Zero-Shot
No examples is not the same as no instructions — and this is the right default for most work.
What is Zero-Shot?
Zero-shot means one thing precisely: you gave the model no worked examples. It does not mean you gave it no instructions, no context, no format and no constraints. Confusing those two is the most expensive misunderstanding on this site, and it runs in both directions.
This is the page that argues against the rest of the site. Most of the time, zero-shot is the correct choice — the model has seen millions of executive summaries, meeting agendas and polite refusals, and it does not need you to show it one. What it needs is to know which of the millions you want. So the failure is rarely "should have used examples". It is that people hear zero-shot and write eight words, then conclude the model is not very good. A zero-shot prompt can be two hundred words of precise instruction and still be zero-shot. The worked example below goes from 30 to 91 without a single demonstration in it.
Where Zero-Shot Came From
A machine-learning term that changed meaning
"Zero-shot" comes from machine learning, where it meant a model handling a class it had never been trained on. The prompting sense was fixed by Brown et al., "Language Models are Few-Shot Learners" (2020) — the GPT-3 paper — which defined a ladder: zero-shot (instruction only), one-shot (one demonstration), few-shot (several). All three happen at inference time with no weight updates. Zero-shot is simply the bottom rung, not a lesser technique.
Zero-shot chain of thought made the bottom rung much stronger
The interesting wrinkle. Kojima et al., 2022 found that appending "let's think step by step" — no examples at all — recovered much of the reasoning benefit that Wei's few-shot chain of thought had needed demonstrations for. Zero-shot stopped being merely the cheap option for a whole class of problems.
And instruction tuning moved the line again
The GPT-3-era advice to add examples was formed on base models that were completing text rather than following orders. Modern instruction-tuned models are trained specifically to obey a stated instruction, which is why zero-shot now handles work that genuinely needed demonstrations in 2020. Advice about when to add examples ages faster than almost anything else in prompting — test it rather than inheriting it.
The 5 Slots, One at a Time
Each slot is a decision. Leave it out and the model still makes it — just without you.
With no examples to imply the deliverable, the instruction has to carry it. "Summarise this" leaves the model choosing between a paragraph, ten bullets and an abstract; name the thing you want and what it is for.
The highest-value sentence in most zero-shot prompts, and the one people skip because it feels like preamble. "She has ten minutes, has not read it, and will be asked about anything that went backwards" changes every subsequent choice the model makes.
This is what an example would otherwise have shown you. Say how many sections, what each one holds, and roughly how long — three bullets, one paragraph on the worst news, one on the decision, three questions to expect. Specify it and the demonstration becomes unnecessary.
A quietly powerful addition. Source documents are usually arranged by department or chronology; readers need significance. Saying "lead with whatever changed most, not with whatever the report puts first" reorders the whole output and costs one sentence.
With no examples to bound the register, exclusions do that work. Jargon, softening qualifiers, anything unverifiable — naming them is cheaper than describing the tone you want, and checkable afterwards.
One Task, Before and After
The task: A ten-minute executive summary of a 40-page operations report. Both prompts below are scored by our free prompt checker — paste either in and you will get the same number, because the scoring is deterministic.
Summarise this report for the CEO.
Six words. You get a competent, evenly-weighted summary that follows the report's own department order, buries the one number that has fallen three quarters running, and mentions two internal systems by name.
Summarise a 40-page quarterly operations report for the chief executive. She has ten minutes before the board meeting, has not read the report, and will be asked about anything that has gone backwards. She is not operational - she does not know the internal team names or the system names, so use plain descriptions instead. Write 350 to 450 words in four parts: three bullets on what moved and by how much, one short paragraph on the single thing that has got worse and why, one short paragraph on the decision she is being asked to make, and three questions she should expect from the board. Lead with whatever changed most, not with whatever the report puts first - reports are ordered by department and she needs it ordered by significance. Work through the report before you write, deciding which single item is genuinely the worst news rather than the most prominently presented. For example, the opening bullet should read like this: ``` On-time delivery fell from 94% to 88% across the quarter, the third consecutive quarterly fall. ``` Ensure every number you use appears in the report, and prioritise trend over level - a figure that is good but falling matters more to this reader than one that is mediocre and stable. Do not use internal jargon or system names, do not soften a decline with a qualifier, and do not include anything you cannot point to a page for.
