AI Assessment
Automated Feedback vs Manual
Feedback: What Students Actually
Prefer
What the evidence says about timing, specificity, and whether the source
of feedback actually changes how students engage with it.
Your students submit thirty essays. You have two weeks. The feedback you return will
almost certainly arrive after they have moved on to the next assignment, the next topic, and
the next set of worries. Most of them will skim the grade and close the document. A few
will read every comment carefully. Almost none will act on it in time for it to help.
This is not a failure of effort. It is a structural problem with manual feedback at scale. The
question is whether automated feedback genuinely addresses that problem and whether
students actually want it.
What does the research say?
Students do not have a categorical preference for human-written feedback over automated
feedback. What they value is feedback that is timely, specific, and criterion-referenced.
When automated feedback meets those criteria, students rate it as highly or more highly
than manually written comments — particularly when feedback arrives quickly enough to
act on. The key variable is quality and speed, not the source.
What do students actually say they want from feedback?
Research on student feedback preferences consistently identifies the same three priorities:
speed, specificity, and actionability. Students want to know what they did wrong, why it
was wrong, and what to do differently next time — and they want this while the assignment
is still fresh enough to learn from.
A systematic review of student feedback needs in higher education (Winstone et al., 2019)
found that students most frequently cited timeliness and specificity as the factors that
determined whether they found feedback useful. Generic comments — “good analysis” or
“argument needs development” — were consistently rated less valuable, regardless of
whether they came from a human or an automated system.
The implication for the automated vs manual debate is important: the question students are
answering is not “did a human write this?” It is “does this tell me something useful that I can
act on?”
98%
of students rated automated
feedback as timely in a higher
education study (NCBI, 2023)
90%
rated it as high quality in the
same study
87%
said it was specific enough to
enhance their learning
Does it matter who or what wrote the feedback?
Yes, but perhaps not in the way you might expect. Research published in 2025 in
Assessment and Evaluation in Higher Education found that disclosing the source of
feedback significantly affects how students evaluate it. When students were told feedback
was AI-generated, their ratings of it dropped even when the feedback was objectively
identical to feedback labelled as human-written.
This is a disclosure effect, not a quality effect. The same comment reads differently
depending on who students believe wrote it. One implication is that how institutions
communicate AI feedback to students matters enormously. Framing automated feedback
as “AI-generated, reviewed by your lecturer” produces different responses than framing it
as purely machine-generated.
The key finding: students who could not identify the source of their feedback tended to rate AI
feedback equally or higher than human feedback. The preference for human feedback
emerged primarily when students were explicitly told the source. This suggests the preference
is partly attitudinal rather than based on experienced quality.
A further nuance comes from a 2025 paper in the British Educational Research Journal by
Flodén, which compared LLM-generated and human-generated grades on university
exams. LLMs showed systematic overconfidence, routinely assigning high scores even to
weaker responses. This underlines a real limitation: automated feedback without human
review carries accuracy risk which is precisely why the human-in-the-loop model,
where a lecturer reviews and approves before release, is the approach that combines the
best of both.
Student rating of feedback quality by source and disclosure condition
0
2
4
6
8
Rating (out of 10)
7.8
Human
(disclosed)
7.3
AI feedback
(source unknown)
5.5
AI feedback
(disclosed)
8.1
AI + lecturer
review (disclosed)
Illustrative based on research synthesis — not a single study
When students know the source, human-labelled feedback scores higher than AI-only. AI feedback reviewed
by a lecturer and disclosed as such scores highest of all — combining perceived credibility with speed.
Why timeliness changes everything
Here is the practical problem with manual feedback: it arrives too late. Research published
in Assessment and Evaluation in Higher Education in 2025 found that students showed
significantly lower motivation levels when feedback took more than ten days to arrive. The
learning opportunity — the window when students can process feedback and apply it —
closes quickly.
