Our short-form video statistics methodology has one dull and useful rule: if we cannot show where a claim came from, it does not receive an approved label. Every published figure needs a source, locator, geography, reporting period, scope, caveat and review date.
That sounds obvious. It is also where most statistics pages begin to wobble. A parent company’s annual revenue gets called Reels revenue. A registered-member count becomes an active audience. An economic-impact model turns into platform sales somewhere between the source and the headline.
This page explains how we stop that happening.
The short version
- Primary platform documents and statutory filings sit at the top of the source order.
- Every number carries its platform scope, geography and measurement period.
- Product or bundle revenue, segment revenue, parent-company revenue, creator payouts and economic impact use different labels.
- View counts are compared only after their definitions and methodology changes have been checked.
- Platform rules receive a 30-day review cycle. Usage data normally receives a 90-day cycle.
- GPT Image 2.0 supplies decorative artwork. Approved records supply every displayed number, label and chart position.
- Corrections are dated and recorded. We do not quietly replace inconvenient history.
What sources qualify for the research hub?
We use a source order rather than treating every page with a logo as equally useful.
| Source class | What qualifies | How we use it |
|---|---|---|
| Regulator or statutory filing | SEC filings, annual reports filed with a regulator and equivalent statutory disclosures | Financial totals, segment scope and formal risk or methodology statements |
| Official parent-company source | Investor calls, earnings releases and corporate reports from Alphabet, Meta, Microsoft or Snap | Company and product figures, kept within the scope the parent actually disclosed |
| Official platform source | Platform newsrooms, business updates and published research pages | Product usage, feature changes, programme announcements and platform-reported studies |
| Official platform documentation | Help centres, developer documents, policy pages and API changelogs | View definitions, eligibility rules, feature availability, data fields and technical limits |
| Government or regulator | Official enforcement, legal and statistical publications | Rules or claims that require an independent public authority |
| Academic or nonprofit research | Published work with a visible method and a relevant sample | Context that the platforms do not publish themselves |
| Commissioned third-party research | A named research firm publishing its method, sample and commissioning relationship | Survey findings and modelled impact, with the commission and limits kept visible |
| Credible third-party estimate | A named research firm with a disclosed method and clear estimate language | Gaps that cannot be filled officially, always labelled as estimates |
Every source record stores the publisher, title, URL, publication date where available, access date and a precise locator. “It was somewhere in the investor deck” is not a locator.
A reachable URL is not proof that the claim remains on the page. Material facts are checked against the named section, table or paragraph. If an official site blocks automated access, the record receives a manual check rather than a dramatic red cross.
How are statistics approved?
A record moves through a fixed review path:
- Candidate. The claim may be useful, but it is not ready to publish.
- Primary-sourced. An approved researcher has found the original source and exact locator.
- Fact-checked. Scope, date, geography, unit and wording match the evidence.
- Voice-reviewed. The public claim and caveat are clear enough to publish without burying the awkward bit.
- Approved. The record can appear in HTML, charts and downloads until its next review date.
Records can move to needs review, superseded or withdrawn. The page generator refuses to use an unapproved record in a finished statistics page.
How are revenue and payout figures labelled?
Financial language is where a small wording mistake becomes a very large false claim. We use the following labels:
| Financial label | Meaning | What it must not be called |
|---|---|---|
| Official product or bundle revenue | Revenue explicitly disclosed for a named product or stated bundle of products | Segment revenue, parent-company revenue or revenue for one format inside the bundle |
| Official segment revenue | Revenue explicitly disclosed for a named reporting segment | Parent-company revenue or revenue for one product inside the segment |
| Official parent-company revenue | Revenue for the whole company | Segment, product or format revenue |
| Credible third-party estimate | A non-official estimate with a named method | Official revenue |
| Creator payout | Money paid or allocated to creators | Platform revenue |
| Advertiser spend | Money spent by advertisers | Platform revenue unless the source says it was recognised as revenue |
| Economic impact | Modelled economic activity, jobs or value associated with a platform | Platform sales, employment or revenue |
| ShortFormVideo.co calculation | Arithmetic derived from approved inputs with the formula shown | A figure reported by the platform |
Alphabet reported more than $60 billion in combined YouTube Ads and Subscriptions revenue for 2025. That is product-and-subscription bundle revenue. Alphabet did not publish standalone Shorts revenue. Meta reported $198.759 billion in Family of Apps segment revenue, which includes several products. That segment figure cannot be relabelled as Meta parent-company revenue, Instagram Reels revenue or Facebook Reels revenue.
The validator rejects financial records when the classification, scope and public wording conflict. A product bundle does not become segment revenue, and segment revenue does not become parent-company revenue through enthusiasm.
How are views and engagement defined?
A view is whatever the platform says it is at that point in time. There is no universal short-form video view.
YouTube began counting every Shorts start and replay as a public view on 31 March 2025, with no minimum watch time. It kept the previous measure as Engaged views and uses that separate measure for Partner Program eligibility and Shorts advertising revenue sharing.
