actually shows
license_note	Public domain	Whether reuse is permitted

Two practices make this file worth keeping. First, write it during production, not after publication. Retrofitted credits drift from the edit; entries generated at render time cannot. Second, ship the file with the upload package so the description paste and the on-screen lines come from the same single source of truth. ViewMade generates this file automatically with each render, writing the origin of every clip into the credit document as part of the output rather than leaving it to manual bookkeeping.

Where the footage comes from determines what you can cite

Citation is downstream of sourcing. A Short built from AI-generated visuals has no external referent: there is no archive item, no original broadcaster, nothing to point at beyond "generated". A Short built from real archive material, maps, and charts has a chain you can name link by link. That distinction decides whether your credit line is a citation or just a watermark.

When evaluating any tool or workflow against this standard, ask one question per visual asset: can I name the holding institution and the specific item? If the answer is yes, cite it using the structure above. If the answer is no because the imagery was synthesized, say so plainly instead of borrowing the language of citation. An honest label reading "visuals generated" is more defensible than a fake archive reference, and audiences increasingly notice the difference.

One caution applies to stock libraries too. Stock footage of, say, a city skyline used to illustrate a 1962 event is not a citation of that event. Either label it illustrative, or find footage from the actual period. Mixing the two without labels is the fastest way to lose credibility with the comment section, which functions as a distributed fact-checking layer on history channels.

Placement and readability rules for on-screen credits

Vertical framing constrains where text survives. The bottom fifth of a 1080x1920 frame is covered by platform UI: the caption, the audio toggle, the progress bar. The right edge hosts the action rail. That leaves a safe band across the upper-middle third of the frame for burned-in text.

Three placement rules follow:

• Put the credit line in the upper third, below the headline hook but above center, so it does not collide with subtitles.
• Match the subtitle treatment: same font family, high contrast, sized for a phone screen. A credit in 14-point equivalent type is unreadable at scroll speed.
• Time the credit to the clip, not to fixed intervals. The citation should appear exactly while its footage is on screen, then hand off to the next clip's credit.

If your tool offers multiple caption styles, pick one that keeps body text and credit text visually distinct, for example subtitles centered and credits smaller and top-aligned. ViewMade burns captions into the frame across seven selectable styles, and the credit layer follows the same rendering pass, which keeps the two texts from drifting out of sync during edits.

What this does not tell you

This article describes structure and placement. It does not report measured outcomes, because we have not measured them. Specifically, we have no data on whether on-screen citations change watch time, subscriber conversion, or comment sentiment compared with description-only credits, and we will not invent percentages to fill that gap. We also have not measured how platforms treat cited versus uncited historical content in distribution, and any claim that citations boost reach would be unsupported. The legal side is likewise outside scope: license terms vary by archive and by jurisdiction, and the license_note field above records what a tool knows, not what clears your specific use. If your channel monetizes heavily, verify rights with the holding institution before relying on a public-domain assumption.

<!-- faq -->

Frequently asked questions

Do citations hurt retention in a Short? We have not measured this, so treat the following as reasoning rather than data. A two-second credit line occupies a small fraction of a 30-to-60-second video and competes with the narrative for attention only briefly. The larger retention risk is a disputed claim in the comments, which pulls future viewers away. If you want certainty, run your own split test across two uploads of comparable topics and compare your own analytics.

Should I cite the search engine or the archive? Cite the holding institution, not the discovery route. "Found via a web search" tells the viewer nothing checkable. The archive name plus the item reference lets someone request or locate the exact material. If the material came through an aggregator, list the aggregator as a secondary line beneath the primary archive credit.

How do I handle a clip whose origin I could not verify? Label it. A line such as "origin unverified" attached to the specific clip is honest and keeps the rest of the video's citations credible. Removing the unverifiable clip entirely is the stronger option when the claim it supports is central to the video. Never substitute a plausible-looking guess for a missing source, because a wrong citation is worse than none.

Does the description still need a full credit list? Yes. On-screen lines are compressed and timed for viewing; the description carries the complete file, including license notes and any secondary references. Paste the media credit file into the description on upload so both layers exist. Viewers who screenshot a frame and search the description should find an entry matching exactly what they saw on screen.