Manual feedback on a cohort of 200 submissions takes a human marker approximately 50-
100 hours. In practice, this means feedback arrives after the next assignment has already
been set, sometimes after the next assignment has already been submitted. At that point,
feedback is retrospective rather than formative — it tells students what they did wrong on a
task that is now finished, rather than what to do differently on a task they are still working
on.
Automated feedback, returned within hours or overnight, changes this fundamentally. A
2023 longitudinal randomised field experiment published in Frontiers in Psychology found
that daily automated feedback significantly enhanced self-regulated learning compared to
delayed feedback conditions. Students who received faster feedback were more likely to
revise and resubmit, more likely to apply the guidance to subsequent tasks, and showed
measurably better performance over time.
“Students express significantly lower levels of motivation when feedback
takes more than ten days. They do not distinguish between ‘timely’ and
‘extremely timely’ — but the threshold matters.”
Assessment and Evaluation in Higher Education, 2025
How do automated and manual feedback actually compare?
The comparison is not binary. In practice, the question is not “automated or human?” but
“what combination, in what workflow?” The table below compares the two approaches
across the dimensions students and institutions care about most.
Dimension
Manual feedback
Automated feedback
AI + human review
Turnaround time
7-21 days (typical)
Minutes to hours
Hours to 1-2 days
Consistency across
cohort
Variable — inter-rater
variance 20-40%
High consistency
against rubric
High + human
calibration
Specificity
Varies by marker
workload
Criterion-referenced
if rubric is clear
Criterion-
referenced,
lecturer-adjusted
Student perceived
quality (disclosed)
High
Moderate (disclosure
effect)
Highest (trust +
speed)
Scalability
Linear with cohort
size
Does not scale with
cohort
Does not scale with
cohort
Coverage (every
criterion)
Often incomplete
under time pressure
Every criterion,
every submission
Every criterion,
every submission
Discipline-specific
accuracy
High (expert
markers)
Depends on model
— general AI is
weaker
High with domain-
trained model
NSS impact
Bottleneck on Q6
(promptness)
Addresses
promptness directly
Addresses
promptness +
quality
EU AI Act
compliance
Not applicable
Requires human
oversight (Article 14)
Satisfies Article 14
Why quality, not source, is the real driver of student improvement
The research literature on feedback effectiveness consistently points to the same factors:
feedback must be timely, specific to the task criteria, forward-looking (telling students what
to do next rather than just what went wrong), and linked to learning outcomes. Hattie and
Timperley's landmark 2007 synthesis of feedback research found an average effect size of
d=0.73 — one of the largest positive effects on student learning of any educational
intervention — but noted that poorly designed feedback produced negative effects. The
quality of the feedback, not the medium of delivery, determined the outcome.
Nicol and Macfarlane-Dick (2006) identified seven principles of good feedback: it clarifies
what good performance looks like; it facilitates self-assessment; it delivers high-quality
information about learning; it encourages dialogue; it fosters positive motivation; it closes
the gap between current and desired performance; and it informs teaching. None of these
principles are inherently incompatible with automated feedback — they are design
requirements that apply regardless of whether a human or an AI produces the comment.
Where automated feedback currently underperforms is in the nuanced, dialogic elements
of Nicol and Macfarlane-Dick's framework: encouraging a back-and-forth conversation,
responding to a student's specific learning history, or adapting to emotional context. These
are genuine limitations. The practical response is not to reject automation, but to design
workflows where automated feedback handles the high-volume, criterion-referenced
elements — freeing lecturers to focus on the dialogic, relational dimensions where human
judgment genuinely adds value.
What does this mean in practice for lecturers?
The research points towards a clear practical recommendation: the most effective
feedback model combines automated first-pass assessment with structured human review
before release. This is not a compromise — it is, by the available evidence, the approach
students rate most positively and that produces the strongest learning outcomes.