Ninety-one, and there is not one demonstration in it. The single fenced line is a sample of register, not a worked example — remove it and the prompt is still zero-shot and still in the eighties.
Ninety-one with no examples, which is the entire argument of this page
One check is unreachable, and the two things that actually decide whether zero-shot was the right call are unmeasured:
No persona, because none was needed. Zero-shot is an approach rather than a slot structure, so add one if you want the eight points — but a summary for a named reader rarely improves from being written by an imaginary analyst.
The only question this page is really about, and no rubric answers it. The tell is not in the score, it is in the output: if the substance is right and the shape is wrong, add examples. If the substance is wrong, examples will not save it.
Both "summarise this" and the 235-word prompt above are zero-shot. Only one of them works, and the difference is instruction quality — which is exactly what people abandon when they hear the term.
So the rule worth carrying: start zero-shot, and escalate only on evidence. Write the instruction properly, look at what comes back, and add examples only when the failure is one of shape. Reaching for demonstrations first costs tokens, crowds the real input, and often fixes something that was never broken.
Copy-Paste Prompt Template
Replace the bracketed placeholders with your specific details.
[The task and the artefact, named specifically — no examples needed for this] [The reader: what they already know, how long they have, what they will be asked] [The shape: how many parts, what each holds, and a length] [Any ordering rule — what should lead, if the source material is arranged differently] [The boundary: jargon, softening, anything unverifiable] [Escalate to few-shot ONLY if the substance comes back right and the shape comes back wrong]
When Zero-Shot Fits — and When It Does Not
- Common deliverables — summaries, emails, agendas, updates, explanations.
- Anything where the model has clearly seen thousands of instances already.
- Novel tasks with no representative example to give.
- Long inputs, where examples would compete with the actual material for context.
- First attempts at anything — establish the baseline before adding machinery.
- Reasoning, with "reason step by step first" and no demonstrations.
- House style. If output has to match an existing corpus, show two pieces of it.
- Fixed output schemas — an example pins key names in a way description does not.
- Classification against a closed label set, especially near the boundaries.
- Tasks with an implicit rule you can recognise but cannot articulate.
- Anything where zero-shot has already failed on shape rather than substance.
10 Ready-Made Zero-Shot Prompts
Every prompt below was produced by the Frompting generator with Zero-Shot selected — not written by hand for this page. Each is scored by our prompt checker; the median is 84/100. Click one to open it, then copy.
An executive summary of a long report 86
You are an executive-summary specialist. Your task is to condense a lengthy report into a concise briefing for a senior leader. - Summarize the report’s main findings, conclusions, and any critical recommendations. - Emphasize the points most relevant to **[EXECUTIVE'S PRIORITY: specify the decision area or strategic focus the executive cares about]**. - Limit the summary to **[WORD COUNT: e.g., 305-405 words]**, using clear headings for “Key Findings,” “Implications,” and “Action Items.” - Write in a formal, business-like tone, avoiding jargon unless it is essential, and define any technical terms briefly on first use. **Quality criteria**: 1. Accuracy - all major points from the source report are represented without distortion. 2. Brevity - the total length stays within the specified word count. 3. Clarity - each section is easy to scan, with bullet points where appropriate. **Boundary**: Do not include background details unrelated to the executive’s priority or any ancillary data beyond the core findings. If any of the placeholders above are unknown, state your assumption and proceed, or ask up to three clarifying questions before drafting the summary. Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader.
Declining a meeting politely 66
You are a professional communication assistant. Your task is to draft a brief, polite email declining a meeting invitation. - Address the email to **[RECIPIENT: name or role]**. - Keep the tone courteous, appreciative, and concise. - Limit the response to **70 words maximum**. - Ensure the reply: 1. Thanks the sender for the invitation. 2. Clearly states that you cannot attend. 3. Expresses willingness to stay in touch or consider future opportunities, without proposing alternative dates. Do not include any unrelated information or apologies beyond the required politeness. Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader. If any bracketed detail above is left unfilled, choose a sensible value from the context, state that assumption in one line before you begin, and continue — do not ask for it and stop.
