How We Calculate Game Scores
A transparent look at our methodology for aggregating streamer opinions into a single, trustworthy score.
Traditional game reviews come from professional journalists who often have different priorities than everyday gamers. At StreamerVerdict, we believe the content creators you already watch and trust provide more relevant opinions.
Our scoring system aggregates verdicts from trusted streamers to give you an honest, community-driven score that reflects what real players think.
1. Curated Streamer Selection
We monitor a curated list of trusted YouTube creators — established reviewers known for honest, in-depth game analysis, not hype or drama. Every channel is added deliberately; nothing is pulled in automatically.
What we look for:
- A consistent track record of game review content
- Genuine expertise in the genres they cover
- Reviews that reach a clear verdict, not just reaction
We screen each channel before adding it, and we re-screen when a channel's content changes. Channels that turn out to be mostly let's-plays, news, or hardware coverage get filtered out.
2. AI-Powered Analysis
When a streamer posts a review, we analyze the video transcript using AI to extract:
- Their verdict — the sentence where they actually deliver it
- A score, where there is one (0-100) — see below for what happens when there isn't
- Sentiment (positive, mixed, negative) — The overall tone of the review
- Key pros and cons mentioned — What the streamer loved or hated
- Notable quotes and highlights — Direct quotes that capture their opinion
Stated scores and derived scores
Plenty of excellent reviewers never say a number. "This is a must play" is a clear verdict; "8 out of 10" is a clear verdict. If we only counted the second kind, whole channels would contribute nothing to a game's score — and the channels that don't use numbers aren't a random sample. They skew toward commentary, retrospectives, and non-English reviewers. Ignoring them isn't neutrality; it's a quiet bias about whose opinion counts.
So we do both, and we label which is which:
- Stated — the creator said a number. We normalize the scale ("8/10" becomes 80) and use it. If someone states a figure outside the scale, we bring it back inside rather than throw the review away, and we keep the original so the adjustment is on the record.
- Derived — the creator gave a verdict but no number. We read their verdict into one of five classes and convert it with the table below.
Derived scores are labelled wherever they appear: the review shows the verdict we read, the band it falls in, and the creator's own words we read it from. A score shown without that label is one the creator stated. A derived score is our reading of what someone said, and you're entitled to know when you're looking at one.
The conversion table
This is the whole table. It's the same for every creator — there is no per-channel version, because "must play" should not be worth more on one channel than another.
| What they said | Score | Range |
|---|---|---|
| Must play | 90 | 85–95 |
| Recommend | 77 | 70–84 |
| Mixed | 60 | 50–69 |
| Skip | 40 | 30–49 |
| Avoid | 20 | 10–29 |
Two things this table deliberately does not do:
No derived score is ever 0 or 100. Those are the strongest statements the scale can make, and we reserve them for creators who actually make them. Our interpretation of someone's words should never outrank their own number.
"Wait" is not one of the five. When a creator says to wait, the reason is the verdict: "wait for a sale" means the game is good and the price isn't, so it reads as recommend; "wait until it's patched" means the game isn't finished, so it reads as skip. A "wait" with no reason given produces no score at all — those two readings point in opposite directions, and we would rather show you nothing than pick one by coin flip.
The table is versioned, and each derived score records the version it came from. If we change the table, we don't sweep back through old reviews restating what a creator was taken to have meant two years ago — a published score stays put, and stays explainable against the version that produced it. The exception is honest to state: if we re-analyse a video from scratch, the new reading uses the current table, because at that point it's a new reading.
3. Verification
Let's be straight about this: StreamerVerdict is a small independent project. We can't hand-check every review, and we'd rather tell you how it actually works than imply a review team we don't have.
So most reviews are verified automatically, and humans handle the ones that need judgment.
Automated checks. After the AI extracts a verdict, separate checks try to disprove it against the original transcript:
- Did the creator actually say the sentence we're quoting as their verdict?
- Does the game name we extracted actually appear in what they said or wrote — and does it match a real game?
- If we recorded a score they stated, did they really say that number, and were they talking about the whole game rather than one part of it?
- If we derived the score from their verdict instead, does the verdict sentence appear in the transcript, and does re-running the conversion on it produce the same number we published?
A claim that can't be traced back to the source doesn't publish.
Human review. This is the exception, not the routine. When the automated checks genuinely can't resolve something — an ambiguous game title that could be two different releases, a number whose meaning isn't clear, a video that may not be a review at all — the review doesn't publish and doesn't get guessed at. It goes into a queue and waits there until a person works through it.
Nothing in that queue appears on the site in the meantime. A review we're unsure about is simply absent rather than shown with a number we can't stand behind.
This is the honest trade-off: automated verification is what makes covering hundreds of channels possible, and human attention goes where it actually changes the outcome.
4. Score Aggregation
The final score is a recency-weighted average of streamer scores. Newer reviews count for more, because a game at launch and the same game two years later are often not the same game — patches, content updates and price changes all move the needle.
- Reviews from the last 30 days carry full weight
- Older reviews fade gradually, losing half their influence roughly every 6 months
We don't weight by subscriber count or views — every trusted voice counts equally, so popular creators can't dominate the score. Time is the only thing we weight by.
We also show how many reviews a score is based on. 78 (n=2) is us being honest that the data is thin, rather than dressing up two opinions as a verdict.
Category Breakdowns
Beyond the global score, we also calculate category-specific scores. If you primarily care about what "Souls-like experts" or "RPG enthusiasts" think, you can see scores filtered by streamer categories.
Each streamer is tagged with their areas of expertise, allowing for personalized score filtering.
What Makes Us Different
| Traditional Reviews | StreamerVerdict |
|---|---|
| Single journalist opinion | Aggregated from multiple creators |
| May have publisher relationships | No ads or sponsorships |
| Written reviews only | Video-based, captures nuance |
| Static scores | Updated as new reviews come in |
Transparency First
We believe in complete transparency:
- You can see every individual streamer's verdict, their score, and whether that score is one they stated or one we derived from their verdict
- The conversion table above is published in full and applies to every creator equally
- Our methodology is public (you're reading it!)
- We don't hide low scores or cherry-pick reviews
- Score changes over time are tracked
Found Something Wrong? Tell Us
No system that reads thousands of hours of video gets everything right. So there's a report button on every review and every game page.
If a score, summary, timestamp or game match looks wrong to you, report it. Reports go into a queue we actually read, and we re-check the review against the original video. If we got it wrong, we fix or remove it.
If you're the creator: your reviews are here because we think your opinion is worth surfacing — but it's your work, and it's your call. Email us at derekxinzhewang@gmail.com and we'll take your channel and its reviews down. No argument, no process to fight through.
Questions?
Have questions about our methodology, or want to suggest a creator we should be covering? Email us at derekxinzhewang@gmail.com — we're always looking to improve.
Ready to see honest scores? Browse Games →