Your
prompt
works
fine
in
Claude.
Then
you
paste
it
into
GPT-4o
and
get
garbage.
You
switch
to
Gemini
and
the
response
is
formatted
wrong.
This
isn’t
a
sign
you
need
better
prompting—it’s
a
sign
you
need
prompts
built
for
the
specific
model
in
front
of
you.
Each
model
has
different
training,
different
token
handling,
and
different
instruction-following
patterns.
A
prompt
that
exploits
Claude’s
strengths
will
waste
tokens
on
GPT-4o.
What
works
for
structured
extraction
in
Gemini
might
confuse
Mistral.
This
is
not
theoretical.
It’s
the
difference
between
a
70%
success
rate
and
a
94%
success
rate
on
the
same
task.
Why
One
Prompt
Doesn’t
Fit
Three
Models
Claude
(especially
Sonnet
4)
was
trained
with
Constitutional
AI,
which
makes
it
respond
well
to
direct
instruction
and
explicit
reasoning
chains.
It’s
efficient
with
tokens
and
handles
edge
cases
without
over-apologizing.
GPT-4o
is
optimized
for
instruction-following
on
a
massive
scale—it
knows
200+
prompt
engineering
tricks
because
millions
of
users
tried
them.
Gemini
(particularly
the
latest
2.0
models)
excels
at
multimodal
tasks
and
has
different
instruction
prioritization.
Token
efficiency
matters.
Claude’s
context
window
is
200K
tokens,
but
GPT-4o
charges
differently
for
input
vs.
output.
A
prompt
that
wastes
2,000
input
tokens
in
Claude
costs
you
$0.30.
The
same
waste
in
GPT-4o
costs
$1.20.
Gemini
pricing
changes
yet
again.
More
important:
each
model
weights
instructions
differently.
What
you
emphasize
in
the
system
prompt,
where
you
place
examples,
and
how
you
structure
reasoning—these
directly
affect
output
quality
per
model.
I
tested
this
extensively
building
AlgoVesta’s
trading
signal
extraction
pipeline.
Same
task,
three
models,
three
completely
different
prompt
structures
to
hit
90%+
accuracy
on
all
three.
The
Claude-First
Principle:
Direct
Instruction
Claude
responds
best
to
clarity
without
hedging.
It
doesn’t
need
you
to
ask
permission
or
soften
requests.
#
Bad
prompt
(overly
cautious)
Could
you
possibly
help
me
extract
the
key
financial
metrics
from
this
earnings
report?
I'd
really
appreciate
it
if
you
could
also
summarize
the
risks
mentioned.#
Improved
prompt
(direct,
structured)
Extract
these
financial
metrics
from
the
earnings
report:
-
Revenue
(total
and
by
segment)
-
Operating
margin
-
Cash
flow
from
operations
-
Key
risks
mentioned
in
the
MD&A
section
Format
as
JSON
with
these
exact
keys.Claude’s
training
makes
it
penalize
verbosity.
Every
hedge
word—