Facebook currently counts a Reel view after one millisecond of playtime. LinkedIn organic video needs at least two continuous seconds. LinkedIn paid video also requires at least 50% of the video to be on screen. Instagram counts starts and replays as views, while accounts reached is its estimated unique-account measure.
We therefore do not place raw view totals in one comparison chart and pretend the units match. If definitions differ, the chart says so. If a platform changes its rule, the time series receives a visible methodology break rather than a neat line across two different measures.
Other terms receive the same treatment:
- Members means registered LinkedIn accounts, not monthly active users.
- Monthly active users belongs to the whole product unless the source names a specific format.
- Reach usually estimates unique accounts and should not be reported as views.
- Watch time is cumulative unless the platform defines an average or completion measure.
- Eligible means a rule has been met. It does not guarantee recommendation, monetization or payment.
How often is the data checked?
Freshness depends on what can change.
| Record type | Routine review |
|---|---|
| View definitions, monetization rules and platform policies | Every 30 days |
| Usage, adoption and official algorithm statements | Every 90 days |
| Financial filings and annual results | At the next relevant filing or within 12 months |
| Dated platform changes | When the platform announces a material change |
| Trending audio during the pilot | Weekly |
| Curated video examples | Monthly dead-link and embed check |
Each approved record has a last-verified date and a next-review date. A build fails when an approved visible record is overdue. A page should never look current merely because someone changed the year in its title.
Automated checks can flag a redirect, missing locator, title change or failed request. They cannot decide that a replacement claim means the same thing. That decision remains a manual evidence review.
What happens when a source or statistic changes?
Material corrections follow a visible process:
- The reported problem is checked against the original source and any new official evidence.
- The affected record is paused if the published wording can no longer be defended.
- A corrected or replacement record receives its own verification date and caveat.
- The page revision note explains what changed and when.
- The previous data snapshot remains available where practical.
If the old figure was accurate for its period and a newer figure replaces it, the old record becomes superseded rather than wrong. If the old claim was wrong, the correction says so plainly.
We do not delete a statistic simply because an automated request receives a 403 response. Official investor and help sites often block bots while remaining available to a human reader. The source gets a manual review.
Can the data and graphics be reused?
Yes. Individual facts can be quoted with their reporting period, caveat and original source. A link to the relevant ShortFormVideo.co research page helps readers inspect the maintained record, but it does not replace crediting the primary publisher.
ShortFormVideo.co charts and summary graphics may be reused with a visible link to the research page. Keep the source line and caveat visible. Do not crop a company, segment or product-bundle label off a chart and then present the number as revenue for one format. It has been tried elsewhere. It remains a poor method.
CSV downloads are provided for research and citation. The original source material, platform names and any third-party rights remain subject to their respective owners and terms.
How is GPT Image 2.0 used?
GPT Image 2.0 creates editorial artwork and visual texture. It does not create the statistics.
The artwork follows the ShortFormVideo.co visual system: near-black backgrounds, electric-yellow accents, a text-safe upper area and no generated platform logos or synthetic client faces. Prompts, model, date, dimensions and associated statistic IDs are recorded in the image manifest.
Exact titles, numbers, labels, legends, chart geometry, sources and branding are added programmatically with SVG or Pillow. Those values come from approved records. The final image is checked against the same IDs used for the HTML and CSV.
If the image model draws something that looks like a number, it is decoration until removed. Machines can be very confident typographers and still spell “billion” with artistic freedom.
What are the limits of this research?
The first release does not use private client analytics. Daniel’s historic client performance data is not aggregated into a public benchmark.
Official sources also have limits. Platforms choose what to disclose, change definitions and often publish global or company-wide figures without a product breakdown. We leave those gaps visible. We do not fill them with an unlabeled estimate or a scraper that the platform does not permit.
Platform-funded studies are labelled as such. For example, the TikTok-commissioned Oxford Economics study used a 1,000-business survey, TikTok business-account data and US business-population estimates. Survey responses do not prove causation. Economic-impact models are not revenue. An API field can lag, disappear or require approval, which is recorded as part of the evidence rather than hidden in the technical notes.
A later original report will use a consented survey of qualified professionals. It will not publish until at least 250 responses pass the qualification and methodology rules.
Frequently asked questions
Are all short-form video statistics taken from primary sources?
Material platform claims favour official platform documents, parent-company reports and statutory filings. Academic research or credible third-party estimates can be used when official evidence does not exist, but the source class and estimate status remain visible.
Can view counts be compared across platforms?
Not without checking the definitions first. YouTube Shorts, Instagram Reels, Facebook Reels and LinkedIn video use different play, time and visibility rules, so identical view totals can describe different audience behaviour.
How often do you update the methodology?
The policy changes when a new source type, calculation or correction case exposes a genuine gap. Each update receives a visible date and revision note rather than a quiet rewrite.
What should I do if I find an error?
Send ShortFormVideo.co the statistic’s permanent link, the conflicting source and the relevant section or table. We will review the evidence, pause an indefensible claim and record any material correction.
Does GPT Image 2.0 generate any statistics?
No. GPT Image 2.0 supplies decorative artwork only. Every displayed number and chart position comes from an approved data record and is added programmatically.
