Home / Methodology
Evidence policy · updated Aug 18, 2026

We separate product facts from buyer experience.

Our guides are research syntheses. We use official sources to establish what a product sells, neutral review platforms to identify recurring buyer signals, and attributed independent walkthroughs to understand workflow and output.

EDMaintained by AI Video Ad Generators Editorial Desk·Updated August 18, 2026
Current testing status: we do not claim comprehensive hands-on testing. Screenshots captured by our editor are labelled as primary captures; third-party stills remain visibly attributed and are used only for what the shown interface supports.
Aofficial facts and logged-in captures
Bneutral review platforms with sample context
Cindependent, attributed workflow evidence
Dcommunity posts for issue discovery only
Source hierarchy

What each source is allowed to establish.

A · PrimaryOfficial pricing, help centers, policies and logged-in capturesFacts, limits, terms, interface state
B · NeutralG2, Capterra, Trustpilot and Product HuntRecurring praise, complaints and risk signals
C · IndependentNon-competing creators and publishers showing the workflowAttributed experience and output context
D · DiscoveryForums and public social postsQuestions to verify, never standalone proof
Bias control

Sources excluded from our verdict.

Competitor-owned comparisons

An AI video vendor may help us discover a feature or question, but its comparison of a rival does not support our recommendation.

Vendor demo verdicts

Vendor videos can illustrate an interface or stated workflow. They are labelled as vendor material and never treated as independent testing.

Unattributed claims

We do not use copied review text, anonymous statistics without a visible sample, or screenshots without enough context to identify the plan and date.

Single anecdotes

One post can reveal a question worth checking. It does not establish typical quality, support, billing, or reliability.

Review-platform rules

Stars are context, not a leaderboard.

Different platforms have different audiences, invitation systems, moderation, and review prompts. A 4.8 on G2 and a 3.6 on Trustpilot are not controlled measurements of the same population.

  • Show the rating, review count, capture date, and platform together.
  • Disclose claimed profiles, paid platform features, or invitation context when visible.
  • Use repeated themes, not isolated quotes, to describe buyer experience.
  • Keep review-platform snapshots separate instead of averaging them into a fabricated score.
  • Recheck changeable facts before major content updates.
Visual evidence rules

A screenshot inherits the limits of its source.

Primary capture

Captured from a logged-in account by our editor. It can establish the interface, displayed price, plan, date, and visible terms, but not normal output quality.

Creator walkthrough still

Attributed and linked to the original video and time point. It can illustrate workflow and controls, but it remains a third-party account state rather than our test.

Creator-reported result

Campaign numbers remain attributed and sit beside missing methodology. They can reveal a testing pattern, but never become a typical-result promise or an editorial performance score.

Decision model

We evaluate the job, not the longest feature list.

01Starting inputScript, product URL, footage, or existing video
02Core workflowActor generation, ad assembly, or localization
03Real costCredits, retries, editing time, and unused allowance
04Output riskConsistency, controllability, and failed generations
05Exit riskRefund, cancellation, rollover, and overages
Primary metric

Cost per ad you would actually launch. Monthly subscription price and generated clip count are inputs; publishable output is the buying outcome.

Corrections

Every commercial fact has a shelf life.

Pages display a last-checked date and link to their source ledger. When pricing, credits, policies, or review counts change, we update the affected claim and preserve the distinction between the captured snapshot and the current offer.

Readers and vendors can report a factual error with the exact page, claim, and primary source. Commercial relationships do not determine whether a correction is accepted.