Mastering AI for Construction: How to Avoid Common Prompting Mistakes and Leverage RAG Search for Better Results

The Role of AI in Building Design

Artificial Intelligence (AI) is a powerful tool that many people nowadays use daily. Many construction professionals use AI tools even in their work to streamline work processes and access information quickly. In the modern workspace, AI tools are quickly becoming the new normal and are helping workers become more productive than ever before. But whether you use AI for drafting reports quickly, analyzing data, or finding specific guidelines, it’s crucial to learn how to use them properly to gain the maximum benefit.

Prompting is the method of communicating requests to an AI and is essential in harnessing the full potential of these tools. However, prompting isn’t easy.
Many users, and even more experienced ones, often make mistakes in prompting that can lead to less accurate results and frustrating experiences with AI tools. Understanding how to prompt more effectively can allow you to take full advantage of AI’s advanced capabilities in your work.

Understanding Retrieval Augmented Generation

Retrieval Augmented Generation (RAG) search can revolutionize your workflow, blending the power of AI with reliable, up-to-date information. RAG search is a powerful AI approach that combines the best of both worlds: it retrieves information only from pre-defined relevant sources. It generates responses in natural language based on that information. This significantly reduces the risk of getting misleading answers, known as AI hallucinations, which are more common with traditional AI models, such as the base GPT4. Thus, RAG is essential for anyone working with AI in work that demands accuracy, such as navigating regulations and legal requirements.

For construction professionals, RAG can streamline processes by combining external compliance documents and internal guidelines in a single search. It's a simple yet highly efficient way to ensure your work meets the latest regulations and client requirements.

Now let’s understand the common mistakes people make in prompting in RAG search and how to improve it.

1. Being Too Vague

Ineffective Prompt: “Tell me about building codes”

The mistake here is not giving the AI enough information or context about the question.

To improve this prompt, be specific and provide context. This could be, for example, providing information on the type of building or location of the site you are looking for information on. Sometimes you already know which source or database the information can be found in, which you can also specify in your prompt to guide the search in the right direction.

Effective Prompt: “What are the fire safety requirements for apartment buildings in the city of Helsinki?”

When you give the AI more context and information, it can provide more specific answers with more detail. This will help the AI answer your specific question rather than giving vague and general answers.

2. Overloading the Prompt With too Many Questions

Ineffective Prompt: "I'm building a mixed-use building with residential units and a ground-floor commercial space in Helsinki, Finland. What are the fire safety requirements, occupancy limits, accessibility standards for the commercial space, and any environmental or energy efficiency regulations I need to comply with according to the latest Finnish building codes and standards?"

While providing context is important, it doesn’t mean that a longer prompt is always better. Asking multiple questions in one prompt usually leads to poorer answers. This prompt combines four complex questions into one, covering fire safety requirements, occupancy limits, accessibility standards, and environmental regulations.

By overloading the AI with information, it may struggle to provide a comprehensive answer that adequately addresses all these areas in a single response. AI might lack focus because so many topics are included and provide a general answer that doesn't delve deeply into any specific area, reducing the usefulness of the information.

Additionally, the benefit of using AI is to piece large amounts of information into smaller chunks that are easier for you to comprehend. By asking one question at a time, you make it easier for your brain to process each answer before moving on to the next topic.

Split the questions into separate prompts and ask them individually from the AI. Avoid overloading the AI by providing too much information in one prompt.

Effective Prompts:

Safety Regulations: "What are the fire safety requirements for a mixed-use building with residential units and a ground-floor commercial space in Helsinki?"

Occupancy Regulations: "What are the occupancy limits for multi-family residential units in urban areas under Finnish building regulations?"

Accessibility Standards: "What are the accessibility requirements for commercial spaces in mixed-use buildings in Helsinki?"

Environmental Regulations: "Which environmental or energy efficiency regulations for new mixed-use constructions in Helsinki do I need to comply with?"

When you divide the overloaded prompt into separate questions, you get clearer and more comprehensive answers. This is because the AI can focus better, reducing the risk of the AI providing incomplete or generalized answers. It is thus important to find the balance between providing the AI with enough context without overloading it to achieve the best answers.

Tip:
You can even give the AI some constraints to help reduce the amount of information it needs to go through and improve the relevance of the answer provided by the AI. For example, you could ask “What are the building permit requirements in Helsinki, but exclude anything related to multi-family homes?”

However, with software like Complink, you don’t have to type out all the specifics every time. It includes built-in filters and tools that automate much of this process, making it easier to get the exact answers you need ensuring a faster workflow.

3. Using Subjective Language

Ineffective Prompt: "Tell me about the best materials for building."

The question has subjective language, which can lead to the AI providing vague or biased answers. This is because what is considered “best” depends on so many different factors. Focus on measurable criteria, like strength, durability, or industry standards, instead of subjective terms like “best”.

Again, avoid being vague and be specific about the materials you’re asking for and how they’ll be used. Are you thinking about the structure of the building, the facade, or even indoor surface finishes? This will have a dramatic effect on how useful your answer will be. If you want to know about all of these, stick to the habit of asking one question at a time, as discussed in section 3.

Effective Prompt: "What are the recommended materials for load-bearing walls in residential construction?"

When you ask for measurable attributes such as durability and specify the context, it makes it easier for the AI to provide an unbiased, useful answer.

4. Not Iterating on Your Prompts

Ineffective Prompt: "Explain the safety regulations for staircases."

The AI will likely give you a lengthy, unstructured paragraph that is hard to follow and lacks specific details. A common mistake is accepting the confusing or poorly formatted answer without attempting to refine or clarify the prompt. Also, as discussed earlier, not providing additional instructions to guide the AI can lead to a less useful response.

Don’t settle for information that is hard to digest, and recognize that the initial response isn't meeting your needs in terms of clarity or format. Modify your prompt to request the information in a specific format or to focus on particular aspects that help the AI meet your needs in its answer.

Effective Prompt: "Could you provide the safety regulations for staircases in commercial buildings in Turku? List the key requirements in bullet points."

By iterating on the prompt, you guide the AI to provide you with the answers you want in the way that you want them presented, and is easier for you to comprehend. It’s always good to iterate and test out new ways of prompting, especially if the initial answers were not what you were looking for. This helps the AI tailor the response to your needs.

Conclusion

Getting the prompting right is key to getting the most out of AI tools such as RAG Search. By understanding and avoiding the common mistakes of being too vague, overloading the prompt, using subjective language, and not iterating on your question you can turn your AI experience into one that gives you highly relevant and useful information. So, you get the exact info you need whether you’re looking for information on regulatory compliance or project-specific guidance.
RAG Search with good prompting becomes a superpower, simplifying complex queries, combining multiple data sources, and streamlining your workflow. As AI becomes more and more a part of modern construction management, getting your prompting skills up to scratch will mean you can solve problems faster and make better decisions.

Start implementing these tips today on Complink, and you can quickly see how AI can help you streamline operations, minimize errors, and keep your projects aligned with the latest standards and regulations.

Book a demo