Prompt design strategies

This page introduces you to some general prompt design strategies that you can employ when designing prompts. While there's no right or wrong way to design a prompt, there are common strategies that you can use to affect the model's responses. Rigorous testing and evaluation remain crucial for optimizing model performance.

Large language models (LLM) are trained on vast amounts of text data to learn the patterns and relationships between units of language. When given some text (the prompt), language models can predict what is likely to come next, like a sophisticated autocompletion tool. Therefore, when designing prompts, consider the different factors that can influence what a model predicts comes next.

Give clear and specific instructions

Giving the model instructions on what to do is an effective and efficient way to customize model behavior. Ensure that the instructions you give are clear and specific. Instructions can be as simple as a list of step-by-step instructions or as complex as mapping out a user's experience and mindset.

Define the task to perform

Describe in detail the task that you want the model to perform. The task can be as simple as a step-by-step list of instructions or as complex as mapping out a user's experience and mindset. The following prompt tells the model to summarize a block of text:

Summarize this text.
Text: A quantum computer exploits quantum mechanical phenomena to perform calculations exponentially
faster than any modern traditional computer. At very tiny scales, physical matter acts as both
particles and as waves, and quantum computing uses specialized hardware to leverage this behavior.
The operating principles of quantum devices is beyond the scope of classical physics. When deployed
at scale, quantum computers could be used in a wide variety of applications such as: in
cybersecurity to break existing encryption methods while helping researchers create new ones, in
meteorology to develop better weather forecasting etc. However, the current state of the art quantum
computers are still largely experimental and impractical.
  
Quantum computers utilize quantum mechanical phenomena to perform computations significantly faster
than traditional computers. By exploiting the particle-wave duality of matter at tiny scales,
quantum computers leverage specialized hardware to solve problems beyond classical physics.
Potential applications include breaking encryption, developing better weather forecasting, and more.
However, current quantum computers are still largely experimental and impractical.

Specify any constraints

Specify any constraints on reading the prompt or generating a response. You can tell the model what to do and not to do. From the previous example, the summarization is shorter than the original text, but maybe you want the summary to be shorter. In this case, you can specify a constraint in the prompt on how long you want the summary to be:

Summarize this text in two sentences.
Text: A quantum computer exploits quantum mechanical phenomena to perform calculations exponentially
faster than any modern traditional computer. At very tiny scales, physical matter acts as both
particles and as waves, and quantum computing uses specialized hardware to leverage this behavior.
The operating principles of quantum devices is beyond the scope of classical physics. When deployed
at scale, quantum computers could be used in a wide variety of applications such as: in
cybersecurity to break existing encryption methods while helping researchers create new ones, in
meteorology to develop better weather forecasting etc. However, the current state of the art quantum
computers are still largely experimental and impractical.
    

Quantum computers use quantum mechanical phenomena to perform calculations much faster than
traditional computers. They have the potential to revolutionize various fields, but current quantum
computers are still experimental and impractical.

Define the format of the response

You can also give instructions that specify the format of the response. For example, you can ask for the response to be formatted as a table, bulleted list, elevator pitch, keywords, sentence, or paragraph. The following prompt instructs the model to give bullet points of the key information in the text:

Summarize this text as bullets points of key information.
Text: A quantum computer exploits quantum mechanical phenomena to perform calculations exponentially
faster than any modern traditional computer. At very tiny scales, physical matter acts as both
particles and as waves, and quantum computing uses specialized hardware to leverage this behavior.
The operating principles of quantum devices is beyond the scope of classical physics. When deployed
at scale, quantum computers could be used in a wide variety of applications such as: in
cybersecurity to break existing encryption methods while helping researchers create new ones, in
meteorology to develop better weather forecasting etc. However, the current state of the art quantum
computers are still largely experimental and impractical.
  
- Quantum computers leverage quantum mechanical phenomena for exponentially faster calculations than
  traditional computers.
- Quantum devices utilize specialized hardware to exploit tiny-scale matter behavior as both
  particles and waves.
- Their operating principles transcend classical physics.
- Potential applications include cybersecurity, weather forecasting, and more.
- Current quantum computers are largely experimental and impractical.

Summary

  • Give the models instructions on what to do.
  • Make the instructions clear and specific.
  • Specify any constraints or formatting requirements for the output.

Include few-shot examples

You can include examples in the prompt that show the model what getting it right looks like. The model attempts to identify patterns and relationships from the examples and applies them when generating a response. Prompts that contain a few examples are called few-shot prompts, while prompts that provide no examples are called zero-shot prompts. Few-shot prompts are often used to regulate the formatting, phrasing, scoping, or general patterning of model responses. Use specific and varied examples to help the model narrow its focus and generate more accurate results.

