You’re
grading
120
essays
on
Friday
night
while
planning
next
week’s
lessons.
An
AI
assistant
could
handle
the
essay
feedback
in
45
minutes
instead
of
three
hours
—
but
only
if
you
set
it
up
right.
Most
teachers
who
try
AI
give
up
because
they
don’t
know
what
to
ask
for,
get
back
generic
responses,
or
spend
more
time
fixing
the
output
than
writing
it
themselves.
This
guide
covers
three
specific
workflows
—
lesson
planning,
grading
with
feedback,
and
personalizing
assignments
—
using
free
tools.
Each
includes
a
working
prompt
you
can
copy
today.
Why
Free
Tools
Actually
Work
for
Teachers
ChatGPT’s
free
tier
and
Claude’s
free
version
both
run
on
capable
models.
Claude
3.5
Sonnet
(free)
outperforms
GPT-4o
on
reading
comprehension
tasks
—
a
straight
benchmark
from
Anthropic’s
January
2025
eval.
For
teaching
work,
you’re
not
trading
quality
for
cost.
The
constraint
is
context:
free
versions
limit
the
amount
of
text
you
can
paste
at
once.
A
5,000-word
assignment
submission
fits.
Your
entire
curriculum
probably
doesn’t.
Work
within
those
bounds,
and
the
tool
does
its
job.
Speed
matters
more
than
you
think.
A
10-minute
prompt-writing
session
that
saves
you
90
minutes
of
grading
is
worth
the
learning
curve.
Workflow
1:
Generating
Lesson
Plans
with
Specificity
The
default
approach
fails
immediately.
#
Bad
prompt
Create
a
lesson
plan
on
photosynthesis
for
high
school
students.You’ll
get
a
generic
outline:
intro,
explain
the
process,
activity,
assessment.
It’ll
be
correct
and
useless
because
it
doesn’t
know
your
class
size,
your
materials,
your
9th
graders’
reading
level,
or
whether
you
have
lab
access.
Here’s
what
works:
#
Improved
prompt
I
teach
9th
grade
biology
to
28
students.
We
have
45
minutes,
basic
lab
equipment
(test
tubes,
heat
lamps,
plants),
and
my
students
average
7th-grade
reading
level.
Create
a
lesson
plan
on
photosynthesis
that:
-
Opens
with
a
3-minute
hook
that
doesn't
require
reading
-
Includes
one
hands-on
demo
I
can
run
in
20
minutes
-
Has
a
closing
activity
that
takes
10
minutes,
checks
understanding,
and
doesn't
require
writing
essays
-
Lists
all
materials
I
need
upfront
Format
as
a
table
with
time
stamps.That
prompt
pulls
multiple
constraints
into
one
request.
You’ll
get
a
structured,
time-locked
plan
you
can
actually
follow.
Add
one
more
detail
if
your
school
blocks
certain
websites
or
you
want
the
demo
tied
to
a
specific
standard
—
the
model
adjusts.
For
lesson
planning,
ChatGPT’s
free
version
works
fine.
You’re
generating,
not
analyzing.
If
the
output
misses
something,
ask
a
follow-up
question
in
the
same
chat
—
the
model
remembers
your
constraints.
Workflow
2:
Grading
Essays
with
Consistent
Rubrics
This
is
where
most
teachers
try
AI
and
quit.
The
feedback
comes
back
vague
or
harsh.
#
Bad
approach
Here's
my
student's
essay.
Give
me
feedback.
[paste
800-word
essay]Output:
“Good
effort.
You
could
strengthen
your
thesis
and
add
more
evidence.”
That
doesn’t
tell
the
student
what
to
fix
or
why
they
lost
points.
The
fix
is
a
rubric.
Hand
the
model
your
grading
criteria
first.
#
Improved
prompt
You
are
grading
9th-grade
essays
using
this
rubric:
Thesis
(25
points):
Clear
argument
stated
in
intro.
Guides
the
essay.
Evidence
(25
points):
3+
cited
sources.
Each
quote
connected
to
thesis.
Reasoning
(25
points):
Student
explains
WHY
evidence
supports
thesis,
not
just
what
it
says.
Organization
(15
points):
Logical
paragraph
order.
Transitions
between
ideas.
Mechanics
(10
points):
Spelling,
grammar,
punctuation.
Grade
this
essay.
For
each
category,
give:
1.
Points
earned
(out
of
category
max)
2.
One
specific
sentence
or
phrase
that
exemplifies
the
score
3.
One
actionable
revision
for
the
next
draft
Keep
feedback
brief
—
no
more
than
2
sentences
per
category.
[paste
essay
here]Now
the
output
matches
your
grading
standard.
Students
see
exactly
where
they
lost
points
and
what
to
change.
Copy
this
rubric
template
into
a
new
chat
for
each
essay
—
don’t
paste
10
essays
at
once.
The
model
handles
one
essay
better
than
a
batch.
For
grading,
use
Claude’s
free
version
(not
GPT-4o
free).
Claude
performs
better
on
close
reading
and
nuance
—
critical
for
essay
feedback.
It
won’t
hallucinate
citations
or
overstate
student
claims
the
way
weaker
models
do.
Workflow
3:
Personalizing
Assignments
Without
Manual
Remixing
One
assignment
for
28
students
doesn’t
work.
Some
need
a
reading
level
two
grades
lower.
Others
need
more
challenge.
Rewriting
assignments
yourself
multiplies
your
workload.
Instead,
ask
the
model
to
generate
multiple
versions
simultaneously.
#
Improved
prompt
Create
3
versions
of
this
math
assignment
on
linear
equations.
Version
1:
For
students
reading
at
6th-grade
level.
Use
fewer
words
per
problem.
Include
one
worked
example
at
the
top.
Version
2:
Standard
difficulty.
Match
the
original.
Version
3:
For
advanced
students.
Add
one
problem
requiring
multi-step
reasoning.
Format
each
version
as
a
numbered
list.
Keep
vocabulary
consistent
across
all
three
—
only
change
complexity,
not
terminology.
Original
assignment:
[paste
your
assignment
here]You
get
three
differentiated
versions
in
one
request.
Print
different
sets
for
different
groups.
This
cuts
prep
time
from
30
minutes
(rewriting
manually)
to
5
minutes
(reviewing
the
output
and
making
minor
tweaks).
For
differentiation,
ChatGPT
free
works
fine.
You’re
generating
variations,
not
analyzing
deep
context.
The
One
Thing
to
Do
Right
Now
Create
a
folder
in
your
notes
app
or
Google
Drive
and
save
three
template
prompts:
one
for
lesson
planning,
one
for
grading,
one
for
differentiation.
Customize
each
with
your
specific
class
details
(grade
level,
class
size,
time
constraints,
any
accessibility
needs).
Tomorrow,
when
you’re
planning
Thursday’s
lesson,
use
the
template
instead
of
starting
from
scratch.
The
time
investment
pays
back
immediately.
After
three
uses,
you’ll
have
faster
output
because
you’ve
refined
what