What this looks like in practice
In a well-designed AI-assisted workflow, the automated system processes each
submission against the rubric, generates per-criterion scores and written comments, and
holds everything pending lecturer review. The lecturer sees a dashboard of all submissions,
can spot-check, adjust any score or comment, add personal context where relevant, and
release to students in bulk once satisfied. The turnaround drops from three weeks to two
days. Every student receives criterion-referenced feedback. The lecturer's time is
redirected from typing generic comments to reviewing, calibrating, and having targeted
conversations with students who need them.
The Eduface approach: Eduface gives lecturers the choice between two review modes. In blind
mode, lecturers mark first and the AI score is revealed afterwards — protecting against
anchoring bias. In standard mode, AI scores are shown upfront for the lecturer to review and
adjust. Grades are never released to students until the lecturer explicitly approves.
What to tell your students
Given the disclosure research, how you frame AI-assisted feedback matters. Telling
students their feedback was “AI-generated” without context tends to reduce engagement
with it. Telling them their feedback was “generated by an AI trained on your module's rubric
and reviewed by me before release” produces a very different response. Transparency
about the human review step — and about the fact that the lecturer has accountability for
every grade — is the framing that research suggests produces the best student response.
Frequently asked questions
Do students trust automated feedback as much as feedback from their lecturer?
When students know the source, they tend to rate human-labelled feedback slightly higher.
However, research shows this preference is largely attitudinal rather than based on experienced
quality differences. When automated feedback arrives quickly and is specific to the rubric
criteria, students engage with it at least as much as — often more than — delayed manual
feedback that arrives weeks later.
What makes automated feedback feel impersonal, and can it be fixed?
Generic or formulaic comments are the main driver of impersonality — not the automated origin
of the feedback. This is a quality problem, not a technology problem. Domain-specific AI models
that understand discipline conventions, combined with rubrics that require criterion-referenced
justification rather than template phrases, produce feedback that students experience as
specific and relevant. The lecturer review step also allows personal additions where appropriate.
How does feedback turnaround time affect NSS scores?
Assessment and Feedback is consistently one of the lowest-scoring sections in the National
Student Survey. Question 6 asks specifically whether feedback has been provided in time to be
useful. Manual marking at scale makes this question structurally difficult to answer positively.
Automated feedback that arrives within 24-48 hours directly addresses this bottleneck. In
Eduface pilots at UK institutions, faster feedback turnaround has been associated with
measurable improvements in student-reported assessment satisfaction.
Is automated feedback suitable for all types of assignment?
Automated feedback performs best on rubric-based written assignments where criteria are
clearly defined: essays, reports, case studies, open-ended exam questions, and short-answer
tasks. It is less suited to highly creative work where assessment criteria are deliberately
subjective, or to portfolio assessments that require evaluating development over time. For most
standard HE written assessment formats, automated feedback — with lecturer review — is a
practical and effective option.
What does the EU AI Act say about automated feedback in higher education?
The EU AI Act (Regulation 2024/1689) classifies AI systems used to evaluate and assess
students as high-risk under Annex III. Article 14 requires that human oversight is built into the
system's design and daily operation. In practice, this means a lecturer must have the genuine
ability to review, override, and take responsibility for AI-generated assessments before they
reach students. Workflows where lecturers approve grades before release — rather than
automatically accepting AI output — satisfy this requirement.
Give every student better feedback, faster
Eduface returns rubric-grounded, criterion-specific feedback on every submission —
reviewed by you before release. Free to try, no card required.
Create free account
Book a demo
AI Assessment
Automated Feedback vs Manual
Feedback: What Students
Actually Prefer
What the evidence says about timing, specificity, and
whether the source of feedback actually changes
how students engage with it.
Your students submit thirty essays. You have two
weeks. The feedback you return will almost certainly
arrive after they have moved on to the next
assignment, the next topic, and the next set of
worries. This is not a failure of effort. It is a structural
problem with manual feedback at scale.
What does the research say?