We recommend to always include few-shot examples in your prompts. Prompts without few-shot examples are likely to be less effective, because they show the model how to apply instructions. In fact, you can remove instructions from your prompt if your examples are clear enough in showing the task at hand.

Zero-shot vs few-shot prompts

The following zero-shot prompt asks the model to choose the best explanation.

Please choose the best explanation to the question:

Question: How is snow formed?
Explanation1: Snow is formed when water vapor in the air freezes into ice crystals in the
atmosphere, which can combine and grow into snowflakes as they fall through the atmosphere and
accumulate on the ground.
Explanation2: Water vapor freezes into ice crystals forming snow.
Answer:
  
Explanation1

If your use case requires the model to produce concise responses, you can include examples in the prompt that give preference to concise responses.

The following prompt provides two examples that show preference to the shorter explanations. In the response, you can see that the examples guided the model to choose the shorter explanation (Explanation2) as opposed to the longer explanation (Explanation1) like it did previously.

Please choose the best explanation to the question:

Question: Why is sky blue?
Explanation1: The sky appears blue because of Rayleigh scattering, which causes shorter blue
wavelengths of light to be scattered more easily than longer red wavelengths, making the sky look
blue.
Explanation2: Due to Rayleigh scattering effect.
Answer: Explanation2

Question: What is the cause of earthquakes?
Explanation1: Sudden release of energy in the Earth's crust.
Explanation2: Earthquakes happen when tectonic plates suddenly slip or break apart, causing a
release of energy that creates seismic waves that can shake the ground and cause damage.
Answer: Explanation1

Question: How is snow formed?
Explanation1: Snow is formed when water vapor in the air freezes into ice crystals in the
atmosphere, which can combine and grow into snowflakes as they fall through the atmosphere and
accumulate on the ground.
Explanation2: Water vapor freezes into ice crystals forming snow.
Answer:
  
Explanation2

Find the optimal number of examples

You can experiment with the number of examples to provide in the prompt for the most desired results. Models like PaLM and Gemini can often pick up on patterns using a few examples, though you may need to experiment with what number of examples lead to the desired results. For simpler models like BERT, you may need more examples. At the same time, if you include too many examples, the model may start to overfit the response to the examples.

Use examples to show patterns instead of antipatterns

Using examples to show the model a pattern to follow is more effective than using examples to show the model an antipattern to avoid.

⛔ Negative pattern:

Don't end haikus with a question:
Haiku are fun
A short and simple poem
Don't you enjoy them?

✅ Positive pattern:

Always end haikus with an assertion:
Haiku are fun
A short and simple poem
A joy to write

Use consistent formatting across examples

Make sure that the structure and formatting of few-shot examples are the same to avoid responses with undesired formats. One of the primary objectives of adding few-shot examples in prompts is to show the model the response format. Therefore, it is essential to ensure a consistent format across all examples, especially paying attention to XML tags, white spaces, newlines, and example splitters.

Summary

  • Including prompt-response examples in the prompt helps the model learn how to respond.
  • Give the model examples of the patterns to follow instead of examples of patterns to avoid.
  • Experiment with the number of prompts to include. Depending on the model, too few examples are ineffective at changing model behavior. Too many examples cause the model to overfit.
  • Use consistent formatting across examples

Add contextual information

You can include in the prompt instructions and information that the model needs to solve a problem instead of assuming that the model has all of the required information. This contextual information helps the model understand the constraints and details of what you're asking for it to do.

The following example asks the model to give troubleshooting guidance for a router:

What should I do to fix my disconnected wifi? The light on my Google Wifi router is yellow and
blinking slowly.
  
1. **Check the Cables:**
- Make sure that all cables are securely connected to your Google Wifi router and modem.
- Try replacing the cables with new ones to ensure they are not faulty.


2. **Reboot the Router:**
- Unplug your Google Wifi router and modem from the power outlet.
- Wait for at least 30 seconds.
- Plug the modem and router back in and wait for them to start up.


3. **Factory Reset the Router:**
- Resetting the router will restore it to its default settings.
- Find the small reset button on the back of the router.
- Press and hold the reset button with a paper clip or similar object for 10 seconds.
- Release the button and wait for the router to restart.


4. **Check Your Internet Connection:**
- Use a wired connection to connect a device directly to your modem.
- If the internet works, the issue may be with your router.
- If the internet still doesn't work, contact your Internet Service Provider (ISP).


