One highly practical way that artificial intelligence (AI) agents can be utilised is for troubleshooting technical problems.

From everyday office workers seeking help with complex Excel formulas to software developers resolving broken code, AI is increasingly being used as a first-line diagnostic tool to speed up issue resolution.

However, this use of AI is not always straightforward, and relying blindly on its output can often lead to frustration. Using AI effectively requires at least some technical know-how and an understanding of how to craft the right queries. This page explains how you can use AI for technical troubleshooting, including how to develop a good prompt, and how to verify the information you receive.

How Can AI Help?

Using AI agents to troubleshoot can speed up issue resolution because they can:

• Analyse complex systems quickly;

• Identify potential root causes of problems; and

• Suggest ways to resolve those problems, including providing step-by-step instructions.

For example, you can show an AI agent a screenshot of an error message, and ask it to analyse the error message, identify the potential problem or issue, and then suggest several ways to resolve it, with detailed instructions.

However, there is a drawback.

As our pages on Understanding Large Language Models (LLMs) and Using LLMs make clear, these types of AI agents—which include ChatGPT—don’t know anything. LLMs work by putting together words in the most logical order in response to a prompt or query. They have been trained on a LOT of data, including all the pages of the internet, which can sometimes mean that they sound like they have knowledge. However, that is not true.

This means that you may find that you are going round in circles with the AI agent providing what sounds like useful instructions that actually lead nowhere.

The user reporting the text concluded that it “often feels like ChatGPT is reciting Reddit threads from 9 years ago and doesn’t know how to pinpoint problems…”. However, others have reported success using ChatGPT—but only with persistence and the use of a few practical tips.

---

---

Practical Tips for AI Troubleshooting

What, therefore, can you do to improve your AI troubleshooting?

Some useful practical tips include:

• Provide enough context for the AI agent to understand your request

  As always when you are developing a prompt for AI, make sure that you give it enough context. Describe the issue, the environment, and any error messages in detail.

  There is more about how to develop good prompts in our page on Using AI Effectively: Prompt Engineering.

• Start simple and work up to more complex requests

  Don’t overcomplicate your first prompt.

  Ask your chosen AI agent to give you a basic overview first, to make sure that it has properly understood your request. Once you are clear that you are both working on the same problem, you can start exploring some of the options provided in more detail.

• Review, test and iterate until you get something workable

  You will very rarely get a usable solution from your first prompt.

  Instead, you will probably need to review the answer, test any proposed solutions, and iterate if it doesn’t work. If so, tell the AI what happened, and provide further screenshots or details of error messages. Ask it to refine its solution further, in the light of this new information.

• Pick and choose your AI agent

  All AI agents are not the same. It may therefore be worth trying different AI agents for different problems.

  For example, Microsoft’s Copilot is likely to be much better at troubleshooting problems in Windows than a more general AI agent like ChatGPT. This is because it has been developed to help Microsoft users to get more out of Microsoft products.

  Google Gemini is linked to Google Search, and can therefore find more up-to-date information from the internet, and provide you with links to the sources of its information. This makes it easier to check back and make sure that it is talking about the same problem as you.

• Use audit and performance logs to spot anomalies

  AI agents excel in scanning large amounts of data to detect patterns and anomalies.

  You can therefore give your AI agent access to audit and performance logs, and ask it to review them and spot anomalies. For example, Google suggests that you can use Gemini to help you to troubleshoot slow database performance using logs. The AI might, for example, identify specific queries that are inefficient, and suggest a partitioning strategy.

  WARNING! Don’t share any sensitive information

  ---

  Remember, whatever AI agent you are using, don’t share any sensitive information.

  At work, that includes any details of clients or any company data. At home, just be careful that you haven’t included any information that could identify you or your computer.

• Choose the right problems to troubleshoot

  There are some circumstances when AI excels—and others where it will be largely useless.

  You need to pick your moment (or rather, use case).

  For example, AI is very good in predictable environments, where historical patterns and experience largely predict future performance. However, it is not very good in more variable or complex situations, such as when ethical frameworks come into play. You also need to be confident that you have suitable data and information to provide, and that a ‘wrong’ answer is unlikely to be catastrophic.

The Effect of AI

Does using AI for troubleshooting actually work? That is, can it improve the quality, speed up the process and make things more efficient?

Experience at SkillsYouNeed would suggest that this is true.

However, our ‘sample of one’ is also supported by academic research among technicians involved in engineering projects.

A study on the effect of using AI

In 2025, a paper was published in the International Journal of Product Research, entitled “Can a troubleshooting AI assistant improve task performance in industrial contexts?”.

The authors reported on a study among 19 technicians completing 143 tasks in the area of train commissioning and manufacturing.

These technicians were highly skilled, and used their expertise to diagnose and solve problems detected during the commissioning process in train manufacturing. The authors developed a specific AI chatbot that used a technique called Retrieval Augmented Generation, to support troubleshooting during the process by giving access to detailed technical information.

The authors found that users supported by AI scored significantly higher both overall and for individual tasks than those without AI. In other words, their solutions were more accurate. However, the number of tasks achieved was very similar, suggesting that AI did not speed up the process of troubleshooting even though it improved the quality.

Source: Löwhagen, N., Schwendener, P., & Netland, T. (2025). Can a troubleshooting AI assistant improve task performance in industrial contexts? International Journal of Production Research, 1-22.

This study therefore suggests that using AI can improve the quality of troubleshooting, if not the speed.

---

A Final Thought

AI tools have already been proven to work to support troubleshooting.

However, this does not mean that every AI agent will work in every troubleshooting scenario.

Pick your moment, pick your tool, and be prepared to iterate extensively to achieve success.

---