Skip to content
The Citation Journal

Manager guidesArticle

How to write prompts that actually measure something

The prompts decide what the visibility measurement tells you. How to write prompts that sound like real customer questions, and how to set them up in Manager.

Krister RossFounder & CEO, CitationLab

Published approx. 4 min read

In this article+

Curious how AI talks about your brand?Run a free visibility check

A visibility measurement is never better than the questions it is built on. The prompts are the questions Manager asks ChatGPT, Gemini, Perplexity and Google on behalf of your customers. If they match what customers actually ask, you get a picture of the market. If they miss, you get numbers that look precise but don't mean anything.

The short version: Write prompts the way your customers talk, with context such as industry, size, location and need. Keep most of the weight on questions without your company name. Spread them across the whole buyer journey, and tag them with the right stage and focus in Manager.

Three examples of weak and strong prompts

  • Weak: "What is YourCompany Inc.?" A question with your company name tells you how AI describes you, but nothing about whether you get recommended.
  • Weak: "Accounting". No customer asks like that, and the answer is a general explanation with no providers.
  • Strong: "Which accounting firm is right for a small business in Manchester that wants a fixed monthly price?" A real buying question with context, where the AI model has to choose and name providers.

Five signs of a good prompt

1. It sounds like a real question

People write longer and more naturally to AI than to Google. They explain their situation and ask for advice. Write the prompts as full sentences, ideally with "I" or "we", and use the words your customers use, not the internal names you have for your services.

2. It has context

Add what sets your customers apart: industry, size, location, budget or a specific need. Context makes the answers more specific and closer to what a real customer actually gets.

3. It leaves out your company name

The most important insight is whether AI recommends you while the customer hasn't made up their mind yet. You only see that in questions without your company name. Use a few brand prompts to check how you are described, but keep them a small minority.

4. It belongs to a clear stage

A question about what a problem is gets different answers than a question about which provider to choose. Spread the prompts across TOFU, MOFU and BOFU, so you see where in the buyer journey you drop out.

5. It can be repeated

AI answers vary from one time to the next. The value lies in asking the same questions over time and following the trend. So avoid prompts tied to dates or events that quickly go out of date.

How to add prompts in Manager

Go to AI + Agents and Prompts. Under New Prompt you have several options:

  • Prompt: add one question manually.
  • Prompt Builder: let Manager suggest 3 to 20 prompts based on stage, focus, topic and, optionally, a persona. Useful when you want to fill a cell in the matrix quickly.
  • Generate Conversations: create conversations with several questions in a row, the way customers often use AI in practice.

The prompts suggested in the setup wizard are already here. Go through them and deactivate the ones that don't sound like real customer questions.

The fields that matter most

  • Scope: General, Contextual or Brand. This is where you separate questions without your company name from brand questions. This field decides how your numbers can be filtered later.
  • Funnel Stage: TOFU, MOFU or BOFU.
  • Focus Level: whether the question is about the problem, the solution, the product, the provider or the customer.
  • Topic and Tags: link the prompts to the business area, so you can see visibility per area.
  • Personas: link the question to an audience, if you have set up personas.
  • AI models and Search engines: which models and search engines the question runs against.
  • Run Frequency: how often the question runs. Weekly is a good starting point.

Choose a model during the trial

During the free trial, each prompt can run against one AI model and one search engine. For most businesses, ChatGPT and Google are the best choice, because that is where most customers ask. The Google measurement includes AI Overview when Google shows one. Note that Google AI Mode currently only supports English prompts.

Clean up and adjust in bulk

Select several prompts in the list to activate or deactivate them, change frequency, add models, link personas, add tags or run them right away with Run Now. Deactivate prompts you are unsure about rather than deleting them. That way you keep the history.

Next guide

How to read the visibility measurement

Frequently asked questions

How many prompts should I start with?

15 to 25 prompts per business area is a good starting point. That gives enough breadth for the numbers to mean something, without it getting hard to follow.

Which language should the prompts be in?

Write them in the language your customers use when they ask AI. If you sell in several countries, measure each market in its own language.

Why do I get different answers to the same prompt?

The AI models generate the answer from scratch every time, and the sources they use change. That is why Manager measures over time and shows trends. Read more in How to read the visibility measurement.

Can I import prompts from a spreadsheet?

Not directly today. Use Prompt Builder to create many prompts quickly, or add them one at a time.

Was this page helpful?

ShareLinkedInX

The best question is yours

Is AI citing your company?

Find out how ChatGPT, Gemini, Perplexity and Google AI Overview describe your brand. AI Marketing Manager tracks it and turns the findings into measured actions.

Explore AI Marketing Manager

Stay up to date

Get articles, product news and analyses straight to your inbox.