What is Prompt Chaining?

AI prompting often feels more like an art than an exact science, especially when you want AI to carry out a complex task.

After all, how many times have you nailed your desired output on the first try?

If you’re like most, it probably took some trial and error, plus your fair share of debugging.

To make your job easier, we’ll be taking a closer look at a prompt engineering technique that will help you unlock AI’s potential – prompt chaining.

Definition of prompt chaining

Prompt chaining is a special prompt engineering technique that uses a sequence of prompts to walk generative AI through a complex task.

Each prompt builds on the previous output, helping AI generate outputs that are more contextually relevant than if you were to enter one huge prompt. (Note: This might come in really handy if you decide to distill a large model like GPT-4o.)

Similar to chunking, this involves breaking down the larger task into smaller, more manageable steps. Doing so allows you to maintain better control over the output while smoothing your path to your desired end goal.

For instance, instead of telling AI to write an entire blog post in one prompt, prompt chaining would involve having AI generate an outline, then expanding each section of the outline.

This approach allows your final output to come out closer to what you’re looking for.

What are the benefits of prompt chaining?

Prompt chaining offers several benefits when it comes to improving the quality of AI-generated content:

• Increased accuracy – Breaking complex queries into smaller parts helps refine AI’s responses, leading to more accurate and relevant results.
• Greater depth of detail – Follow-up prompts allow you to dive deeper into specific aspects, enabling more comprehensive outputs.
• Fewer gaps – Prompt chaining helps you address any gaps or issues in earlier outputs.
• Scalability – Since prompt chaining already made you break a huge task into smaller parts, scaling and tailoring get easier as you can simply add another ‘link’ to the chain if you need. In other words, you just need to add a new follow-up prompt if you want to do something else.
  • Pro tip – If you find that you’re using the same chains repeatedly, you can cache each chain and reduce your LLM costs up to 90%.

What are the drawbacks of prompt chaining?

While prompt chaining is highly effective at getting AI to carry out difficult tasks, there are some drawbacks:

• Management difficulty – Working with a series of interrelated prompts can prove challenging, especially if the chain becomes overly long and intricate. This ends up creating room for error.
• Hyper-dependency on the first prompt – Prompt chaining is naturally reliant on the quality of the preceding prompt in the chain. And if the first prompt is inherently flawed, it will lead to a cascade of failures that compound throughout the entire sequence.
• Time-consuming – AI prompting is already a very iterative task, but prompt chaining takes this to the next level as you need to keep testing to check the quality of the next chainlink. Issues in prompts also become more challenging to find and debug.

How to implement prompt chaining

Prompt chaining requires you to systematically break down larger tasks into smaller, well-defined steps that a GPT can understand.

Here’s a quick walkthrough of how to set up prompt chaining.

Define your goal

Generative AI dislikes ambiguity and loves clarity. As a rule, the more details and context it has, the better the results. (Just don’t overfeed it with too much info).

You should clearly outline what you want the AI to do. This will also help you frame and construct your initial and follow-up prompts. 

Create your initial prompt

Your initial prompt is your most important. It sets the stage for follow-ups and refinements.

The initial prompt should provide a broad overview of the task and the general direction you want AI to take toward your goal.

Review preliminary responses

Once your initial prompt is ready, enter it into the AI you’re using and review the output.

You’ll need to decide if the output aligns with your goal, as well as identify any areas that need extra detail.

If you’re not sure how AI arrived at the output it generated, you can simply ask it to explain its logic.

Importantly, you’ll want to repeat this step several times to confirm that the AI is consistently and reliably generating the output you need.

Refine with follow-up prompts

Using a set of initial responses (we hope you didn’t stop at 1 or 2), create follow-up prompts that target specific areas and details that weren’t covered yet you want to be addressed.

And if any outputs seem lacking and need extra explanation, this is where you ask AI to clarify or expand on what it generated.

Iterate and improve

Adjust your follow-up prompts based on the AI’s responses. This might require changes to the wording or focus of your prompts.

Then repeat this process of generating responses and refining prompts until you reach the level of detail and accuracy you want.

Consolidate

Once you think you’ve built the entire chain, test it.

Check for coherence, relevance, and completeness.

If all’s good, assemble the final prompt chain or outputs and use as you wish.

Example of prompt chaining in action

Here’s a simplified example of what a prompt chain sequence might look like.

You get the idea. In the interest of time and space, we won’t share every single prompt and output.

If you were wondering, we used ChatGPT to generate the outputs.

And lastly, if you were to use ChatGPT like this to create a blog post, we’d recommend editing the final output to remove words that AI overuse.