How We Review AI Tools
Our reviews, comparisons and decision tools are built around workflow fit, verifiable evidence and real production economics. Affiliate relationships never add ranking points or guarantee inclusion.
1. Evidence levels
We separate what we directly observe from what a provider publishes. This prevents vendor marketing language from being presented as independent testing.
| Evidence level | What it means | How we describe it |
|---|---|---|
| Direct observation | Behavior or output we can inspect directly in a product, demo, interface or reproducible workflow. | We describe the observed result and its limitations. |
| Official-source verification | Pricing, credits, plan limits, licensing, documentation or policy information published by the provider. | We attribute the fact to the provider and treat volatile data as time-sensitive. |
| Vendor claim | A capability or performance statement that we cannot independently reproduce. | We label it as a vendor claim rather than presenting it as our test result. |
We do not write “we tested” when the evidence does not support that statement.
2. What we evaluate
- Workflow fit: whether the product is suitable for the actual job — cinematic generation, avatar video, faceless content, localization, repurposing, ecommerce ads or another defined use case.
- Output usefulness: whether the output is realistically usable after generation, not merely visually impressive in a demo.
- Control and correction: the ability to revise errors, timing, captions, scenes, voices, brand elements and other production details.
- Production economics: subscription price, credits, generation cost, retries and the cost of reaching an approved output.
- Commercial constraints: export limits, watermarking, commercial-use rights and plan-specific restrictions where these can affect a buying decision.
- Localization: language, dubbing, voice and translation support when international workflows are relevant.
- Operational friction: setup complexity, generation time, workflow handoffs and hidden work that reduces the value of automation.
- Limitations: weaknesses are part of the recommendation, not an afterthought.
3. How our decision tools score products
Interactive tools such as the AI Video Tool Finder use deterministic workflow-fit rules. A reader's answers add or subtract fit points based on the requested workflow. An affiliate relationship contributes zero points.
Where a page presents a ranked shortlist, the ranking should reflect the stated use case and methodology on that page. We do not use a single universal score to pretend that one product is best for every job.
4. Pricing and credit methodology
Credits are not a common currency. Ten credits from one provider can represent a completely different amount of generation from ten credits elsewhere. For that reason, we avoid cross-vendor credit comparisons unless the underlying unit is normalized.
Our AI Video Cost Calculator and Budget Planner focus on production economics. When the data allows it, we model:
This retry factor matters because a cheap first generation can become expensive when several attempts are required before the result is usable.
5. Commercial-use and licensing checks
Commercial rights can differ between free and paid plans and can change over time. Our Commercial Use Rights Checker distinguishes confirmed permissions, plan-dependent permissions and cases where terms remain conditional or ambiguous. It is a research aid, not legal advice.
6. Affiliate independence
AI Video Signal may earn a commission from some outbound links. That commercial relationship is kept separate from the editorial decision:
- Affiliate status does not add points in the Tool Finder or other scoring logic.
- A tool can rank above an affiliate partner when it fits the use case better.
- A tool without a verified tracking link can still be recommended.
- We do not fabricate tracking URLs or describe an unverified partnership as active.
See the current program status and link policy in our Affiliate Disclosure.
7. Freshness and volatile facts
AI products change quickly. Pricing, credit allowances, model names, free-plan limits, export rules and commercial-use terms are especially volatile. We date material checks where practical and direct readers to official provider documentation for purchase-critical facts.
A page being recently reviewed does not mean every vendor will keep the same terms after publication. Readers should confirm current terms before paying.
8. Corrections policy
If we identify a material factual error, stale price, broken source, incorrect affiliate status or misleading wording, we correct the page rather than preserving the old statement for consistency. Material corrections may also require updates to related comparisons, tools or trackers.
Readers and providers can report a factual issue through our contact page. A correction request does not guarantee a change: we verify the underlying evidence first.
9. What we do not do
- We do not invent hands-on tests, testimonials, user counts or performance data.
- We do not rank a product higher because its commission is larger.
- We do not compare raw credit totals as if provider credits were equivalent.
- We do not present commercial-use summaries as legal advice.
- We do not keep an affiliate label when the tracking relationship has not been verified.
Authorship and page ownership
Guides maintained under the AI Video Signal Editorial Team byline use a shared editorial process. That byline does not imply first-hand product testing. The evidence basis is disclosed on analysis pages, and first-hand language is used only when supported.
Sources and corrections
The separate Sources & Corrections Policy documents our source hierarchy, handling of conflicting vendor information and correction process.