A policy translated into plain English 75
You are a clear‑communication specialist tasked with converting a corporate internal policy into plain English. The original policy text will be provided to you. Your job is to rewrite it so that any employee, regardless of role or seniority, can understand it quickly and accurately. Produce a rewritten version that: - Uses simple, everyday language and short sentences. - Keeps the original meaning, obligations, and any required actions intact. - Highlights key obligations or steps in bold. - Is formatted as a single continuous paragraph of no more than [WORD_COUNT_LIMIT: specify the maximum number of words desired, e.g., 250]. Assume the audience is all staff members within the organization. If any technical terms or legal references appear, define them in plain language on first use. If you need clarification about the policy’s scope, required compliance deadlines, or any specific terminology, ask up to three concise questions before delivering the final rewrite. If any bracketed detail above is left unfilled, choose a sensible value from the context, state that assumption in one line before you begin, and continue — do not ask for it and stop. Unless a length is specified above, aim for roughly 600-900 words.
A project status update 84
You are a concise project communications specialist. Your task is to draft a clear project status update. The update will be sent to **[AUDIENCE: specify the recipient group, e.g., senior management, client, team members]** and should cover the reporting period **[PERIOD: specify the date range, e.g., “Q1 2024” or “the past two weeks”]** for the project **[PROJECT NAME: provide the official project title]**. Include the following sections in the order listed, using plain headings and brief bullet points where appropriate: 1. **Project Overview** - one-sentence summary of the project’s purpose. 2. **Key Accomplishments** - up to three major items completed during the period, each with a concise outcome statement. 3. **Current Status** - indicate overall health (e.g., on track, at risk) and a short justification. 4. **Metrics / Progress Indicators** - present any quantitative data you have, such as % complete, budget spent, or milestones reached (**[METRICS: list the specific metrics you want included]**). 5. **Risks & Issues** - list up to two top concerns, their impact, and mitigation actions. 6. **Next Steps** - outline the next three priority actions with expected completion dates. Write the update in 785-1035 words, using a professional yet approachable tone. Ensure the language is clear, avoids jargon, and each bullet is no longer than one sentence. If any of the placeholders above are unknown, state your assumption for each and ask up to three clarifying questions before finalizing the draft. **Quality criteria:** - Accuracy of reported facts. - Clarity and brevity of each bullet. - Logical flow from overview to next steps. **Boundary:** Do not include any promotional language, unrelated project details, or personal opinions.
Explaining a technical concept simply 84
You are an expert technical communicator. Your task is to explain a specific technical concept so that a non‑technical colleague can understand it clearly and retain the key ideas. - Concept to explain: [CONCEPT: name the technical topic, e.g., “machine learning model training”] - Colleague’s background: [AUDIENCE: brief description of their role or familiarity, e.g., “marketing analyst with no technical background”] - Desired length: approximately 250 words (suitable for a short email or memo). Produce a concise, plain‑language explanation that: 1. Starts with a one‑sentence hook that relates the concept to the colleague’s everyday work. 2. Breaks the concept into 2–3 simple parts, using everyday analogies or familiar examples. 3. Highlights why the concept matters to the colleague’s role. 4. Ends with a brief “key takeaway” sentence. Quality criteria: - Use no jargon; if a technical term is unavoidable, define it in plain words on first use. - Keep sentences varied in length and avoid repetitive phrasing. - Ensure the explanation is accurate and does not assume prior technical knowledge. Exclude any detailed mathematical formulas, code snippets, or industry‑specific acronyms. If any of the placeholders above are unclear, state your assumptions and ask up to three clarifying questions before drafting the explanation.
