Prompts for Product: Examples for Product Managers and Marketers

Using AI tools like ChatGPT is deceptively simple. Chatbot builders go out of their way to make them as user-friendly as possible, however there are hidden pitfalls that can trip you up if you don’t do your research. This can lead you to getting results you weren’t expecting, without knowing why.

On the plus side, where there is challenge there is opportunity. If you can learn the secrets behind great prompt writing, and gather a bank of tried-and-tested examples, you can really get the most out of AI tools. Especially if you’re a Product Manager or Product Marketer. Product people like you are already harnessing the power of AI to create project documentation, manage meetings, and generate actionable insights from data.

Whether you’re curious about how to harness the power of AI, or you’re wondering why you’re not seeing the results you want from your prompts, we’re here to help! We’ll be going through the technology behind LLMs, some do’s and don’ts of prompting, and give you a wide range of example prompts to get you started.

How Do Chatbots and LLMs Work?

Large language models (LLMs) are the machine learning models behind most modern chatbots, operating through a neural network architecture allowing them to understand and respond to human-like text. The training process involves feeding them vast datasets scraped from the internet, books, articles, and many other sources. The model is then taught to predict the next word in each sentence based on the preceding words, learning to recognize patterns through repetition. The more the model is used, the better it gets, through another process called back propagation - the model learns from its mistakes.

The engineers behind modern LLMs ensure quality responses through a few different techniques. An attention mechanism allows the model to assign varying ‘weights’ of importance to different words in a sentence, to help it understand context. Text is broken down into tokens for easier processing, and then converted into dense vectors through embedding layers, helping the model to further understand the relationships between words.

While LLM-driven chatbots like ChatGPT and Perplexity appear to be all-knowing, they have their flaws. Depending on the tool you’re using, it may not be able to search the internet for the most up to date information, meaning it only has knowledge of the dataset it was trained with. If the output you want is reliant on current statistics, you’ll want to make sure the chatbot you’re using has web scraping capabilities.

How Does Prompting Work?

A prompt is the input provided to an AI model, and while chatbots are great at performing complex tasks, their conversational UI means that typically prompts are instructions to perform one task or answer one question. ‘Give me some data points on how many businesses are adopting AI automation tools in 2024’ or ‘Check this email for grammar mistakes.’

Chatbots are designed to maintain a contextual awareness to allow for more of a dialogue. This allows the user to execute multi-step tasks within a session. It’s not dissimilar to working with a junior colleague during their first week. Similarly, your junior colleagues need context and specifics, as they’re limited in their abilities to read between the lines and perform tasks without key information.

For example, if you’re asking a chatbot to write a blog post for you, it’ll be helpful to supply it with your brand voice. Or if you want to create user personas, the output will be more accurate if you provide the tool with up to date user information.

Best Practices for Prompts

1. Be clear. Chatbots only mimic understanding, but they lack human intuition. There are only so many blanks they can fill in. Think back to how you felt on the first day of your job, without fully understanding the company and what you were expected to do. That’s how your AI tool feels!
   ‍
2. Be specific - AI is very smart, so don’t be afraid to get creative. You can guide brand voice, make special requests, and many more. Use the fact that chatbots have ‘memory’ to your advantage, and go through a task step by step, adding layers of complexity as you go.
   ‍
3. Provide context and examples - If you want something done a particular way, give the tool something to refer to. Let’s say you’re announcing a feature launch to your users, and want to write an email blast just like the one you used in the previous launch. Feed the tool your last announcement email as an example.
   ‍
4. Work directly from source material - Or better yet, get a tool like Bash to help you craft your announcement email based off of your product documentation and user information. AI is always more powerful when it’s working straight from your own sources.
   ‍
5. Learn from the best! - There’s a whole community of people getting excited about the potential of AI, and they’re making and sharing huge repositories of prompts, like this one. Do some digging and see what other product people are up to.

Let’s see what this looks like in practice, by asking ChatGPT 4o to give us a description of Bash.

What went wrong? For starters, Bash means many things. I didn’t specify that I wanted it to tell me about Bash. I also didn’t specify what kind of description I wanted, or what I deem to be ‘short.’

So let’s try again, this time telling ChatGPT exactly what I want.

That’s better, but it’s still missing something. What ChatGPT doesn’t know is how Product Managers and Product Marketers can benefit from using its features for project documentation. Luckily, ChatGPT remembers our conversation and can give me some edits.

Common Pitfalls for Prompt Writing

1. Overloading with information: Some users feel the need to get an incredibly complex task done with a single prompt. Asking a chatbot to take too many steps at once can overwhelm the model and give you confusing results. Make the most of the chatbot’s ‘memory’ and break tasks down step by step.
2. Not checking data sources: Most bots do not cite sources unless specifically directed to, and they have little information on whether a source is considered reliable or not. Believe it or not, AI is also capable of making mistakes…

Image source

3. Hallucinations: Be aware that when AI models lack information for a task, they may ‘hallucinate’, or come up with answers based on non-existent data. These can be tricky to spot, but there are things you can look out for. When hallucinating, AI’s language can become clunky or overly complex, no longer mimicking human speech patterns. You can also keep an eye out for wildly high or low data points. “99% of Product Managers say that they love how much time they spend in meetings.” Yeah, you may want to fact-check that!
4. Leading questions: Be careful of phrasing your questions in ways that lead to biased answers. For example, ‘Why is it better to have employees work from the office, rather than from at home?’ will give you a list of reasons why in-office work wins over WFH. Because that’s exactly what you asked for. But the answer will be pro-office, because your question was pro-office. To get a more balanced answer, your prompt could be; ‘Give me bullet-point lists of pros and cons for working in the office, and working from home.’