Students do not have a categorical preference
for human-written feedback over automated
feedback. What they value is feedback that is
timely, specific, and criterion-referenced. When
automated feedback meets those criteria,
students rate it as highly or more highly than
manually written comments — particularly when
it arrives quickly enough to act on.
98%
of students rated automated feedback
as timely in a higher education study
(NCBI, 2023)
90%
rated it as high quality in the same study
87%
said it was specific enough to enhance
their learning
What do students actually want from
feedback?
Research consistently identifies the same three
priorities: speed, specificity, and actionability.
Students want to know what they did wrong, why it
was wrong, and what to do differently next time —
while the assignment is still fresh enough to learn
from.
A systematic review of student feedback needs in
higher education (Winstone et al., 2019) found that
students most frequently cited timeliness and
specificity as the factors that determined whether
they found feedback useful. Generic comments —
"good analysis" or "argument needs development" —
were consistently rated less valuable, regardless of
whether they came from a human or an automated
system.
The implication is important: the question students
are answering is not "did a human write this?" It is
"does this tell me something useful that I can act on?"
Does it matter who wrote the feedback?
Yes — but not in the way you might expect. Research
published in 2025 in
Assessment and Evaluation in
Higher Education
found that disclosing the source of
feedback significantly affects how students evaluate
it. When students were told feedback was AI-
generated, their ratings dropped — even when the
feedback was objectively identical to feedback
labelled as human-written.
The key finding: students who could not identify
the source tended to rate AI feedback equally or
higher than human feedback. The preference for
human feedback emerged primarily when students
were explicitly told the source — suggesting this is
partly attitudinal rather than based on experienced
quality.
A 2025 paper in the
British Educational Research
Journal
found that LLMs showed systematic
overconfidence, routinely assigning high scores even
to weaker responses. This underlines why the
human-in-the-loop model — where a lecturer reviews
and approves before release — is the approach that
combines the best of both.
Why timeliness changes everything
Research published in
Assessment and Evaluation in
Higher Education
in 2025 found that students showed
significantly lower motivation when feedback took
more than ten days to arrive. The learning
opportunity — the window when students can
process feedback and apply it — closes quickly.
Manual feedback on a cohort of 200 submissions
takes a human marker approximately 50–100 hours. In
practice, feedback arrives after the next assignment
has already been set, sometimes after it has already
been submitted. At that point, feedback is
retrospective rather than formative.
"Students express significantly lower levels of
motivation when feedback takes more than ten
days. They do not distinguish between 'timely'
and 'extremely timely' — but the threshold
matters."
Assessment and Evaluation in Higher Education, 2025
A 2023 randomised field experiment published in
Frontiers in Psychology found that daily automated
feedback significantly enhanced self-regulated
learning compared to delayed feedback conditions.
Students who received faster feedback were more
likely to revise and resubmit, and showed measurably
better performance over time.
How do automated and manual
feedback compare?
The question is not "automated or human?" but "what
combination, in what workflow?"
Dimension
Manual
Automated
AI + human
Turnaround
7–21 days
Minutes–
hours
Hours–2 days
✓
Consistency
Variable
High vs rubric
High +
calibration ✓
Specificity
Varies
Criterion-ref
Criterion-ref,
adjusted ✓
Student rating
(disclosed)
High
Moderate
Highest ✓
Scalability
Linear cost
Fixed cost
Fixed cost ✓
Every criterion
covered
Often
incomplete
Always ✓
Always ✓
NSS impact
Bottleneck
on Q6
Addresses
promptness
Addresses
both ✓
EU AI Act Art.
14
N/A
Requires HITL
Satisfies Art.
14 ✓
Why quality, not source, is the real
driver
The research literature consistently points to the
same factors: feedback must be timely, specific to
the task criteria, forward-looking, and linked to
learning outcomes. Hattie and Timperley's landmark
2007 synthesis found an average effect size of
d=0.73 for feedback — one of the largest positive
effects on student learning of any educational
intervention — but noted that poorly designed
feedback produced negative effects.
