Which film rating site should you trust?

Interstellar looks like an obvious audience favourite in the records below: 8.7 on IMDb and 4.46 out of five on Letterboxd. Its Metascore is 74. That is not a small difference after rescaling, and it cannot be explained by saying one side watched the film correctly.

The film offers something unusually powerful to a viewer who accepts its emotional bargain: an immense physical journey organised around a parent's absence and a child's experience of being left. It also gives a sceptical viewer plenty to resist, including dialogue that can explain a feeling or a concept more insistently than the scene needs. Admiring the scale and finding the explanation heavy are compatible responses.

That is the point at which a ratings site becomes useful—or starts to mislead. Do you want to know how many included critics approved, how strongly a selected group scored the work, or how an audience that chose to rate it responded? Those are different questions. The biggest number on a page is not necessarily the answer to yours.

My practical answer is this: use Rotten Tomatoes for the breadth of critical approval, Metacritic for the strength of selected critics' responses, and IMDb or Letterboxd for recorded audience enthusiasm. Use the individual reviews when the decision matters to you. MEANRANK's job is to put the recorded measures together and keep the differences available, not to declare one population infallible.

Here is how the choice works in practice.

Four familiar films, in the sites' own units

This comparison was calculated from the saved catalogue on 30 October 2026. The source ratings may have been collected at different times; the date does not mean all four sites were checked together. The columns retain their native scales because a Tomatometer percentage and an average out of five describe different things, however neatly they fit beside one another.

The examples are not a random sample. They are useful because they make different patterns visible. Interstellar's audience averages are especially high alongside its Metascore. Paddington 2 has both broad critic approval and a high Metascore, while its audience averages express enthusiasm in a different numerical form. The Shining shows how a familiar reputation can coexist with a less elevated figure in one selected critical measure.

None of those patterns tells us why each individual voted. They tell us where opening the reviews may be more useful than continuing to compare the size of the printed numbers.

Rotten Tomatoes answers a narrower question than its reputation suggests

A 100% Tomatometer can arise from ten mildly positive reviews. An 80% can arise from eight ecstatic recommendations and two dismissals. The first result has broader approval within the reviews counted. The second may involve more intense enthusiasm.

That distinction is not a defect in the arithmetic. It is what the measure is for. The mistake occurs when an approval percentage is read as an average grade for excellence.

Paddington 2 makes the difference easy to see without inventing an adversarial debate. A critic can admire the film's visual organisation, its generosity and the construction of an elaborate comic sequence without saying it is nearly a perfect film in every possible sense. A very high share of positive notices can coexist with a range of judgements inside the positive group.

A useful check is the language inside the positive group. One review may recommend a film as an undemanding evening; another may call its construction exceptional. Both can help the Tomatometer rise while offering quite different reasons to choose it. Where an average review rating is also available, it can add information about intensity. The individual arguments still tell you whether that intensity concerns something you value.

The audience percentage is another approval measure from a different group. Ticket verification can change who qualifies for a particular tally; it cannot establish that the eligible voter shares your taste. Check which audience figure is displayed before comparing records collected under different conditions.

Metacritic preserves more intensity, with its own choices

Metacritic converts selected professional reviews into a weighted Metascore. That can be useful when a simple positive-or-negative division would conceal how strongly people responded.

The score also carries editorial choices: the included publications, the conversion of unscored reviews and the weights applied to critics. Metacritic says its movie scores are additionally normalised. Rescaling a review and publishing a Metascore are therefore not simply the same arithmetic step. Nor does an average preserve the argument inside an individual review. One critic may find a film exquisitely constructed and emotionally empty; another may find the construction awkward and the emotional effect overwhelming.

The number between them is not a newly discovered third response that resolves the dispute. It is a summary of the included ones.

Interstellar is a good reason to use that summary as an invitation rather than a barrier. If a critical objection concerns the insistence of the dialogue, you can ask whether that is something that generally troubles you. If the attraction is the joining of intimate feeling to enormous scale, you can decide whether the apparent excess may be part of what you want.

A middling or moderately positive Metascore can therefore be more informative than a strong generic recommendation. The distribution and the reasons may reveal a film with a clear appeal and a clear cost.

IMDb's size is useful, but size does not remove selection

IMDb publishes a weighted user average rather than a raw arithmetic mean of every submitted vote. The visible vote count helps distinguish a score supported by a large body of activity from one based on relatively little. It is not a record of how many people watched the film, and IMDb does not disclose enough of its weighting method to let the public reconstruct the displayed number from that tally alone.

That distinction matters. It should not be expanded into the claim that enough votes eventually become a sample of everybody. People still choose which films to watch, which to finish, which to rate and whether to revise a rating later.

An acclaimed older film may mostly be encountered through a recommendation that already tells the viewer it is important. A specialist release may attract people actively seeking its particular pleasures. A widely promoted disappointment may receive ratings from many viewers who would not have chosen it under a more accurate description. Those mechanisms can produce different voting populations without anybody participating dishonestly.