5. **Update Router's Firmware:**
- Check if there is a firmware update available for your Google Wifi router.
- Download and install the latest firmware update from the Google Wifi app.


6. **Interference and Obstacles:**
- Make sure the router is not near electronic devices that can cause interference, like microwaves or cordless phones.
- Remove any obstacles between the router and your devices, such as furniture or thick walls.


7. **Change Wi-Fi Channel:**
- In the Google Wifi app, go to Settings > Network & General > Advanced Networking.
- Try changing the Wi-Fi channel to a less congested one.


8. **Contact Google Support:**
- If you've tried the above steps and the issue persists, contact Google support for further assistance.

The response looks like generic troubleshooting information that's not specific to the router or the status of the LED indicator lights.

To customize the response for the specific router, you can add to the prompt the router's troubleshooting guide as context for it to refer to when providing a response.

Answer the question using the text below. Respond with only the text provided.
Question: What should I do to fix my disconnected wifi? The light on my Google Wifi router is yellow and blinking slowly.

Text:
Color: Slowly pulsing yellow
What it means: There is a network error.
What to do:
Check that the Ethernet cable is connected to both your router and your modem and both devices are turned on. You might need to unplug and plug in each device again.

Color: Fast blinking yellow
What it means: You are holding down the reset button and are factory resetting this device.
What to do:
If you keep holding down the reset button, after about 12 seconds, the light will turn solid yellow. Once it is solid yellow, let go of the factory reset button.

Color: Solid yellow
What it means: Router is factory resetting.
What to do:
This can take up to 10 minutes. When it's done, the device will reset itself and start pulsing white, letting you know it's ready for setup.

Color: Solid red
What it means: Something is wrong.
What to do:
Critical failure. Factory reset the router. If the light stays red, contact Wifi customer support.
  
Check that the Ethernet cable is connected to both your router and your modem and both devices are turned on. You might need to unplug and plug in each device again.

Summary

  • Include information (context) in the prompt that you want the model to use when generating a response.
  • Give the model instructions on how to use the contextual information.

Add prefixes

A prefix is a word or phrase that you add to the prompt content that can serve several purposes, depending on where you put the prefix:

  • Input prefix: Adding a prefix to the input signals semantically meaningful parts of the input to the model. For example, the prefixes "English:" and "French:" demarcate two different languages.
  • Output prefix: Even though the output is generated by the model, you can add a prefix for the output in the prompt. The output prefix gives the model information about what's expected as a response. For example, the output prefix "JSON:" signals to the model that the output should be in JSON format.
  • Example prefix: In few-shot prompts, adding prefixes to the examples provides labels that the model can use when generating the output, which makes it easier to parse output content.

In the following example, "Text:" is the input prefix and "The answer is:" is the output prefix.

Classify the text as one of the following categories.
- large
- small
Text: Rhino
The answer is: large
Text: Mouse
The answer is: small
Text: Snail
The answer is: small
Text: Elephant
The answer is:
  
large

Let the model complete partial input

Generative language models work like an advanced autocompletion tool. When you provide partial content, the model can provide the rest of the content or what it thinks is a continuation of that content as a response. When doing so, if you include any examples or context, the model can take those examples or context into account.

The following example provides a prompt with an instruction and an entity input:

For the given order, return a JSON object that has the fields cheeseburger, hamburger, fries, or
drink, with the value being the quantity.

Order: A burger and a drink.
  
```
{
"cheeseburger": 0,
"hamburger": 1,
"fries": 0,
"drink": 1
}
```

While the model did as prompted, writing out the instructions in natural language can sometimes be challenging and it leaves a lot to the model's interpretation. For example, a restaurants menu might contain many items. To reduce the size of the JSON response, you probably want to omit the items that weren't ordered. In this case, you can give an example and a response prefix and let the model complete it:

Valid fields are cheeseburger, hamburger, fries, and drink.
Order: Give me a cheeseburger and fries
Output:
```
{
  "cheeseburger": 1,
  "fries": 1
}
```
Order: I want two burgers, a drink, and fries.
Output:
  
```
{
"hamburger": 2,
"drink": 1,
"fries": 1
}
```

Notice how "cheeseburger" was excluded from the output because it wasn't a part of the order.

Prompt the model to format its response

The completion strategy can also help format the response. The following example prompts the model to create an essay outline:

Create an outline for an essay about hummingbirds.
  