AI Prompts for Product Management

Tools like ChatGPT are extremely helpful when it comes to external research and data analysis. But if you’re building project documentation it’s better to do that with an AI tool that’s directly linked to your source material like Bash.

You’ll notice that many of these prompts are asking the AI tool to handle information that’s probably not accessible to the public, like your meeting notes and user research. This is why smart product people are adding tools like Bash to their stack. It more easily allows you to connect your documents and internal information with the power of AI without the need to remember long and complicated prompts.

Optimizing Meetings
‍

1. "Summarize the key discussion points and action items from this meeting transcript."
2. "Based on this meeting recording, identify the main challenges discussed and propose potential solutions."
3. "Extract the top 3 priorities mentioned in this meeting and create a brief action plan for each."
4. "Analyze this meeting transcript and suggest ways to improve the efficiency of future meetings on this topic."

When you take your meetings with Bash, you’ll automatically get a recording and a transcript to work with. This is already incredibly useful, and makes sure you never miss an insight or action point again. Instead of prompts, Bash has templates built in that will return more detailed meeting minutes or other types of meeting follow-ups.

Try this out by uploading or recording something yourself:

‍

Creating Product Requirement Documents (PRDs)

1. "Outline a PRD structure for [product name], including sections for user stories, feature specifications, and success metrics."
2. "Generate a list of potential user stories for [new feature], considering different user personas and use cases."
3. "Based on these market research findings, suggest key features to include in the PRD for [product name]."
4. "Create a template for documenting technical requirements and dependencies for [product feature]."

Instead of dropping these into ChatGPT, check out our PRD template, and write a PRD, user stories, or other product documents in no time.

Creating a Product Roadmap

1. "Develop a high-level product roadmap for the next 12 months, considering our current resources and market priorities."
2. "Suggest a framework for prioritizing features on our product roadmap based on user impact and development effort."
3. "Generate ideas for potential product milestones to include in our Q3 and Q4 roadmap."
4. "Create a visual representation of our product roadmap, highlighting key releases and their estimated timelines."

Managing User Feedback and Running Retrospectives

1. "Analyze this set of user feedback and categorize the comments into themes of praise, concerns, and feature requests."
2. "Based on these sprint retrospective notes, identify recurring issues and suggest actionable improvements for the next sprint."
3. "Summarize the key learnings from our latest user testing session and propose next steps for product iteration."
4. "Create a template for a user feedback survey that will help us gather actionable insights for our next product update."

Getting Insights from Analytics

1. "Analyze these user engagement metrics and identify potential areas for improvement in our product's onboarding flow."
2. "Based on these conversion funnel data, suggest hypotheses for why users might be dropping off at [specific stage]."
3. "Interpret these usage statistics for [feature] and recommend ways to increase adoption among our user base."
4. "Compare our key performance indicators (KPIs) from the last quarter to industry benchmarks and highlight areas where we're excelling or lagging."

AI Prompts for Product Marketing

It’s not just Product Managers who get to have all of the AI fun. AI tools and chatbots are also extremely helpful when it comes to marketing efforts.

But just like Product Managers, remember that a lot of the tasks you have will be based on internal documentation not readily available inside of LLMs. Make sure that you feed relevant documents like your brand voice and persona to any AI tool you use, or choose a tool like Bash which can be instantly connected to your documents.

Iterating on Messaging and Positioning

1. "Analyze our product features against competitor offerings and suggest unique selling points for our marketing messaging."
2. "Based on these customer personas, propose positioning statements that highlight how our product solves their specific pain points."
3. "Generate ideas for potential market segments we could target with our product, considering its current features and capabilities."
4. "Create a value proposition canvas for our product, focusing on customer jobs, pains, and gains."

Developing a 360 Product Marketing Strategy

1. "Outline a comprehensive product marketing strategy for [product name], including goals, target audience, channels, and key messages."
2. "Suggest a launch timeline for our new product, including pre-launch, launch day, and post-launch marketing activities."
3. "Develop a content calendar for promoting our product across various channels (social media, blog, email) for the next quarter."
4. "Create a framework for measuring the success of our product marketing efforts, including relevant KPIs and tracking methods."

Creating Content

1. "Generate ideas for blog post topics that would help educate our target audience about how [product] solves [problem]."
2. "Create an outline for a case study showcasing how [customer] successfully used our product to achieve [specific outcome]."
3. "Develop a script for a 60-second explainer video introducing our product and its main value proposition."
4. “Create an email blast announcing the launch of [feature name] for [product name], highlighting the key benefits for existing users.”

AI Misbehaving? Check Your Prompts!

If you’re not getting the results you expect, run through this quick checklist to make sure you’ve covered all of your bases:

1. Is your prompt too short? This might mean you’re being too vague and not providing enough detail. Make sure you’re explaining yourself properly.
2. Are you being specific? AI can’t read your mind. Tell it exactly what you want the outcome to look like.
3. Is your prompt too long? If you’re trying to do something complex, break it down step by step.
4. Are you using source material? Make sure you’re not asking AI to complete a task based on data it has no access to.
5. Are you checking for cited sources? Data looking funky? Ask the AI to cite all of its sources and check they’re reliable.

Too complicated? Use built-in Bash templates or add your own custom templates

Instead of having to remember or copy paste all these prompts, you can also use the built-in templates in Bash to have AI and a LLM to write what you need. Bash templates are basically pre-written prompts based on best-practice templates so you don't need to do any of the thinking or experimentation to get your output just right.

If you do want to write your own prompts, you can also create your own custom templates in Bash which allows you to store these prompts for later use. You can even upload a document you use within your company and turn this into a template.

Each of these custom templates can then be applied to any information you have in Bash making it easy to get consistent and accurate project documentation.