AI content scoring is a method of grading a draft against fixed dimensions and returning a number. ForgeRank's system scores four dimensions — information gain, specificity, source compliance, and argument depth — then averages them into one headline figure. The score summarizes four separate judgments. It is not a single impression, and it does not measure originality against the whole web.
Why This Matters
A score of 7.0 is a pass line, and plenty of drafts land below it. The ForgeRank published dataset (https://forgerankai.com/static/data/ai-content-quality-29-drafts.json) shows a distribution that clusters at the bottom: 4.0 ten times, 4.8 ten times, 7.0 three times, 7.2 four times, 7.8 once, and 8.5 once. Nothing landed between 5.0 and 6.9.
That gap tells you something about how the scoring works. Drafts either fail on the fundamentals or clear them and score in the sevens and eights. There is no comfortable middle.
The dataset covers 29 drafts and 2,839 claims. Of those claims, 51 were judged to need a source, and 38 had none attached. That is the kind of finding a headline score hides. A draft can score well on argument depth and still ship 38 unsourced claims.
How Is the Headline Score Calculated?
The headline score is a weighted average, not a plain mean. According to the ForgeRank methodology page, the four dimensions are weighted 30 percent information gain, 25 percent specificity, 25 percent source compliance, and 20 percent argument depth.
Weighting matters because it changes what a draft has to do to pass. Information gain carries the most weight, so a draft that restates what already exists online starts at a disadvantage no amount of polish fixes. Source compliance and specificity together account for half the score, which means unsourced and vague claims drag a draft down faster than weak argument structure does.
The boundary is that the weights are fixed. A draft strong in argument depth cannot compensate for a 4.0 in information gain, because 20 percent cannot outvote 30 percent.
Why Does Each Dimension Get Scored Three Times?
Each dimension is scored as the median of three samples, according to the ForgeRank methodology page. The median exists because a single sample of a judgment call is unstable, and three samples let the middle value absorb an outlier.
The mechanism is straightforward. If a scorer rates information gain a 7, then a 4, then a 7, the median is 7. One low sample does not sink the dimension. Because the headline score is the average of the four dimensions, that stability propagates upward: one noisy dimension reading cannot swing the total as far as a single sample would.
The consequence for you is that a score reflects a repeatable judgment rather than one pass. The boundary: the median protects against a single outlier, not against a scorer who reads the same draft the same wrong way three times.
What Do the Tier Anchors Mean?
Dimensions are judged in tiers, with 4, 7, 8, and 9 all observed, according to the ForgeRank methodology page. That is why the headline score moves in steps rather than sliding continuously.
A tier anchor is a fixed reference point a scorer compares the draft against. If information gain is judged at the 7 tier, it does not drift to 7.1 or 7.3 based on feel. The score snaps to the anchor. Because the headline is the average of four tiered dimensions, the total lands on a limited set of values — which is exactly what the published dataset shows, with scores at 4.0, 4.8, 7.0, 7.2, 7.8, and 8.5 and nothing between.
The writer's takeaway: you are trying to move a draft up a tier, not nudge a decimal. The boundary is that tier boundaries are coarse. A draft sitting just under the 7 tier and one sitting just over it can read almost identically to a human.
What Does Each Dimension Actually Measure?
Information gain measures whether the draft adds something a reader could not get elsewhere. It matters because, per the ForgeRank methodology page, it carries 30 percent of the weight — the single largest share. The consequence is that a well-written draft with nothing new still fails. The boundary: gain is judged against what the scorer can see, not against the entire web.
Specificity measures whether claims carry a number, a named entity, or a concrete instance. It matters because vague claims cannot be checked, and 25 percent of the score rides on it. The consequence is that abstract sentences cost you directly. The boundary: a specific claim that is false scores worse than a vague one that is true.