I. Introduction
A. Hook: Begin with an intriguing fact or description of hummingbirds.
B. Provide some background information on hummingbirds.
1. Mention their small size and long, thin beaks.
2. Highlight their unique ability to hover and fly backward.

II. Body
A. Anatomy and Physiology
1. Elaborate on their tiny size, making comparisons to other animals.
2. Describe their wings and how they enable hovering and swift flight.
3. Discuss their metabolism and energy requirements, emphasizing their caloric needs.

B. Feeding Habits and Diet
1. Explain their primary food source, nectar, and how they extract it from flowers.
2. Describe their ability to feed from a wide variety of flowers.
3. Mention their consumption of insects and the nutritional value they provide.

C. Migration Patterns
1. Discuss the incredible long-distance migrations of some hummingbird species.
2. Explain the phenomenon of seasonal migration and the environmental factors driving it.
3. Describe the routes taken by different species during migration.

D. Conservation and Threats
1. Highlight the importance of hummingbirds in ecosystems as pollinators.
2. Discuss the challenges and threats they face, such as habitat loss, nectar scarcity, and climate change.
3. Mention conservation efforts and ways to support hummingbird populations.

III. Conclusion
A. Summarize the main points covered in the essay.
B. Reinforce the importance of hummingbirds and their ecological role.
C. Provide a final thought or call to action to protect these fascinating creatures.

The prompt didn't specify the format for the outline and the model chose a format for you. To get the model to return an outline in a specific format, you can add text that represents the start of the outline and let the model complete it based on the pattern that you initiated.

Create an outline for an essay about hummingbirds.
I. Introduction
   *
  
I. Introduction
* Capture the reader's attention with an interesting anecdote or fact about hummingbirds.
* Provide a brief background on hummingbirds, including their unique characteristics.
* State the main points that will be covered in the essay.

II. The Unique Characteristics of Hummingbirds
* Discuss the hummingbird's exceptionally rapid wing-beat, allowing them to hover and fly backward.
* Highlight their tiny size, making them the smallest birds in the world.
* Describe their long, slender beaks and tongues, allowing them to reach deep into flowers to extract nectar.
* Explain their iridescent plumage, which creates brilliant and vibrant colors.

III. Hummingbird Migration and Distribution
* Discuss the extensive migration patterns of hummingbirds, covering vast distances across continents.
* Describe the different habitats where hummingbirds can be found, including tropical rainforests, deserts, and urban areas.
* Explain the seasonal changes that influence hummingbird migration and habitat selection.

IV. Hummingbird Diet and Feeding Habits
* Focus on the hummingbird's primary food source, which is nectar from flowers.
* Describe their ability to consume insects, spiders, and other small invertebrates for protein.
* Explain how hummingbirds use their long, slender tongues to extract nectar from flowers.
* Discuss the importance of flowering plants in providing sustenance for hummingbirds.

V. Reproduction and Nesting Behavior
* Describe the mating rituals and courtship behavior of hummingbirds.
* Explain the process of nest-building, including the materials used and the unique characteristics of hummingbird nests.
* Discuss the role of the female hummingbird in laying and incubating eggs.
* Describe the development of hummingbird chicks and the role of both parents in providing care.

VI. Threats to Hummingbirds and Conservation Efforts
* Highlight habitat loss, climate change, and deforestation as significant threats to hummingbird populations.
* Describe the role of habitat conservation initiatives in protecting hummingbird habitats.
* Explain the importance of creating hummingbird-friendly gardens and landscapes.
* Discuss the efforts of organizations and individuals working to conserve hummingbird populations.

VII. Conclusion
* Summarize the main points covered in the essay, emphasizing the unique characteristics, migration patterns, diet, reproduction, and conservation of hummingbirds.
* Provide a closing statement that leaves a lasting impression on the reader and reinforces the significance of these fascinating creatures.

Summary

  • If you give the model a partial input, the model completes that input based on any available examples or context in the prompt.
  • Having the model complete an input may sometimes be easier than describing the task in natural language.
  • Adding a partial answer to a prompt can guide the model to follow a desired pattern or format.

Break down prompts into simple components

For use cases that require complex prompts, you can help the model manage this complexity by breaking things down into simpler components.

Break down instructions

Instead of having many instructions in one prompt, create one prompt per instruction. You can choose which prompt to process based on the user's input.

Chain prompts

For complex tasks that involve multiple sequential steps, make each step a prompt and chain the prompts together in a sequence. In this sequential chain of prompts, the output of one prompt in the sequence becomes the input of the next prompt. The output of the last prompt in the sequence is the final output.

