Unlocking the intelligence of Azure Open AI
AlfaPeople DK |
Sep 20, 2023

Unlocking the intelligence of Azure Open AI

Have you ever wondered about the strategies to achieve the best results with ChatGPT and Azure Open AI?

Its natural language understanding capabilities make it valuable for language learning, customer support chatbots, and virtual assistants. On the other hand, it occasionally produces incorrect or nonsensical responses and may exhibit inconsistent behavior.

In this blog, you can discover how to optimize your interactions with ChatGPT by creating efficient request descriptions and leveraging smart techniques for processing the information obtained (Azure Open AI).

Devloping communication skills with ChatGPT: Mastering effective request descriptions

As a user of ChatGPT (Azure Open AI), it is essential to understand that how we formulate our requests greatly influences the quality of the generated responses. Providing clear and specific prompts can guide ChatGPT to answer our questions appropriately and accurately.

To enhance your experience, follow these guidelines for creating efficient request descriptions:

  • Define the goal of the question: By establishing your objective or the purpose of your question, you can provide context, guiding it to provide a more targeted and relevant response.

    Example: Meet the requests of my clients

  • Be clear and specific: Articulate the information you seek clearly, avoiding ambiguities and making your request precise and detailed.

    Example: Identify the customer’s name and the intent of the request [customer’s request.

  • Provide context: Share relevant information, defining the role you want ChatGPT to perform. This context enables the system to generate targeted and useful information.

    Example: You are an agent and have received a request from a customer or a lead. [customer’s request]

  • Specify constraints: Provide relevant constraints, rules, and limitations. This helps to limit the scope.

    Example: You are an agent who only deals with financial matters and has received a request from a customer or a lead. Analyze the request and identify if it is a financial request. [customer’s request]

  • Break down complex requests: Consider dividing them into sub-questions if you have a complex request.

    Example: ‘You are an agent who only deals with financial matters and have received a request from a customer or a lead. Analyze the request by identifying:
    • Is it a financial request?
    • Who is the requester?
    • Sentiment analysis
    • [customer’s request]

  • Provide examples: If you are looking for specific formats, structures, examples, illustrations, or comparisons, mention them explicitly. ChatGPT can provide more accurate and targeted responses when given clear instructions about the desired output format.

    Example: ‘You are an agent who only deals with financial matters and have received a request from a customer or a lead. Analyze the request by identifying:
    • Is it a financial request? Select one of the following options YES or NO
    • Who is the requester? Just the requester’s last name
    • Sentiment analysis: Select one of the following options VERY HAPPY, HAPPY, NEUTRAL, UNHAPPY, or VERY UNHAPPY
    • [customer’s request]’

    • Test and refine: It’s important to test and refine your questions over time. This iterative process allows you to adjust the content of the request, increasing the chances of receiving a relevant and focused response.

Remember, the more precise and well-defined your request is, the better ChatGPT will be able to understand your intention and generate accurate and relevant responses.

Intelligent data exploration: Efficiently analyzing user responses

To further enhance your experience with ChatGPT, take advantage of information analysis techniques to expedite the processing of user responses.

Here are some examples of how to analyze and tabulate user information:

  • Sentiment Analysis: Determine the sentiment expressed in the user’s response (positive, negative, neutral). This analysis helps to understand the emotional state or level of customer satisfaction regarding the topic discussed.

    Example: ‘…Analyze the request identifying:

    Sentiment Analysis: Select one of the following options VERY HAPPY, HAPPY, NEUTRAL, UNHAPPY, or VERY UNHAPPY

    [customer’s request]’
  • Intent Classification: Classify the user’s intention or purpose based on their response, distinguishing between categories like questions, complaints, suggestions, feedback, or support requests.

    Example: ‘…Analyze the request identifying:

    Intent Classification: Select one of the following options: QUESTION, COMPLAINT, or PRAISE

    [customer’s request]’
  • Entity Extraction: Identify specific entities mentioned in the user’s response, such as names, dates, places, or product names. This analysis helps to extract actionable information and can be useful for further processing or categorization.