Source compliance measures whether claims that need a source have one attached. It matters because the dataset recorded 51 claims across 29 drafts judged to need a source, with 38 carrying none — a failure mode that repeats. The consequence is that unsourced claims are the most common fixable defect. The boundary: not every claim needs a citation, only those making a factual assertion a reader would want to verify.
Argument depth measures whether the draft explains mechanism and edge cases rather than listing points. It matters because it carries 20 percent, the smallest share. The consequence is that depth alone will not carry a draft. The boundary: depth stops helping once the other three dimensions are already failing.
What Most People Get Wrong About AI Content Scoring
The common belief is that a high score predicts ranking. The evidence supports a narrower claim: a score measures four internal qualities of a draft, and Google's own quality signals are separate.
Google Search Central's March 2024 core update aimed to reduce low-quality, unoriginal content in search results by 40 percent. That is a search-results outcome, not a draft-grading outcome. A 7.2 on a ForgeRank score says nothing about whether Google will rank the page, because the score never checks the live index, backlinks, or query intent.
The dataset itself shows why the belief persists. The ten 4.8 drafts score 4.0/4.0/7.0/4.0, which is the ten 4.0 drafts plus one source-compliance tier, and the four 7.2 drafts score 7.0/7.0/8.0/7.0, which is the three 7.0 drafts plus the same tier. A writer sees a number climb after one narrow fix and assumes the number tracks quality. It tracks the four dimensions. Ranking is a different system with different inputs.
Key Takeaways
- The headline score is a weighted average: 30 percent information gain, 25 percent specificity, 25 percent source compliance, 20 percent argument depth, per the ForgeRank methodology page.
- Each dimension is the median of three samples, which stabilizes the reading against one bad pass.
- Scores move in tiers (4, 7, 8, 9 observed), so the total lands on a limited set of values, not a smooth scale.
- The published dataset's 29 drafts cluster at 4.0 and 4.8, with nothing between 5.0 and 6.9.
- A score does not measure originality against the whole web, does not predict rankings, and drifts between runs.
Frequently Asked Questions
What is a good AI content score?
A 7.0 is the pass line in the ForgeRank dataset, and scores of 7.2, 7.8, and 8.5 appear above it. A 4.0 or 4.8 signals a fundamentals failure. The number alone will not tell you which dimension failed.
Does a high AI content score mean Google will rank the page?
No. The score grades four dimensions of a draft. Google Search Central's March 2024 core update targeted low-quality, unoriginal content in search results, a separate system. A high score and a ranking are independent outcomes.
Why does the same draft score differently on two runs?
Each dimension is the median of three samples, so a draft near a tier boundary can land on either side. A draft sitting at the 7 tier edge can read as 7.0 on one run and 7.2 on another without any edit.
How do I raise a score fastest?
Attach sources to claims that need them. The dataset shows 38 of 51 source-required claims had none. Source compliance is 25 percent of the score, and fixing it is mechanical rather than structural.
What can't an AI content score measure?
It cannot measure originality against the whole web, because it only sees the draft in front of it. It cannot predict rankings. And it cannot tell you whether the argument is true, only whether it is sourced and specific.
Google Search Central's guidance on AI-generated content draws the same line from the other side. Its spam policies treat using automation to produce "many pages" without adding value as spam, and its systems reward "original, high-quality, people-first content." A scorer can flag the absence of value markers. It cannot confirm the presence of a perspective nobody has published. The 8.5 draft in the published dataset scored 8.0/9.0/9.0/8.0, a profile that says the draft cleared three dimensions at a high tier and says nothing about whether its argument is correct.
The Score Is a Starting Point, Not a Verdict
Four weighted dimensions, each sampled three times, produce a number that tells you where a draft stands on information gain, specificity, source compliance, and argument depth. That number will not tell you if the page ranks, and it will not tell you if the idea is original. Treat it as a checklist with a score attached.
Open your most recent draft tonight and count the factual claims that need a source. The dataset found 38 unsourced claims across 29 drafts — check whether yours is one of them before you publish.