Prompt Engineering Implementation Partner
Module 2: Practical Foundations of
Prompting
Video 2: Zero-Shot and One-Shot Prompting
Sushmitha Dixith
Data Scientist | GITAA Pvt Ltd
Recap: Video 1
Prompt engineering = communicating intent clearly to a probabilistic model.
4 components: Task, Context, Input, Expected Output — missing one causes drift.
Explicit prompts outperform vague ones: specificity narrows the output distribution.
Stating your purpose shifts how the model calibrates depth, tone, and vocabulary.
Zero-Shot Prompting - Asking Directly, No Examples
Zero-shot = task only, no examples provided. Works when the model already knows how to do the job.
ZERO-SHOT EXAMPLE
✓ Works best when…
"Classify the sentiment of this review as Positive, Neutral, or
Negative:
Task is familiar & unambiguous · Standard
'The delivery was fast but the packaging was damaged.'" → Neutral format acceptable · Speed > precision
When Zero-Shot Breaks Down
✗ Non-standard format expected ✗ Ambiguous interpretation ✗ Specific tone or persona required
Model guesses structure — may give "Analyze this" could mean sentiment, Without instruction, model defaults
prose when you needed JSON, or financial, rhetorical, or competitive to a neutral, generic register that may
bullets when you needed a table. analysis. Model picks one. not match your brand or audience.
One-Shot Prompting - One Example, More Clarity
One-shot = you show the model one example before the task. The example defines the implicit contract.
ANATOMY OF A ONE-SHOT PROMPT
① THE EXAMPLE (shows the contract) ② THE TASK (applies the contract)
Example: Now do the same for:
Input: "Meeting ran over. Key decision: delay Input: "Budget approved for Q3. Sarah to lead
launch." procurement."
Output: "Launch delayed. Decision owner: PM team." Output: ???
What one example communicates implicitly:
Output Structure Vocabulary Level Appropriate Length Edge Case Handling
How to deal with
Exact format to follow — Register, formality, Rough word count and
ambiguous or complex
no guessing required technical vs. plain language density expected
inputs
Choosing the Right Example
The example you choose defines the implicit specification. Get it wrong and one-shot backfires.
Representative, not exceptional ✓ A normal meeting summary ✗ A 3-hour all-hands with 12
01 Use a typical case — not your most interesting or difficult
one. If your example is unusually long or complex, the
from last Tuesday agenda items
model calibrates to that level and over-elaborates on
simpler inputs.
Format matches desired output ✓ Exact format you want ✗ A "close enough" example
02 Every formatting choice in the example becomes a rule. returned with slight differences
Bullet style, capitalisation, punctuation, field order — if it's
in the example, the model will replicate it.
Length & tone mirror expectations ✓ 200 words if you want ✗ A 600-word example when you
03 If you want concise, make your example concise. If you
want formal, use formal language in the example. The
200-word outputs need 150-word outputs
model interprets length and tone as specifications, not
coincidences.
Zero-Shot vs. One-Shot
Task: Convert rough meeting notes into an executive email.
INPUT NOTES
"Talked to Priya re Q4 budget. Procurement blocking software purchase. Need escalation. End of month deadline.
Tom handles."
Zero-Shot Result One-Shot Result (example email provided)
Dear Team,
Owner: Tom Chen
I wanted to share a brief update following my conversation with Action: Escalate Q4 software procurement block to resolve before
Priya regarding the Q4 budget. We are currently experiencing a Oct 31.
procurement hold on the software purchase, which will require VS Background: Priya flagged procurement holding approval — budget
escalation in order to meet our end-of-month deadline. I have asked confirmed.
Tom to take ownership of this matter... Next step: Tom to contact procurement director by EOD Friday.
[continues for 3 more paragraphs]
✗ Too long — 4 paragraphs ✓ Owner highlighted immediately
✗ Owner buried in paragraph 2 ✓ Concise — 4 lines
✗ No clear next action ✓ Clear next action with deadline
✗ Passive, padded language ✓ Tone matched the example
Decision Guide - When to use Zero-Shot vs. One-Shot?
Start zero-shot. Upgrade when you see inconsistency.
START: Write the prompt without any examples → Run it once
Is the output consistent and correctly formatted?
YES NO
YES → Stay with Zero-Shot NO → Upgrade to One-Shot
→ Output format drifts between
✓ Output is usable as-is runs
✓ Format is consistent → Tone keeps being off
✓ Tone is appropriate → You have a clear "good" example
✓ No need to add complexity → Task has ambiguous
interpretation
Key Takeaways
1 Zero-shot for simple, unambiguous tasks — Don't add complexity when the model already knows the job
One-shot = minimum useful fix for unstable output — One well-chosen example resolves most format and tone
2
issues
3 Examples set implicit rules — Format, vocab, length, edge-case handling — all taught by the example
4 The right example beats more instructions — Showing is usually clearer than describing
5 Start zero-shot, upgrade when needed — Use one-shot only when you see real inconsistency
Up Next: Few-Shot Prompting Done Well