How ReelCritix Works
Every rule below is public on purpose. A review aggregator is only worth as much as the method behind it, so the method is the product.
What qualifies as an AI film
A title belongs in the catalogue when at least 50% of its visual content is machine-generated. We measure it shot by shot, weighted by runtime:
| Shot type | Counts as |
|---|---|
| Fully generated shot | 100% |
| Hybrid: a live plate with a generated environment | 50% |
| Camera-captured or hand-drawn | 0% |
| Titles, credits, black frames | Excluded entirely |
Enhancement is not generation. AI upscaling, denoising, rotoscoping or colour work applied to real footage does not make a film an AI film. It is recorded separately as post-production assistance.
Voice, music and performance can be entirely human and the film still qualifies. Those are recorded on their own axes so you can filter on them.
Provenance: nine axes, not one label
Asking "is it AI?" is the wrong question by now. Every title carries a graded record across nine axes (imagery, motion, performance, voice, music, sound, writing, editing and post), each marked as human, AI-assisted, hybrid, AI-generated, not applicable, or not disclosed.
We display "not disclosed" rather than hiding it. A half-filled record that is honest is worth more than a complete one that was guessed, and showing the gap is what gets it filled.
The public label on a film, whether Fully Generated, AI-Animated or Hybrid Production, is computed by us, never chosen by the submitter. Filmmakers describe their process; the system decides what to call it. That removes any incentive to shop for a flattering label.
How provenance gets verified
Confidence is accumulated from independent evidence, strongest first:
| Evidence | Weight |
|---|---|
| Creator-authorised generation history from the model vendor | Highest |
| C2PA / Content Credentials manifest on the master | High |
| SynthID or vendor watermark detected | High |
| Staff audit against the rubric | High |
| Festival submission record | Medium |
| Project file / edit decision list | Medium |
| Synthetic-media detector output | Advisory only |
A detector never decides a public label. Synthetic-media detectors are unreliable in both directions. They route a title to a human reviewer; they do not adjudicate it.
Review integrity
This is the part everything else rests on.
- An unverified account cannot move any score. Not a little. Not at all. An account that costs nothing to create is worth nothing to a manipulator.
- Identity is verified by stacked independent signals, scored on a published points ladder. We store a one-way hash from the identity provider, never your document, number or photo.
- Verified does not mean public. Critics publish under a byline; viewers may use a pseudonym. Verification status is public, the evidence never is.
- Filmmakers cannot review their own work, or work competing against theirs. This is enforced in the database, not by asking nicely.
- Filmmakers can pay for a critic's time, never for a verdict. Through Get Your Film Reviewed, a filmmaker can ask a professional critic to review their film. The critic reviews it as they would any other, a negative review is published like a positive one, and these reviews count toward the score on the same terms as every other.
- Corrections are published. When we remove reviews or restate a score, we say what happened and how the number changed. No silent deletions.
The score
We publish one number, the Critic Score. There is no audience score: a rating from an account we cannot tie to a body of published criticism is the easiest thing on a review site to fake, so we do not collect one.
The Critic Score is the percentage of verified critics who rated a title positive or mixed. Not an average out of 100, and not a weighted mean. Our critics publish prose, not numbers: nobody gave a film a 54, so any 54 we printed would be our invention wearing their byline. A verdict is a reading of what a critic concluded, which is the smallest claim that still supports a score and the only one a reader can check against the review linked beside it.
- A mixed verdict counts toward the score. A mixed review usually credits the film with something real, and counting it with the pans made scores read harsher than the reviews. The positive, mixed and negative counts are shown separately above every score, so the number can always be checked against the verdicts.
- Where a critic gives a number, 5 out of 10 is positive. So is 2.5 out of 5, or anything at 50 percent or above. The number is the critic's, so it settles the verdict without any reading on our part.
- A score is published from three reviews. Before that the title shows TBD with its review count, and every review is still listed and linked on its page. One critic's opinion shown as a red or green percentage reads as the site's verdict; three is where it starts to describe a reception.
- Every admitted critic counts once. No multipliers, no pedigree bonus. All of the judgement sits at admission to the roster, which is a human decision made against published criteria.
- Green at 60% and above, red below it. Two colours, one line. A bare majority is not a recommendation, so 55% is not painted as a good review.
- Scores do not decay. A 2026 verdict stays the 2026 verdict.
What is in the catalogue
1702 titles, built by ingesting juried festivals and curated galleries rather than by waiting for submissions. Everything we can measure, we measure:
- Runtime on every title we can resolve one for, shown under every thumbnail. Runtime also decides what is a short and what is a feature, at a 40-minute line, so the kind label is derived rather than claimed.
- Genres inferred from the synopsis and title where the evidence is strong enough, or assigned by an editor who read the synopsis. Titles with nothing to go on carry no genre rather than a guess. A title can carry more than one, and filmmakers and viewers can correct them.
- The film itself where it is legally embeddable, playing in the page from YouTube or Vimeo. Nothing third-party loads until you press play.
- Festival records, kept separately from any score, which is what orders a chart before the reviews arrive.
No tutorials, no behind-the-scenes reels, no software demos. This is a catalogue of finished work.
Where we are today
The critic roster is being assembled and is small on purpose, so most of the catalogue carries no score yet. Those titles say TBD rather than showing a zero. Provenance across the catalogue is self-declared and awaiting human verification unless a title page says otherwise.
We would rather launch with an honest empty state than a fabricated number. If that sounds like a place you want to write for, see the roster criteria.