6 Advanced Prompt Writing Tactics for Improved AI Outputs
Large language models are exciting tools, but they can lose steam with some marketers who don’t immediately experience success with them. If you want to use AI beyond basic tasks, you must move beyond basic prompts. Become a prompt engineer with a few easy-to-learn advanced prompt writing tactics.
By Jared Frank | 6 min read
When OpenAI launched ChatGPT in November 2022, the world was introduced to large language models (LLMs). Early adopters flocked to the new generative AI with curiosity. And the early returns were extremely positive. Minds were blown. Over 100 million people used the chatbot in the first two months, making it the fastest-growing consumer software application in history. Hyperscalers began spending like drunken sailors to calm their waves of FOMO.
Then for some, the bloom wore off the rose. The technology was “cool”, but it didn’t really solve their big issues or save them time. For this cohort of AI users, the flaw is not in the technology, but in “user error.” When LLMs hit a wall, it’s almost always because the prompts being fed to them are not good enough to progress the AI conversation towards the expected solution.
In order to put AI to use on complex tasks, one must master advanced prompt writing. LLMs are only as helpful as we help them. They require detailed context (that they won’t possess unless we give it to them) around our specific use cases.
Advanced prompting doesn’t mean writing longer prompts – those won’t help you. Advanced prompt writing means a user has identified all subtasks within a complex task and has scribed multiple, shorter prompts to solve those complex tasks one step at a time.
The six building blocks studied previously remain the foundation for effective prompts. Now, let’s add six advanced prompt writing tactics on top of that foundation to unleash the full potential of generative AI.
1. Anonymize
The first thing marketers should do before diving headfirst into the AI deep end is ensure the AI is restricted from training on your company’s private or proprietary information. It’s easy enough to go into your ChatGPT settings and disable the training function (if you’re a paying customer). For some small and medium-sized businesses, you might not need to lose sleep over this data share, but just be advised that whatever data you input into ChatGPT OpenAI will use to further train the model. Setup your guardrails as needed.
Beyond toggling your settings, here are some additional practical examples for anonymizing your prompts.
Don’t Say: “I work at Seat 36.”
Do Say: “I work at a content marketing agency.”
Don’t Say: “Use the attached spreadsheet to learn more context on our customer.”
Do Say: “Our customer is a landscaping business in Frisco, Texas that leverages content marketing to enhance SEO and generate more sales leads.”
By simply referring to your company and your customer with general information and excluding any datasets containing sensitive information, you can protect privacy while still achieving the results you need.
2. Template
AI excels at following specific structures that you instruct it to follow – tables, bulleted/numbered lists, etc. Use this feature to your advantage to save the time and effort of having to format outputs later – all you’ll have to do is polish. Templating is especially helpful for high-frequency tasks, including project proposals or contract reviews.
3. Use Delimiters
LLMs have a “lost in the middle” problem. They pay close attention to the beginning and end of your prompt, but sometimes wander in the middle and ignore that information. To help the AI stay focused and remember what’s in the middle, break up your prompt with section titles and/or delimiters, such as hashtags or slashes, to distinguish its components. Consider the following example:
ROLE: You are my expert copywriter at my content marketing agency.
###
CONTEXT: You are writing a series of blog articles to help our clients understand the basics of AI.
###
TASK: Your job is to generate ideas for this blog series and write high-performing headlines for each article.
###
SPECIFIC REQUIREMENTS: Write ten blog topics with ten unique headlines for those topics.
###
CONTEXT: Visit https://seat36.com/client-stories/ to get background information on our ideal customers.
###
AUDIENCE: The audience for this blog series are marketing professionals at small and medium-sized business. Use your knowledge of the pain points of this audience to enhance the performance of the blog series.
4. Try Interactive Prompting
If you are unsure whether the AI has everything it needs to deliver a quality output, try asking it. By instructing the AI to ask you questions before responding, you ensure the AI understands what you need before proceeding to the next steps. An example interactive prompt might be: “Before responding, ask me relevant questions to give you additional information you need.”
Interactive prompting is another way to avoid inadequate context, as well as the “lost in the middle” problem. This technique also assists with breaking up complex tasks into multiple prompts, as well as avoiding hallucinations or generic responses due to missing information.
Interactive prompting captures the conversational spirit of large language models. Let the AI guide you through the process, rather than the other way around.
5. Provide Steps
If the AI continues to have difficulty maneuvering through long, complicated prompts, spell out the task step-by-step for the machine. For example:
Compose a project proposal for launching a new SEO marketing campaign by following these instructions:
Step 1 – Introduce your project idea with a sentence starting with ‘Project Idea:’.
Step 2 – List the objectives of the project with the prefix ‘Objectives:’.
Step 3 – Outline the required resources with the prefix ‘Resources Needed:’.
Step 4 – Estimate the timeline for the project with the prefix ‘Estimated Timeline:’.
6. Provide Examples
LLMs will always perform better after they see an example of the task you’re wanting to accomplish. By prompting with a text example or file upload, you can communicate specific requirements without having to explain them. This tactic is ideal for repetitive tasks, including content creation. For example:
Attached is a blog article I wrote about the benefits of an AI content strategy for small businesses.
Write a new article mimicking my tone and style on the following topic:
The best tools for creating AI-generated content.
Some Final Tips on Advanced Prompt Writing
Prompting is an iterative process, and sometimes you have to try multiple tactics before getting the output you need. So prepare to have a conversation with AI. And if the AI gets lost and your conversation goes off the rails, start over and try again. After you find success with a prompt, store it in a document and create a prompt library for tasks you do regularly. It will get easier with time.
Remember, advanced prompt writing takes practice. If you’re experiencing too many simplistic, cheesy, cringy, painfully cliché responses from AI, don’t give up. As your prompts improve, so will the AI performance. 
Are you struggling to generate high-quality AI outputs? Let’s chat.
Write to Jared at [email protected].
RELATED CONTENT
6 Basic Building Blocks for Marketers to Write Effective AI Prompts
AI models have quickly become powerful assistants for marketing professionals. But to get the best results from these tools, marketers must communicate with them clearly. It’s both an art and science to write effective AI prompts.
Why Your Small Business Should Embrace an AI Content Strategy
If you want new customers to find you and existing customers to love you, you need good content. While marketing tasks can overwhelm a small business’ resources, an AI content strategy enables a simple, affordable process.