Aggregate responses

Aggregation is when you want to perform different parallel tasks on different portions of the data and aggregate the results to produce the final output. For example, you can tell the model to perform one operation on the first part of the data, perform another operation on the rest of the data and aggregate the results.

Summary

  • Break down complex instructions into a prompt for each instruction and decide which prompt to apply based on the user's input.
  • Break down multiple sequential steps into separate prompts and chain them such that the output on the preceding prompt becomes the input of the following prompt.
  • Break down parallel tasks and aggregate the responses to produce the final output.

Experiment with different parameter values

Each call that you send to a model includes parameter values that control how the model generates a response. The model can generate different results for different parameter values. Experiment with different parameter values to get the best values for the task. The parameters available for different models may differ. The most common parameters are the following:

  • Max output tokens
  • Temperature
  • Top-K
  • Top-P

Max output tokens

Maximum number of tokens that can be generated in the response. A token is approximately four characters. 100 tokens correspond to roughly 60-80 words.

Specify a lower value for shorter responses and a higher value for longer responses.

Temperature

The temperature is used for sampling during response generation, which occurs when topP and topK are applied. Temperature controls the degree of randomness in token selection. Lower temperatures are good for prompts that require a more deterministic and less open-ended or creative response, while higher temperatures can lead to more diverse or creative results. A temperature of 0 is deterministic, meaning that the highest probability response is always selected.

For most use cases, try starting with a temperature of 0.2. If the model returns a response that's too generic, too short, or the model gives a fallback response, try increasing the temperature.

Top-K

Top-K changes how the model selects tokens for output. A top-K of 1 means the next selected token is the most probable among all tokens in the model's vocabulary (also called greedy decoding), while a top-K of 3 means that the next token is selected from among the three most probable tokens by using temperature.

For each token selection step, the top-K tokens with the highest probabilities are sampled. Then tokens are further filtered based on top-P with the final token selected using temperature sampling.

Specify a lower value for less random responses and a higher value for more random responses. The default top-K is 40.

Top-P

Top-P changes how the model selects tokens for output. Tokens are selected from the most (see top-K) to least probable until the sum of their probabilities equals the top-P value. For example, if tokens A, B, and C have a probability of 0.3, 0.2, and 0.1 and the top-P value is 0.5, then the model will select either A or B as the next token by using temperature and excludes C as a candidate.

Specify a lower value for less random responses and a higher value for more random responses. The default top-P is 0.95.

Prompt iteration strategies

Prompt design is an iterative process that often requires a few iterations before you get the desired response consistently. This section provides guidance on some things you can try when iterating on your prompts.

Use different phrasing

Using different words or phrasing in your prompts often yields different responses from the model even though they all mean the same thing. If you're not getting the expected results from your prompt, try rephrasing it.

Version 1:
How do I bake a pie?

Version 2:
Suggest a recipe for a pie.

Version 3:
What's a good pie recipe?

Switch to an analogous task

If you can't get the model to follow your instructions for a task, try giving it instructions for an analogous task that achieves the same result.

This prompt tells the model to categorize a book by using predefined categories.

Which category does The Odyssey belong to:
thriller
sci-fi
mythology
biography
  
The answer: mythology

Explanations:

The Odyssey is an epic poem in ancient Greek literature. It is one of two major ancient Greek epic
poems attributed to Homer. The Odyssey is an epic poem that tells the story of Odysseus, a Greek
hero who fought in the Trojan War and then spent ten years trying to return home to Ithaca. The
Odyssey is a work of mythology.

The response is correct, but the model didn't stay within the bounds of the options. You also want to model to just respond with one of the options instead of in a full sentence. In this case, you can rephrase the instructions as a multiple choice question and ask the model to choose an option.

Multiple choice problem: Which of the following options describes the book The Odyssey?
Options:
- thriller
- sci-fi
- mythology
- biography
  
The answer is: mythology

Change the order of prompt content

The order of the content in the prompt can sometimes affect the response. Try changing the content order and see how that affects the response.

Version 1:
[examples]
[context]
[input]

Version 2:
[input]
[examples]
[context]

Version 3:
[examples]
[input]
[context]

Fallback responses

A fallback response is a response returned by the model when either the prompt or the response triggers a safety filter. An example of a fallback response is "I'm not able to help with that, as I'm only a language model."

If the model responds with a fallback response, try increasing the temperature.

Things to avoid

  • Avoid relying on models to generate factual information.
  • Use with care on math and logic problems.

Next steps