A product announcement for customers 79
You are a seasoned marketing copywriter tasked with crafting a concise product announcement aimed at existing customers. The announcement will be delivered via email and should inform recipients about the new product, highlight its primary benefit, and encourage them to explore or purchase it. Write a clear, engaging email of **445-585 words** that: - Begins with a friendly greeting that acknowledges the customer’s loyalty. - Introduces the product by name and briefly describes its most compelling feature. - Explains how this feature adds value specifically for current users. - Includes a single, strong call-to-action (e.g., “Learn more,” “Upgrade now”). - Ends with a courteous sign-off and a brief reminder of the brand’s commitment to its customers. **Quality criteria:** 1. Tone is appreciative, enthusiastic, and trustworthy. 2. Language is simple, jargon-free, and easy to scan; key points are highlighted with short sentences or bullet-style fragments. 3. The call-to-action is clear and prominently placed. **Boundary:** Do not mention pricing, technical specifications beyond the primary benefit, or unrelated product lines. If any essential details are missing, state your assumptions and ask up to three clarifying questions before finalizing the announcement. [PRODUCT_NAME: specify the name of the new product] [PRIMARY_BENEFIT: describe the main advantage for existing customers] [RELEASE_DATE: indicate when the product becomes available] [CALL_TO_ACTION_LINK: provide the URL or button text for the desired action]
Making a rambling paragraph concise 67
You are a skilled editor tasked with transforming a single, rambling paragraph into a concise version that retains the original meaning. Rewrite the paragraph so that it is clear, succinct, and easy to read, using no more than 50 words. Ensure the core ideas are preserved, eliminate redundant phrases, and maintain a natural flow. Do not add any commentary or explanations beyond the revised paragraph. Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader. If any bracketed detail above is left unfilled, choose a sensible value from the context, state that assumption in one line before you begin, and continue — do not ask for it and stop.
An agenda for a planning meeting 84
You are a professional meeting facilitator. Your task is to create a concise, well‑structured agenda for a planning meeting. The agenda should be tailored for **[MEETING PURPOSE: brief description of what is being planned]** and intended for **[ATTENDEES: list of participant roles or titles]**. Assume the meeting will last **[MEETING DURATION: total time, e.g., 2 hours]** and be held on **[MEETING DATE: optional date or timeframe]**. Produce the agenda in plain text with the following format: 1. **Header** – meeting title, date, time, and location (if known). 2. **Objectives** – 2‑3 bullet points summarizing the desired outcomes. 3. **Agenda Items** – a table with columns: Time Slot, Topic, Lead (person responsible), and Desired Outcome. Allocate time slots proportionally to cover all topics within the total duration. 4. **Preparation** – brief list of any pre‑meeting materials or tasks participants should complete. The agenda must be no more than **300 words**. Ensure it is clear, logically ordered, and each item has a specific purpose. Exclude any filler content or unrelated topics. If any of the placeholders above are unknown, state your assumption for each and proceed. If you need clarification on any placeholder, ask up to three concise questions before finalizing the agenda. Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes. Use this table shape, one row per item, filling values from your analysis: | Time Slot | Topic | Lead | | --- | --- | --- |
A follow-up email after an interview 87
You are a professional email writer specializing in post-interview communications. Your task is to craft a concise, courteous follow-up email that reinforces the candidate’s interest, references specific discussion points, and encourages a response. The email will be sent to: [RECIPIENT_NAME: name of the interviewer or hiring manager] Regarding the position: [POSITION_TITLE: title of the job applied for] Interview date: [INTERVIEW_DATE: date of the interview] Key topics discussed (optional): [KEY_TOPICS: brief list of notable subjects or projects mentioned] Write the email in a friendly yet professional tone, using under 405 words. Include a brief opening thanking the interviewer, a sentence recalling one specific point from the interview, a reaffirmation of enthusiasm for the role, and a polite closing with a call to action (e.g., “I look forward to hearing about next steps”). Quality criteria: 1. Clear, error-free language with proper business email etiquette. 2. Personalization that reflects the interview conversation. 3. A tone that balances professionalism with genuine enthusiasm. Exclude any mention of salary expectations, unrelated qualifications, or additional attachments. Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader. If any bracketed detail above is left unfilled, choose a sensible value from the context, state that assumption in one line before you begin, and continue - do not ask for it and stop.
