Skip to content

Writing good prompts

Learn how to write prompts that reflect what your customers actually ask AI, with practical examples and the common mistakes to avoid.

Updated

A good prompt sounds like something a real customer would ask ChatGPT while researching a problem, comparing options, or considering a purchase.

The closer your prompts are to those real questions, the more useful your results become. You don't need complicated wording. You need questions that reflect what your customers actually want to know.

What makes a good prompt?

There are five things worth keeping in mind when writing prompts:

  • Write like your customers. Use natural language and focus on a real need. For example: "What's the easiest accounting software to use as a freelancer?"
  • Ask one question at a time. Combining pricing, features, and customer support in one question makes the answer harder to evaluate. Give each need its own prompt.
  • Keep it short and specific. Good prompts rarely need to be long. As a general rule, aim for fewer than 25 words.
  • Use your customers' language and market. Write questions the way people in your selected country would ask them. A customer in the UK may phrase things differently from someone in the US.
  • Include context when it matters. Company size, location, audience, or budget can make a question more relevant. For example: "What's the best payroll software for a company with 20 employees in Manchester?"

The main thing is that the question should feel realistic. If you can't imagine a customer actually asking it, it's probably worth reconsidering.

Examples: before and after

A few small changes can make your prompts much more useful.

Less usefulBetterWhy
"accounting software""What's the best accounting software for a freelancer who's just starting out?"A keyword doesn't tell you much about the customer's needs. The second question includes an audience and a specific situation.
"office cleaning prices""How much does office cleaning cost for a 200-square-metre office in Manchester?"The size and location make the question more realistic.
"best ecommerce platform""Which ecommerce platform is best for a small UK shop selling 100 products?"The question makes it clear what kind of business needs the solution.
"Which ecommerce platform is the best, cheapest, and has the best support?""Which ecommerce platform is easiest to use for someone starting an online store?"Focusing on one need makes the answer easier to evaluate and compare.
"Act as an expert and list the five best accounting tools""Which accounting tools would you recommend for a small UK business?"The second version sounds more like something a customer would naturally ask.

Not every prompt needs to be highly specific. Broader questions can also be useful, especially when customers are still exploring their options.

What matters is having a good mix.

Use different types of questions

Customers ask different questions depending on where they are in the buying journey.

Some are trying to understand a problem. Others are comparing solutions, while some are ready to buy or sign up.

That's why it's useful to track questions across all three stages:

  • Research: "How do I manage VAT when starting a small business?"
  • Compare options: "What's the best accounting software for small businesses?" or "How does Xero compare with QuickBooks?"
  • Take action: "Where can I buy accounting software for a sole trader?"

As explained on the previous page, you can organize these questions using topics, personas, and stages.

Most of your prompts should be unbranded, meaning they don't mention your brand or product names. This helps you understand whether AI recommends your company without being asked about it directly.

Branded prompts are still useful when you want to know what AI says about your company specifically.

Common mistakes to avoid

A few mistakes tend to come up when teams start creating larger sets of prompts:

  • Only tracking branded questions. If every prompt contains your company name, you won't learn whether AI would recommend you over competitors without being asked.
  • Using internal terminology. Avoid product names, abbreviations, and technical language your customers wouldn't use. Write from the customer's perspective.
  • Combining multiple questions. Focus on one need at a time. It makes the answers easier to understand and compare.
  • Creating too many similar prompts. Ten nearly identical questions rarely give you ten times the insight. Look for questions covering genuinely different needs instead.
  • Giving AI instructions. Avoid phrases like "Act as an expert" or "Write a list of five recommendations". Ask the question as a customer would.

A useful test is: Would a potential customer actually type this into ChatGPT?

If not, there's probably a more natural way to phrase it.

How Klarvu runs your prompts

Every prompt is sent to the AI engines as a new, independent question. There's no previous conversation for the AI to refer to.

That means each question needs to make sense on its own. For example, avoid "What about a cheaper option?" without explaining what kind of product or service you're looking for.

Klarvu's own suggestions follow the same approach: natural customer questions in your selected language and market, one question at a time, and fewer than 25 words.

Add and improve your prompts

You don't need to get every question right from the beginning. Start with a relevant set, then refine it as you learn what works.

Start with real customer questions

Talk to your sales, support, and marketing teams. What questions do customers ask again and again?

Use "Add prompt" to include them in Klarvu. Questions you've actually heard from customers are often a better starting point than guessing what people might ask.

Get more suggestions from Klarvu

Use "Suggest more" to generate additional prompts matched to your topics and personas.

Review the suggestions and choose questions that add something new rather than repeating what you're already tracking.

Review the answers and make adjustments

After your prompts have been running for a few days, take a look at the results.

Are the questions producing useful answers? Are they relevant to your business? Remove prompts that aren't working, and try a different wording as a new prompt where needed.

When you remove a prompt with "Remove", its previously collected answers are preserved.

A useful set of prompts rarely comes together on the first attempt. It improves as you learn which questions tell you the most about your visibility in AI-generated answers.