Nicol and Macfarlane-Dick (2006) identified seven
principles of good feedback: it clarifies what good
performance looks like; it facilitates self-assessment;
it delivers high-quality information about learning; it
encourages dialogue; it fosters positive motivation; it
closes the gap between current and desired
performance; and it informs teaching. None of these
principles are inherently incompatible with automated
feedback.
Where automated feedback currently underperforms
is in the dialogic elements: responding to a student's
specific learning history, or adapting to emotional
context. The practical response is to design
workflows where automated feedback handles the
high-volume, criterion-referenced elements —
freeing lecturers to focus on the relational dimensions
where human judgment genuinely adds value.
What this means in practice
The most effective feedback model combines
automated first-pass assessment with structured
human review before release. In a well-designed AI-
assisted workflow, the automated system processes
each submission against the rubric and holds
everything pending lecturer review. The lecturer
reviews a dashboard of all submissions, adjusts any
score or comment, and releases to students in bulk
once satisfied.
The Eduface approach: Eduface gives lecturers the
choice between two review modes. In blind mode,
lecturers mark first and the AI score is revealed
afterwards — protecting against anchoring bias. In
standard mode, AI scores are shown upfront for
review. Grades are never released to students until
the lecturer explicitly approves.
What to tell your students
Given the disclosure research, how you frame AI-
assisted feedback matters. Telling students their
feedback was "AI-generated, reviewed by me before
release" produces a very different response than
describing it as purely machine-generated.
Transparency about the human review step is the
framing that research suggests produces the best
student response.
What to look for in an AI feedback platform
Feedback arrives within 24 hours of submission
Each comment is rubric-specific, not generic
Lecturer reviews and approves before release
Blind mode available for high-stakes
assessments
EU data hosting and Article 14 compliance
Students see feedback framed as human-
reviewed
Domain-specific models for your discipline
Frequently asked questions
Do students trust automated feedback as much as
feedback from their lecturer?
When students know the source, they tend to rate
human-labelled feedback slightly higher. However,
research shows this preference is largely attitudinal.
When automated feedback arrives quickly and is
specific to the rubric criteria, students engage with it
at least as much as — often more than — delayed
manual feedback that arrives weeks later.
What makes automated feedback feel impersonal,
and can it be fixed?
Generic or formulaic comments are the main driver
of impersonality — not the automated origin of the
feedback. This is a quality problem, not a technology
problem. Domain-specific AI models combined with
rubrics that require criterion-referenced justification
produce feedback that students experience as
specific and relevant.
How does feedback turnaround time affect NSS
scores?
Assessment and Feedback is consistently one of the
lowest-scoring sections in the National Student
Survey. Question 6 asks specifically whether
feedback has been provided in time to be useful.
Automated feedback that arrives within 24–48 hours
directly addresses this bottleneck.
Is automated feedback suitable for all types of
assignment?
Automated feedback performs best on rubric-based
written assignments where criteria are clearly
defined: essays, reports, case studies, open-ended
exam questions. It is less suited to highly creative
work where assessment criteria are deliberately
subjective, or to portfolio assessments that require
evaluating development over time.
What does the EU AI Act say about automated
feedback?
The EU AI Act classifies AI systems used to evaluate
students as high-risk under Annex III. Article 14
requires that human oversight is built into the
system's design and daily operation. A lecturer must
have the genuine ability to review, override, and take
responsibility for AI-generated assessments before
they reach students.
See AI feedback in action
Try Eduface's human-in-the-loop feedback
system with your department — free to start.
Create free account
Or book a demo for institutional rollout.
AI Assessment
Automated Feedback vs Manual
Feedback: What Students
Actually Prefer
What the evidence says about timing, specificity, and
whether the source of feedback actually changes
how students engage with it.
