An AI article quality checker scores a draft across separate dimensions and reports one headline number. That number is an average, so two drafts can land on the same score for completely different reasons. Reading which dimension dragged the score down tells you what to fix; reading the headline alone tells you nothing actionable.
Why This Matters
The ForgeRank published dataset (https://forgerankai.com/static/data/ai-content-quality-29-drafts.json) scored 29 AI-assisted drafts before publishing. Twenty of them landed at or below 4.8. None landed between 5.0 and 6.9.
That gap matters because the drafts below 4.8 weren't failing on one shared flaw. They were failing on different dimensions, and the headline number hid which one.
A writer who sees "4.8" and rewrites the whole piece burns a day. A writer who sees "4.8, information gain tier 4, source compliance tier 7" knows exactly which section to rebuild. The score is a diagnostic, and diagnostics only work when you read the parts.
What Do the Four Dimensions Actually Measure?
ForgeRank's methodology page weights the headline score 30 percent information gain, 25 percent specificity, 25 percent source compliance and 20 percent argument depth. The headline is a weighted average, not a plain mean, so a weak information-gain tier costs more than a weak argument-depth tier.
Information gain measures whether the draft adds something the existing top results don't have. It carries the heaviest weight because a page that only restates other pages gives a reader no reason to stay. The consequence for you: a beautifully written draft that summarizes existing coverage caps out low no matter how clean the prose is. The boundary is a genuinely new angle — a first-party dataset, a named failure mode, a decision rule tied to the reader's situation — which lifts the tier even when the writing is plain.
Why Does Specificity Move the Score So Much?
Specificity measures whether claims carry a number, a named entity, or a concrete instance. It matters because a reader scanning for a fact can't verify "many teams struggle with this," and an extractor can't quote it either. Write an abstraction and you lose the citation; write "twenty of twenty-nine drafts scored at or below 4.8" and the sentence survives on its own.
The consequence for you is that specificity is the cheapest dimension to fix. Swap "most drafts fail" for the actual count. The boundary: a specific number with no source attached fails source compliance, so the two dimensions pull against each other and you have to satisfy both.
Can Two Drafts Share a Score and Need Different Fixes?
Yes. The published dataset shows it directly. Every 4.8 draft scored 4.0 on information gain, 4.0 on specificity, 7.0 on source compliance and 4.0 on argument depth. Every 7.2 draft scored 7.0, 7.0, 8.0 and 7.0. Same neighborhood, different shape.
| Draft score | Info gain | Specificity | Source compliance | Argument depth |
|---|---|---|---|---|
| 4.8 | 4.0 | 4.0 | 7.0 | 4.0 |
| 7.2 | 7.0 | 7.0 | 8.0 | 7.0 |
The 4.8 draft has one strong dimension and three weak ones. The fix is broad: add sourcing depth, sharpen claims, build the argument. The 7.2 draft is balanced and slightly behind on sourcing. The fix is narrow: tighten attribution and leave the structure alone.
Apply the wrong fix and you rewrite a draft that only needed one pass. ForgeRank's tier structure explains why the scores move in steps: dimensions are judged in tiers, so 4, 7, 8 and 9 all appear in the data, and each step in the distribution is a single dimension moving.
What Does the Checker Refuse to Claim?
A checker that scores phrasing separately from substance refuses to say a high score means the draft reads as human. Eight of the nine drafts scoring 7.0 or above still received a heavy-AI-phrasing verdict. The only draft whose phrasing came back clean scored 4.0.
That split is the tool's most honest feature. It tells you a draft can be well-argued and still sound generated, and it tells you a draft can sound natural and still carry nothing worth citing. The boundary: phrasing detection is a separate axis, so a clean phrasing verdict never implies the substance passed, and a high headline score never implies the prose reads clean.
What Most People Get Wrong About Quality Scores
The common belief is that a higher score means a better article. The evidence supports something narrower: a higher score means more of the four dimensions cleared their tier, and the dimensions can move independently.
Tabulating the tiers behind every row shows each step in the distribution is one dimension moving. All ten 4.0 drafts scored 4.0 across all four dimensions. The three 7.0 drafts sat flat at 7.0. The single 8.5 draft scored 8.0, 9.0, 9.0 and 8.0 — no dimension at the ceiling.
The belief persists because a single number is easier to report than a four-part profile. Treat the headline as a summary of four judgments and the fixes become obvious.
Three Questions to Ask Any Checker
Which dimension is weakest? A 4.8 with argument depth at 4.0 needs structural work; a 4.8 with source compliance at 4.0 needs attribution work. Ask for the per-dimension tiers before you touch the draft.
How much does the score move between runs? ForgeRank's method scores each dimension as the median of three samples and averages the four, which is why the headline moves in tier steps rather than decimals. A tool that reports a different score on an unchanged draft is measuring noise.
What does the tool refuse to claim? ForgeRank separates phrasing from substance, which is why eight of nine drafts above 7.0 still carried a heavy-AI-phrasing verdict. A checker that folds phrasing into the headline hides the split you need.
Key Takeaways
- A headline quality 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.
- Two drafts at 4.8 and 7.2 need different fixes because their dimension tiers differ, not because one is "better written."
- Twenty of twenty-nine drafts in the ForgeRank published dataset scored at or below 4.8, with nothing between 5.0 and 6.9.
- Phrasing and substance are separate axes: eight of nine drafts above 7.0 still read as heavy-AI, and the one clean-phrasing draft scored 4.0.
- Ask for per-dimension tiers, not the headline, before deciding what to rewrite.
Frequently Asked Questions
What is an AI article quality checker?
It's a tool that scores a draft across separate dimensions — information gain, specificity, source compliance and argument depth in ForgeRank's model — then reports one weighted headline number. The per-dimension tiers are the useful output; the headline is a summary of them.
Why do two drafts with the same score need different fixes?
Because the headline averages four independent judgments. A 4.8 draft at 4.0/4.0/7.0/4.0 has one strong dimension and three weak ones, while a 7.2 draft at 7.0/7.0/8.0/7.0 is balanced and slightly behind on sourcing. Different shapes, different repairs.
Does a high quality score mean the writing sounds human?
No. ForgeRank's dataset shows eight of nine drafts scoring 7.0 or above still received a heavy-AI-phrasing verdict, and the only draft with clean phrasing scored 4.0. Phrasing is scored on a separate axis from substance.
How much should a score move between runs?
ForgeRank scores each dimension as the median of three samples, so the headline moves in tier steps rather than decimals. A tool reporting a different score on an unchanged draft is measuring variance, not quality.
What's the fastest way to raise a low score?
Fix the weakest dimension first. Specificity responds fastest because swapping an abstraction for a named instance is a local edit. Information gain carries the heaviest weight at 30 percent, so a draft with no original angle needs a new section, not a polish pass.
Open the ForgeRank published dataset tonight and find the row matching your draft's score. Read its four tier numbers before you rewrite a single sentence — the weakest tier tells you which section to rebuild, and the rest of the draft stays untouched.