    Example: ‘…Analyze the request identifying:
    First Name of the Requester
    Last Name of the Requester
    [customer’s request]’
  • Keyword Extraction: Identify terms or phrases representing the main topics or subjects discussed in the user’s response.

    Example: ‘…Analyze the request identifying:
    2 main Keywords
    [customer’s request]’
  • Language Detection: Determine the language used in the user’s response, enabling specific language processing or routing to the appropriate language-specific support teams.

    Example: ‘…Analyze the request identifying:
    Language of the request
    [customer’s request]’
  • Topic Categorization: Categorize the user’s response into predefined topics or themes to understand their primary areas of interest or concern. This analysis allows for the organizing and routing user input to relevant departments or processes.

    Example: ‘…Analyze the request identifying:
    Up to 2 main Topics
    [customer’s request]’
  • Content Summarization: Generate a concise summary of the user’s response to capture the key points or main ideas discussed.

    Example: ‘…Analyze the request identifying:
    Content Summary
    [customer’s request]’
  • Intent Ranking: Rank the user’s intentions based on their response to prioritize actions accordingly.

    Example: ‘…Analyze the request identifying:
    Up to 3 requester’s intentions ranked by priority with up to 10 words each
    [customer’s request]’
  • Conversation Context Tracking: Can refer to previous messages and specific details, recall user preferences, or provide follow-up information based on the ongoing conversation.

    Example: ‘…Analyze the request identifying:
    Conversation Context Tracking with up to 20 words
    [customer’s request]’

By implementing these analysis techniques, you can efficiently process and categorize user responses for the information system.

Azure Open AI.

Practicing the suggestions for creating requests and receiving coherent and accurate responses

For this blog, the following request was created:

  • ‘You are a customer or a lead of my company, and you have exposed your problem. Analyze the customer’s problem by identifying the following points. Turn the points in bold font and the analysis in regular font.
  • Language detection in bullets
  • Sentiment Analysis in bullets
  • Intent Classification in bullets
  • Suggested Ticket Subject
  • Content Translated to English
  • Content Summarization
  • Entity Extraction in bullets
  • Keyword Extraction in bullets
  • Topic Categorization in bullets
  • Intent Ranking
  • Conversation Context Tracking
  • [customer’s request]’

Request 1:

‘My father is trying to log in to the website, but it’s impossible, even changing the password. When he enters the data it shows “oops! some incorrect information”. I have already contacted you to cancel the service, and you ignore my request. A disgrace!
Ah-kum Silva’


Request 1


ChatGPT response

Request 2:

‘I graduated from the first class of the Bachelor of Nursing course at the Faculdade da Santa Cruz de Minas on December 29,2021. My graduation ceremony occured in August 2022, and I still have not received my diploma. We have a WhatsApp group for my class, and the course coordinator is included in this group. The last update she gave us was in May 2023, stating that there was a change in the certifier previously used because it did not meet the MEC (Ministry of Education) prerequisites, while the college had previously been operation with postgraduate courses – how were the postgraduate diplomas issued if the certifier did not meet the MEC prerequisites? Well, I don´t understand.

Moreover, in the same message, we were not given any deadline for issuing the diplomas. I am in the final stage of getting the job I have always dreamed of, and I am depending on this diploma. In other words, I am being prevented from getting a job!

I have sent an email to the secretary, where he gave me a deadline of 180 days. Still, Ordinance No. 1.095, from 2018, of the MEC establishes the deadlines for the issuance and registration of the diploma within 60 days, counted from the date of graduation.

I leave my complaint and await contact.


Blaine Wolfeschledorff’


Request 2


Response from ChatGPT

ChatGPT is a powerful tool that relies on human input to guide responses.

With the right approach, you can demystify ChatGPT and leverage its vast resources to achieve your goals effectively.

If you have any questions about Microsoft AI, Azure Open AI or related products and services, reach out to us at telephone +45 70 20 27 40 or send us a message here.