The Shining is a useful reminder that a rating is also a record of an encounter made at some point in a film's public life. Today's user and a critic represented in an older selection may not have approached it with the same expectations or the same surrounding reputation. The four figures alone cannot reconstruct that history.

The individual reviews often supply what the average cannot: whether someone is responding to pacing, performance, atmosphere, the adaptation or the difference between a frightening image and an emotionally convincing person.

Letterboxd is especially useful when the average stops being the main attraction

Letterboxd also publishes a weighted film average. Its diary, list and following features offer another way to make the information useful: observe a particular person's judgement across several films.

Try a reviewer on a film whose choices you know well. If their objection to Interstellar concerns the dialogue, ask whether they identify the moment where an explanation weakens an exchange. If their praise for The Shining concerns distance, ask what that distance lets them notice. A useful writer makes the reason testable even when the conclusion differs from yours. Matching your star rating without explaining it tells you much less about the next recommendation.

Someone who complains that a scene withholds explanation may be accurately identifying the very quality you value. Someone who praises a film's relentless energy may be warning you away from an exhausting evening without intending to. The reason makes a recommendation transferable in a way that an isolated score is not.

The danger is replacing the imagined wisdom of a large crowd with the equally untested prestige of a smaller circle. A stylish review or a confident list is still a judgement to examine. Familiarity with a writer's habits becomes valuable when it helps you interpret the judgement, not when it makes disagreement embarrassing.

What happens when all four sites are tested on the same films?

A comparison of each site's overall average can be unfair before any arithmetic is done. The sites may hold different films. A collection with more difficult-to-find festival titles is not directly comparable to one with a different mixture of popular releases.

We therefore used 6,507 films present in all four of these catalogue measures. Each needed at least 1,000 IMDb votes, 1,000 Letterboxd ratings, 40 Rotten Tomatoes critic reviews and 10 Metacritic reviews. IMDb and Letterboxd were rescaled to 0–100 for this calculation.

For each source, we calculated its correlation with the mean of the other three. The source being tested was left out of its own benchmark.

In this matched sample, Metacritic has the strongest correlation, at 0.91. That result is useful, but it answers a restricted question: which measure moves most similarly to the other three within these films?

It does not identify a true score against which the others have been tested. The benchmark is another construction from opinions. The populations may overlap, and the Tomatometer still has a different meaning from a graded average.

Correlation also concerns movement rather than equality. Two measures can rank films similarly while one is consistently higher. That is why the mean column and the correlation column should not be used interchangeably, and why the lower mean is not proof of more discriminating taste.

The difference between the leading correlations is small enough that the rounded table should not become a league title. The exact comparison also depends on the reference we chose: each source is tested against the mean of the other three. Changing that reference, the films or the eligibility floors could change the order. The calculation is reproducible; the choice of what would count as trustworthy remains an argument.

What MEANRANK should help you do

The immediate benefit of aggregation is practical. You can see the source ratings in one place instead of remembering one flattering number from a search result and treating it as the whole reception.

MEANRANK puts the qualifying recorded measures on a numerical scale and shows their mean. The title page keeps the source breakdown available, including threshold information. That lets you distinguish a combined score supported by several measures from one formed from fewer qualifying records.

The mean is not calibrated to remove each source's usual generosity or severity. A platform publishing critic and audience measures can contribute more than one component, and a person can use several audience sites. Nor does a million-vote component get a million times the weight of another eligible measure. The calculation gives each included rating measure equal weight after rescaling; the counts help determine eligibility, rather than creating one pooled ballot.

There is another qualification worth knowing when reading the breakdown: a missing count is treated as unknown, not as zero. A stored rating with no count can still enter the mean; a known count below that source's floor is excluded. A total formed from several measures is therefore not a claim that every component has an equally well-documented sample.

That is why two films with similar MEANRANK scores may deserve very different follow-up. One may have relatively close components. Another may have a high and a low measure that the mean brings towards the middle. In the second case, the disagreement is often the useful thing to investigate.

A combined score helps locate the decision. It should not conceal the information that makes the decision yours.

The site I would trust depends on what I need from it

For a quick check that a film was broadly well received by the included critics, I would look at Rotten Tomatoes and its review count. For the intensity of selected professional responses, I would look at Metacritic and the reviews behind a split. For audience enthusiasm, I would compare IMDb and Letterboxd, with their counts and individual reactions.

For a film I am especially interested in, I would not let any one of those checks make the decision alone. I would read a positive review and an intelligent negative one. The disagreement often names the choice more accurately than the average.

Interstellar does not become less ambitious because someone dislikes its explanations. Paddington 2 does not require a claim of numerical perfection to justify its comic craft. The Shining does not have to persuade every viewer that its distance is emotionally productive.

The most trustworthy review is one whose account of the film gives you something specific to test. The best ratings site is the one you are using for a question its numbers can actually answer.