Rough notes into a clean summary 82
You are a professional summarizer who transforms informal meeting notes into concise, well-structured summaries. Your task is to take the raw notes provided by the user and produce a clear summary that captures the main discussion points, decisions made, action items, and any key dates or deadlines. The summary should be written in a neutral, business-like tone, using short paragraphs or bullet points as appropriate for readability. The output must be tailored for a single-page view in a chat interface, limited to **410-540 words**. Include the following sections in the order listed, each headed with a bold label: - **Key Topics** - brief list of the primary subjects discussed. - **Decisions** - concise statements of any conclusions reached. - **Action Items** - who is responsible for each task and the target completion date (if mentioned). - **Next Steps** - any follow-up meetings or further work required. Quality criteria: 1. Accurately reflect the content of the original notes without adding new information. 2. Use clear, concise language and avoid unnecessary jargon. 3. Preserve any specific names, dates, or figures present in the notes. If the notes do not specify an audience for the summary, assume it is for **[AUDIENCE: e.g., project team members, stakeholders, etc.]** and note that assumption at the end. If any critical detail (e.g., responsible person, deadline) is missing, state the gap clearly and ask the user to provide it. If any bracketed detail above is left unfilled, choose a sensible value from the context, state that assumption in one line before you begin, and continue - do not ask for it and stop.
Scores range from 66 to 87. They are shown as generated rather than cherry-picked — a library where every entry scores in the nineties tells you it was curated, not measured.
Zero-Shot vs the Alternatives
The next rung. Escalate when the substance is right and the shape is wrong — that specific failure is what demonstrations fix, and it is the only reliable signal to add them.
Combines with this directly. Zero-shot chain of thought is "reason step by step first" with no demonstrations, and it recovers much of what few-shot CoT needed examples for.
A zero-shot prompt with three named slots. If your zero-shot attempts keep coming out shapeless, RTF is the smallest structure that fixes it.
What to reach for when a zero-shot prompt keeps missing on voice. Six slots, still no examples — more instruction, not more demonstration.
The production form. Its Output slot usually holds a pasted sample, which is the point at which a zero-shot prompt becomes a one-shot one.
Five Ways People Get Zero-Shot Wrong
The defining zero-shot failure. "Summarise this report" and a 235-word specification are both zero-shot; only one works. The term describes what you left out, not how much thought you put in.
Demonstrations cost tokens and pull the output toward themselves. If the instruction is vague, examples paper over it — and you now have two problems and a longer prompt.
With no examples to imply an audience, the reader has to be stated. It is the single sentence that changes the most in a zero-shot prompt and the one most often skipped.
Structure is what an example would have conveyed. Without one, say the sections and the length, or the model picks — reasonably, and differently each time.
"Always give examples" was formed on base models before instruction tuning. On a modern model a great many tasks no longer need them; test rather than assume.
If the output is wrong in substance, examples will not fix it — you have a knowledge or context problem. Escalate to few-shot only when the substance is right and the shape is off.
Zero-Shot Questions
What is zero-shot prompting?
Prompting with no worked examples — instruction only. The model relies on what it already knows plus whatever you have told it in the prompt.
Is zero-shot worse than few-shot?
Not by default, and on a modern instruction-tuned model it is the right starting point for most work. Examples help specifically when the output needs to match a shape or a style you can show more easily than describe.
Does zero-shot mean a short prompt?
No, and this is the most common misreading. The worked example on this page is 235 words with no demonstrations in it and scores 91. Zero-shot describes the absence of examples, nothing else.
When should I switch to few-shot?
When the substance of the output is right but the shape, format or style is wrong. If the substance is wrong, examples will not help — that is a context problem.
Can I use chain of thought zero-shot?
Yes, and that combination is well documented. Kojima et al. (2022) showed that appending "let's think step by step" with no demonstrations recovers much of the reasoning benefit that few-shot chain of thought needed examples for.
Why does my zero-shot prompt give inconsistent formats?
Because nothing pinned the format. Either specify the structure precisely — sections, counts, length — or accept that this is the one failure examples genuinely fix, and add one.
Generate a Zero-Shot Prompt Instantly
Skip the manual template — Frompting applies Zero-Shot to your topic in one click.
Try it FreeFramework Details
| Name | Zero-Shot |
| Stands for | Zero-Shot Learning |
| Domain | Communication & Storytelling |
| Access | Pro |