Learn how the AI Analyzer helps you prompt AI quickly and effectively with this video walkthrough and how-to guide.
When you properly prompt your Hatch AI bots, they conduct more effective and meaningful interactions with your customers, while also saving you time in the building and testing process, but AI prompting isn't always intuitive. Our AI Analyzer tool will help you.
How does the AI Analyzer work and how do I use it?
When you create a Hatch Assistant bot, you create interactions for the bot to use. These interactions are made up of questions and response validation prompts. When adding these prompts, you’ll see a button in the bottom left corner of the interaction panel.
Click the AI Analyzer button (circled in the image above) to grade your prompts and get instant feedback.
Clicking this button initiates the AI Analyzer, which will then rate the prompts on a three-tiered grading scale.
The AI Analyzer grades each of your prompts as “Excellent,” “Good,” or “Needs Improvement”—and provides the reasons why—so you can get each of your bots to behave exactly the right way, every time, and turn them into your biggest asset.
To understand more fully how our tool works, let’s look at an example. We want this bot to get the customer’s address, so the prompt we’ve entered is “Ask to verify the home address.” As you can see, the AI Analyzer has rated this prompt as “Excellent” with the reason being “The instruction is simple and direct, making it easy for an AI to generate a question.”
You also have to prompt your bot with what defines a valid response to the question. So if we type in “Customer replies with answer,” the AI Analyzer tells us this prompt “Needs Improvement” because “The rule is too vague and does not specify what kind of answer or format is expected.”
If we then type in “you need to get the full address including city, zip code and state,” the AI Analyzer tells us this is better, but still doesn’t get the best result. The reason being, “The rule is somewhat ambiguous because it does not specify what constitutes a 'full address' beyond city, zip code, and state. It lacks clarity on whether street name and number are required.”
Now if we type in “the full address must include street name and number, city, zip code and state,” the AI Analyzer tells us “The rule is direct and covers all necessary components of a full address without ambiguity.” We’re good to go!
Now, the full interaction (which includes both the question and response validation) is properly set up:
When you properly prompt your Hatch AI bots, they can conduct more effective and meaningful interactions with your customers, while saving you time in the building and testing process.
AI Analyzer FAQs
What is an AI prompt?
An AI prompt is an instruction you give to AI so that it yields the desired output. In the case of Hatch Assistant, the "desired output" is the conversational text messages it sends to customers to collect their information and move them through your funnel.
Hatch is built on a large language model (LLM) — the same one ChatGPT is built on, actually. When you give it a prompt like "get the customer's full address," it then uses uses the LLM to turn that prompts into a conversational text like "May I have your full address please?"
Why do I need to be able to do AI prompting?
AI prompts are not always intuitive. For example, when creating a Hatch Assistant bot, if you want your bot to ask a customer what their minimum budget is, you may be inclined to enter “what is your minimum budget?” when you actually should enter “find out their minimum budget.” You are not prompting the customer, you are prompting the bot to ask the customer questions. Our AI Analyzer helps you to correct these mistakes and makes it easy for you to control the style, tone, as well as the overall response of the AI bots so that they interact with your customers exactly the way you want.
Why do I need the AI Analyzer?
The AI Analyzer simplifies and shortens the time it takes for you to build, test, and launch your Hatch AI chatbots. It evaluates the question and validation prompts you’ve given your bot and immediately provides feedback. If your prompt is “Good” or “Excellent,” the analyzer tells you why it works, helping you learn for future bot builds. If it “Needs Improvement,” the analyzer will explain why and help you make the appropriate fix so that the bot behaves correctly.
How does the AI Analyzer grade my prompts?
Our Hatch AI engineers have been working with AI for a long time and are constantly staying up to date on the LLM (large language model) that Hatch AI is built on (which is the same one used to make ChatGPT) They have created a rule system for the AI right inside the product, based on prompting best practices and their intimate knowledge of how Hatch Assistant works.
The system runs the prompt through those rules to determine if the prompt falls into one of three categories: Needs Improvement, Good, and Excellent.
Do I need to be able to code to use the AI Analyzer?
Nope! Hatch Assistant AI and the built-in AI Analyzer are a no-code-needed kind of deal. Simple text is enough. However, it may still take some time to adjust to the exact wording and syntax needed to get the AI to understand what you're asking for in your prompts. The AI Analyzer is made to help with this.
How does this help me do my job?
The feedback provided by the AI Analyzer is meant to help you write better instructions for your bot. It also simplifies the bot testing process, saving you time and enabling you to check the prompt quality before you even run the test conversation. Plus, it empowers you to train other team members on Hatch AI more quickly.
What is the difference between “Good” and “Excellent” prompting?
Ideally, you want your prompts to be “Excellent,” but due to the nature of AI, there will be times where “Excellent” just isn’t possible—and that’s okay. In these scenarios, “Good” is enough. Just be sure to test the prompt and make sure your bot is giving you the outputs you’re looking for.
Keep in mind that the AI Analyzer is meant to help you improve prompt performance, but it’s not the end-all-be-all. Like any other AI tool, it works best with human input. So use it as a guide and know that there may be some scenarios where the analyzer tells you one thing but your own experience tells you another. Trust yourself to do the work and let the AI support you.