Specific and Parseable · ai-skills

Asking Clearly · Beginner to Intermediate

On this page

Specificity beats length

Specific prompts name values, signals, and bounds. Long prompts with adjectives but no numbers stay vague.

Added length helps only when each sentence removes an ambiguity. More words without more facts add cost without gain.

Vague line	Specific rewrite	What changed
`Explain the FIFO.`	`Explain FUNCTION and DEPTH of this 8-deep FIFO.`	Names the keys and the parameter.
`Check reset.`	`List registers not reset by rst_n in reset_sync.`	Names the artifact and the check.
`Summarise the log.`	`List each error code with its first time and count.`	Names the fields and the grouping.

Numbers beat adjectives

Numbers state depth, width, counts, and limits. Adjectives like fast or robust state no test.

A number also bounds review. A claim with a count can be checked against the source.

Ask for a format that parses

Ask for lines, keys, or tables a script can split. Fixed keys with one separator parse with a few lines of code.

Name the keys, the separator, and the row order. One example line shows the shape exactly.

example.py

import sys

def is_record(line):
    if ": " not in line:
        return False
    key, value = line.split(": ", 1)
    return key.strip() != "" and value.strip() != "" and key == key.upper()

def main():
    lines = [
        "FUNCTION: store UART bytes",
        "DEPTH: 8 entries",
        "SIGNALS: full empty rd_en wr_en",
    ]
    loose = [
        "store UART bytes",
        "depth: 8 entries",
    ]
    good = 0
    for line in lines:
        if not is_record(line):
            print("FAIL: good line set aside")
            sys.exit(1)
        good = good + 1
    for line in loose:
        if is_record(line):
            print("FAIL: loose line kept as record")
            sys.exit(1)
    if good != 3:
        print("FAIL: count not 3")
        sys.exit(1)
    print("PASS: 3 records parsed, 2 loose lines set aside")

main()

Common mistakes

The checker above is a plain format gate. It proves that fixed KEY: VALUE lines split cleanly and loose lines do not pass.

It calls no model. Format discipline is checked by code, not by reading.

Check yourself A prompt asks for a careful summary with no keys and no example. What is the review cost?

Every answer needs a full reading to check. Fixed keys would turn part of that review into a script.

Check yourself In the program, why does the lowercase line fail the gate?

Its key is not all uppercase. The gate requires KEY: VALUE with an uppercase key and a colon separator.

Related topics

Up nextWeak Versus Strong Prompts

Practice this lesson Report an error

Specificity beats length

Specific prompts name values, signals, and bounds. Long prompts with adjectives but no numbers stay vague.

Added length helps only when each sentence removes an ambiguity. More words without more facts add cost without gain.

Vague line	Specific rewrite	What changed
`Explain the FIFO.`	`Explain FUNCTION and DEPTH of this 8-deep FIFO.`	Names the keys and the parameter.
`Check reset.`	`List registers not reset by rst_n in reset_sync.`	Names the artifact and the check.
`Summarise the log.`	`List each error code with its first time and count.`	Names the fields and the grouping.

Ask for a format that parses

Ask for lines, keys, or tables a script can split. Fixed keys with one separator parse with a few lines of code.

Name the keys, the separator, and the row order. One example line shows the shape exactly.

example.py

import sys

def is_record(line):
    if ": " not in line:
        return False
    key, value = line.split(": ", 1)
    return key.strip() != "" and value.strip() != "" and key == key.upper()

def main():
    lines = [
        "FUNCTION: store UART bytes",
        "DEPTH: 8 entries",
        "SIGNALS: full empty rd_en wr_en",
    ]
    loose = [
        "store UART bytes",
        "depth: 8 entries",
    ]
    good = 0
    for line in lines:
        if not is_record(line):
            print("FAIL: good line set aside")
            sys.exit(1)
        good = good + 1
    for line in loose:
        if is_record(line):
            print("FAIL: loose line kept as record")
            sys.exit(1)
    if good != 3:
        print("FAIL: count not 3")
        sys.exit(1)
    print("PASS: 3 records parsed, 2 loose lines set aside")

main()

Common mistakes

The checker above is a plain format gate. It proves that fixed KEY: VALUE lines split cleanly and loose lines do not pass.

It calls no model. Format discipline is checked by code, not by reading.

Check yourself A prompt asks for a careful summary with no keys and no example. What is the review cost?

Every answer needs a full reading to check. Fixed keys would turn part of that review into a script.

Check yourself In the program, why does the lowercase line fail the gate?

Its key is not all uppercase. The gate requires KEY: VALUE with an uppercase key and a colon separator.