Your students submit thirty essays. You have two
weeks. The feedback you return will almost certainly
arrive after they have moved on to the next
assignment, the next topic, and the next set of
worries. This is not a failure of effort. It is a structural
problem with manual feedback at scale.
What does the research say?
Students do not have a categorical preference
for human-written feedback over automated
feedback. What they value is feedback that is
timely, specific, and criterion-referenced. When
automated feedback meets those criteria,
students rate it as highly or more highly than
manually written comments — particularly when
it arrives quickly enough to act on.
98%
of students rated automated feedback
as timely in a higher education study
(NCBI, 2023)
90%
rated it as high quality in the same study
87%
said it was specific enough to enhance
their learning
What do students actually want from
feedback?
Research consistently identifies the same three
priorities: speed, specificity, and actionability.
Students want to know what they did wrong, why it
was wrong, and what to do differently next time —
while the assignment is still fresh enough to learn
from.
A systematic review of student feedback needs in
higher education (Winstone et al., 2019) found that
students most frequently cited timeliness and
specificity as the factors that determined whether
they found feedback useful. Generic comments —
"good analysis" or "argument needs development" —
were consistently rated less valuable, regardless of
whether they came from a human or an automated
system.
The implication is important: the question students
are answering is not "did a human write this?" It is
"does this tell me something useful that I can act on?"
Does it matter who wrote the feedback?
Yes — but not in the way you might expect. Research
published in 2025 in
Assessment and Evaluation in
Higher Education
found that disclosing the source of
feedback significantly affects how students evaluate
it. When students were told feedback was AI-
generated, their ratings dropped — even when the
feedback was objectively identical to feedback
labelled as human-written.
The key finding: students who could not identify
the source tended to rate AI feedback equally or
higher than human feedback. The preference for
human feedback emerged primarily when students
were explicitly told the source — suggesting this is
partly attitudinal rather than based on experienced
quality.
A 2025 paper in the
British Educational Research
Journal
found that LLMs showed systematic
overconfidence, routinely assigning high scores even
to weaker responses. This underlines why the
human-in-the-loop model — where a lecturer reviews
and approves before release — is the approach that
combines the best of both.
Why timeliness changes everything
Research published in
Assessment and Evaluation in
Higher Education
in 2025 found that students showed
significantly lower motivation when feedback took
more than ten days to arrive. The learning
opportunity — the window when students can
process feedback and apply it — closes quickly.
Manual feedback on a cohort of 200 submissions
takes a human marker approximately 50–100 hours. In
practice, feedback arrives after the next assignment
has already been set, sometimes after it has already
been submitted. At that point, feedback is
retrospective rather than formative.
"Students express significantly lower levels of
motivation when feedback takes more than ten
days. They do not distinguish between 'timely'
and 'extremely timely' — but the threshold
matters."
Assessment and Evaluation in Higher Education, 2025
A 2023 randomised field experiment published in
Frontiers in Psychology found that daily automated
feedback significantly enhanced self-regulated
learning compared to delayed feedback conditions.
Students who received faster feedback were more
likely to revise and resubmit, and showed measurably
better performance over time.
How do automated and manual
feedback compare?
The question is not "automated or human?" but "what
combination, in what workflow?"
Dimension
Manual
Automated
AI + human
Turnaround
7–21 days
Minutes–
hours
Hours–2 days
✓
Consistency
Variable
High vs rubric
High +
calibration ✓
Specificity
Varies
Criterion-ref
Criterion-ref,
adjusted ✓
Student rating
(disclosed)
High
Moderate
Highest ✓
Scalability
Linear cost
Fixed cost
Fixed cost ✓
Every criterion
covered
Often
incomplete
Always ✓
Always ✓
NSS impact
Bottleneck
on Q6
Addresses
promptness
Addresses
both ✓
EU AI Act Art.
14
N/A
Requires HITL
Satisfies Art.
