How good is Fiveable's AI grading for AP® FRQs?
Fiveable grades a class set of AP® free-response questions against the rubric in minutes. Teachers then review every score before students see it. Since teachers can change any score they disagree with, what they keep and what they change is the most honest measure we have. Here is what more than a thousand teachers decided.
96.2%
of FRQ responses that teachers reviewed one at a time kept Fiveable's score exactly as given.
Each circle is a reviewed response. Teachers changed the final score on about 1 in 26.
21,353 responses reviewed individually · 812 changed · Updated September 3, 2026
How grading works
Fiveable makes the first pass. The teacher makes the call.
Every number on this page comes from this three-step flow. The teacher's decision in step three is what we measure.
Upload a class set
Essays come in from Google Classroom, Docs, photos of handwritten work, or paste. The teacher picks the AP® subject and FRQ type, or uses a Fiveable prompt.
Fiveable scores it
Each response gets a point-by-point score with a reason for every rubric row, in the same format AP® scoring guidelines use.
The teacher decides
The teacher reads each score, keeps or changes any rubric row, then approves it. Only then does the student see a grade.
So the question "can I trust the first pass?" has a concrete answer: look at how often teachers who read the scores left them alone.
The evidence, part one
Which responses count toward the 96.2%
Fiveable has scored 80,929 FRQ responses, for students practicing on their own and for teachers grading class sets. The quality number uses only the class-set responses a teacher approved one at a time. Approve All approvals are real, but they do not prove the teacher read each score, so they are set aside.
80,929 FRQ responses scored on Fiveable
53,805 class-set responses, by teacher approval
21,353 responses approved one at a time, by outcome
812 responses had a score changed
What counts as reviewed
A response the teacher approved on its own, not as part of an approve-all action on the class set.
What counts as a change
Any edit the teacher made to the points earned on a rubric row before approving.
What it is not
It is not a comparison with College Board scoring or AP Readers, and it is not a promise about any single response.
The evidence, part two
What the other 3.8% looks like
Kept scores tell you how often teachers agreed. Changed scores tell you how Fiveable misses when it does. A grader that runs harsh or lenient would show up here as a lopsided chart. Instead, increases and decreases are almost exactly even, and nearly every change is a single rubric point.
Rubric points changed on each score, all 1,715 changes
teacher decreased the score
teacher increased the score
1-3 points
33-2 points
831-1 point
818+1 point
32+2 points
1.04
rubric points, average size of a change
96%
of changes were a single rubric point
The evidence, part three
FRQs scored, by subject
Fiveable has scored FRQs in every AP® subject with a free-response section, for teachers grading class sets and for students practicing on their own. Counts below combine both. A teacher review rate appears only where teachers have individually reviewed at least 500 rubric points in that subject, so a small sample never poses as precision.
Enough teacher reviews to publish a rate
Scored on Fiveable, fewer teacher reviews so far
AP World History: Modern
12,352
98.3% unchanged points
AP US History
7,468
99.0% unchanged points
AP English Language
5,441
98.3% unchanged points
AP Human Geography
4,889
99.2% unchanged points
AP Psychology
4,831
99.8% unchanged points
AP European History
4,341
98.6% unchanged points
AP US Government
3,636
97.4% unchanged points
AP Computer Science A
2,748
responses scored
AP English Literature
2,489
99.5% unchanged points
AP Biology
1,735
responses scored
AP Computer Science Principles
1,633
99.6% unchanged points
AP Comparative Government
1,531
100.0% unchanged points
AP Environmental Science
1,519
98.3% unchanged points
AP Statistics
920
responses scored
AP Pre-Calculus
755
responses scored
AP Seminar
621
96.5% unchanged points
AP Spanish Language
527
responses scored
AP Macroeconomics
476
responses scored
Fewer teacher reviews in a subject means fewer teachers have run class sets there so far. It says nothing about how well that subject is scored. Every subject shown has been graded on Fiveable.
The evidence, part four
Teachers who try it come back with the next class set.
Numbers about kept scores show agreement. Behavior shows trust. Of the 1,278 teachers who have graded FRQs with Fiveable, most uploaded another set after the first, and about three in ten have run five or more.
1,278 teachers · 6,811 class sets · average 5.3 per teacher
Read this before you quote us
What this evidence shows, and what it does not.
It shows
- How often teachers kept Fiveable's score on responses they reviewed individually.
- How large teacher changes are and which direction they go.
- Where teachers and students have used Fiveable scoring, by subject.
- Exactly which responses are counted and which are left out.
It does not show
- An accuracy rate for every subject, prompt, or rubric row.
- Agreement with College Board scoring or AP Readers.
- That every teacher reviews with the same care or the same standard.
- That teacher changes retrain a model. They give us cases to evaluate, nothing is retrained automatically.
Questions teachers ask
Common questions about the numbers.
Does 96.2% mean Fiveable is 96.2% accurate?
No. It means that when teachers reviewed responses one at a time, they kept Fiveable's score on 96.2% of them and changed it on the rest. It is what teachers decided in real classrooms, not an independent accuracy study.
Why leave out responses approved with one click?
About a third of approved responses were approved all at once for a whole class set. Those approvals are real, but they do not tell us the teacher read each score, so we do not let them raise the number.
What happens when a teacher changes a score?
The change is saved with the response and the student sees the teacher's score. We look at changes by subject, question type, and rubric row to find repeated patterns worth evaluating. Nothing is retrained automatically.
Why is there no review rate for my subject?
Fiveable has scored responses in every AP® subject. We only publish a review rate once teachers have individually reviewed at least 500 rubric points in that subject, so the number reflects real use rather than a handful of essays.
Try the first pass on your own class set.
Import AP® FRQs, review every rubric point, and keep the final decision on what students see.
Grade a class set