How we evaluate AI clippers: our methodology

VodClipsLab has not yet tested any AI clipper on gaming footage. As of 8 November 2026, every review, comparison and pricing page on this site is built from published documentation, pricing pages and reports. This page explains how those pages are made, and how the gaming VOD tests will work once footage is licensed and paid tool access is approved.

What the labels on our pages mean

Every page on this site falls into one of two evidence states, and the label tells you what the page can and cannot support.

Label on the page	What it means	What you can rely on it for
Source-based (“How this page was built”)	Written from vendor documentation, pricing pages, platform help, published method references and dated reports	Documented features, limits, prices and billing rules, or a published way to measure something, as of the date shown
Measured (none yet)	Would report our own results from a recorded test	Nothing yet. No page on the site holds this label

Method guides explain how to evaluate a tool on your own streams. They are built from published references and report no VodClipsLab results. The tests section lists them.

How source-based pages are made

Source-based pages follow the same steps, so you can see where a claim came from and how far it goes.

Primary sources first. Prices, limits and policies come from the vendor’s or platform’s own page wherever one can be read. Anything taken from a third party is labelled as such.
Attribution at the level the source supports. If a vendor says it supports Twitch links, we report that it says so. We do not turn it into a claim that the tool handles a Twitch VOD well.
Dates on anything that can change. Prices, credit rules and access status carry the date we read them.
Conflicts stay visible. When a vendor’s help article and pricing page disagree, we show both, after checking that the difference is not just annual against monthly billing, API against web limits, or different plans.
Gaps are stated. If a page does not say something, we write that it is not stated. We do not fill it in from inference.
Reasoning is labelled. We sometimes explain why gameplay should be harder for a tool, for example long stretches without speech. That is reasoning about what to check, not a result.

Cost figures are arithmetic from published prices and billing rules, with the inputs shown. They price what you put in, not what you get out, as the cost per VOD hour page explains.

The planned gaming VOD test protocol

The planned protocol runs each tool on the same footage under recorded settings and scores the results without knowing which tool made them. None of it has started. It depends on two things that are not yet in place: footage with documented permission under our test footage policy, and approved paid tool access.

Footage. A fixed set of VODs across six genres, from 2 to 6 hours long, each with its permission on record. The genres are FPS, battle royale, MOBA, low-action sandbox, horror with facecam, and a Just Chatting control with no gameplay. The guides in the tests section cover what to look for in each kind of stream.

Settings. For every run we log the tool, plan, version or date, every setting that was changed, the upload route and the credits charged. Raw outputs are archived so a score can be rechecked.

Clip scoring. The top 10 clips each tool proposes are scored by a rater who does not know which tool produced them, using a published rubric. The rubric covers:

whether the moment is worth posting
whether the clip starts early enough to make sense and ends cleanly
whether facecam and gameplay are framed correctly in vertical
whether captions or crops cover important on-screen information

Rubric weights are planned to differ by genre and to be published before scoring starts. Separate guides explain how to check acceptance rate, facecam layouts and captions against the HUD on your own clips.

Caption accuracy. Word error rate on hand-transcribed, double-checked samples, chosen to include game audio over speech, using the method and sample-size limits on the published caption accuracy page. Word error rate is substitutions, deletions and insertions divided by the number of words in the reference transcript. See caption accuracy on gameplay audio.

Time and cost. Wall-clock processing time and credits used, by VOD length. See how to measure processing time yourself.

A manual baseline. The same VOD is also cut by hand with free tools, with minutes spent and usable clips recorded. Without that, a “time saved” claim has nothing to be measured against.

Published data. Scores, timestamps and settings are planned for release as a test dataset, so that others can check or challenge them. No dataset exists. A log template shows how to record your own evaluations.

Who runs a test

Test pages will name the person who ran the test and the person who reviewed the rubric. Nobody has been appointed yet, because no test has been scheduled. We will not publish under an invented name or credential.

Independence

No vendor can preview, approve or pay for coverage. No page carries a live affiliate link today, and the affiliate disclosure explains how any future links would work. An affiliate relationship does not change a verdict and will not change a score. The FTC’s rule on consumer reviews and testimonials, in effect since 21 October 2024, bars fake reviews, and we hold ourselves to the same line on invented test results.

How pages are kept current

Source-based pages are re-read against their sources on a set schedule, and changes are logged in public.

Prices and credit rules: a monthly check, recorded in the price change log.
Twitch Auto Clips: a weekly status check on the Twitch Auto Clips page, because access can change quickly.
Tests, once they exist: a quarterly re-run for the main tools, and a re-run when a tool ships a major model or credit change.
Mistakes: corrected and listed in the corrections and update log.

Each page shows the date it was last checked, and that date is the only claim we make about how current it is.