14 ✓
Why quality, not source, is the real
driver
The research literature consistently points to the
same factors: feedback must be timely, specific to
the task criteria, forward-looking, and linked to
learning outcomes. Hattie and Timperley's landmark
2007 synthesis found an average effect size of
d=0.73 for feedback — one of the largest positive
effects on student learning of any educational
intervention — but noted that poorly designed
feedback produced negative effects.
Nicol and Macfarlane-Dick (2006) identified seven
principles of good feedback: it clarifies what good
performance looks like; it facilitates self-assessment;
it delivers high-quality information about learning; it
encourages dialogue; it fosters positive motivation; it
closes the gap between current and desired
performance; and it informs teaching. None of these
principles are inherently incompatible with automated
feedback.
Where automated feedback currently underperforms
is in the dialogic elements: responding to a student's
specific learning history, or adapting to emotional
context. The practical response is to design
workflows where automated feedback handles the
high-volume, criterion-referenced elements —
freeing lecturers to focus on the relational dimensions
where human judgment genuinely adds value.
What this means in practice
The most effective feedback model combines
automated first-pass assessment with structured
human review before release. In a well-designed AI-
assisted workflow, the automated system processes
each submission against the rubric and holds
everything pending lecturer review. The lecturer
reviews a dashboard of all submissions, adjusts any
score or comment, and releases to students in bulk
once satisfied.
The Eduface approach: Eduface gives lecturers the
choice between two review modes. In blind mode,
lecturers mark first and the AI score is revealed
afterwards — protecting against anchoring bias. In
standard mode, AI scores are shown upfront for
review. Grades are never released to students until
the lecturer explicitly approves.
What to tell your students
Given the disclosure research, how you frame AI-
assisted feedback matters. Telling students their
feedback was "AI-generated, reviewed by me before
release" produces a very different response than
describing it as purely machine-generated.
Transparency about the human review step is the
framing that research suggests produces the best
student response.
What to look for in an AI feedback platform
Feedback arrives within 24 hours of submission
Each comment is rubric-specific, not generic
Lecturer reviews and approves before release
Blind mode available for high-stakes
assessments
EU data hosting and Article 14 compliance
Students see feedback framed as human-
reviewed
Domain-specific models for your discipline
Frequently asked questions
Do students trust automated feedback as much as
feedback from their lecturer?
When students know the source, they tend to rate
human-labelled feedback slightly higher. However,
research shows this preference is largely attitudinal.
When automated feedback arrives quickly and is
specific to the rubric criteria, students engage with it
at least as much as — often more than — delayed
manual feedback that arrives weeks later.
What makes automated feedback feel impersonal,
and can it be fixed?
Generic or formulaic comments are the main driver
of impersonality — not the automated origin of the
feedback. This is a quality problem, not a technology
problem. Domain-specific AI models combined with
rubrics that require criterion-referenced justification
produce feedback that students experience as
specific and relevant.
How does feedback turnaround time affect NSS
scores?
Assessment and Feedback is consistently one of the
lowest-scoring sections in the National Student
Survey. Question 6 asks specifically whether
feedback has been provided in time to be useful.
Automated feedback that arrives within 24–48 hours
directly addresses this bottleneck.
Is automated feedback suitable for all types of
assignment?
Automated feedback performs best on rubric-based
written assignments where criteria are clearly
defined: essays, reports, case studies, open-ended
exam questions. It is less suited to highly creative
work where assessment criteria are deliberately
subjective, or to portfolio assessments that require
evaluating development over time.
What does the EU AI Act say about automated
feedback?
The EU AI Act classifies AI systems used to evaluate
students as high-risk under Annex III. Article 14
requires that human oversight is built into the
system's design and daily operation. A lecturer must
have the genuine ability to review, override, and take
responsibility for AI-generated assessments before
they reach students.
See AI feedback in action
Try Eduface's human-in-the-loop feedback
system with your department — free to start.
Create free account
Or book a demo for institutional rollout.