Issue #39: Deep Prompt Strategies for Marketing

Good morning.

This issue is dedicated to giving you a rundown, explanation, and examples of effective deep prompt strategies for marketing.

You will always get best results from having a conversation and using several of the strategies below, in one chat session (if you’re using the web version of ChatGPT, Claude, and so on).

Especially if you deploy deep prompt strategies that go beyond a few sentences or even using examples.

And, as it turns out, it’s all about The Trivium.



  • The truth about prompt strategies.

  • Prompting fundamentals you should know by now.

  • Zero-shot vs. Few-shot prompting.

  • Chain-of-Thought Prompting.

  • Multimodal Chain-of-Thought Prompting.

  • Tree-of-Thoughts Prompting.

  • Graph-of-Thoughts Prompting.

  • Skeleton-of-Thought Prompting.

  • What you should do next.

Let’s dive in.

The truth about prompt strategies

This paper by Google Brain at NeurIPS 2022 introduced the concept of Chain-of-Thought Prompting (CoT). 

The paper and its implications set off a new area of research and study with LLMs, which in turn has generated a never-ending stream of related prompt strategies, like Tree-of-Thought, Graph-of-Thought, and so on.

Let me give you the secret behind deep prompting, that very few want me to share but I’m doing so anyway: 

Prompting “discoveries” are people rediscovering and reinventing The Trivium, the liberal arts of Logic, Grammar and Rhetoric. 

They’re “discovering” that language is flexible, hackable, and powerful in and of itself.

Prompting is the application of The Trivium (Logic, Grammar, and Rhetoric) in the context of LLMs.

What the heck is The Trivium? 

I asked Perplexity for a good summary and got: 

The Trivium, originating from the Latin term meaning "the place where three roads meet" (tri + via), is a classical model of education that forms the foundational lower division of the seven liberal arts, which were central to medieval universities. 

This educational framework is comprised of three subjects: grammar, logic (also known as dialectic), and rhetoric. 

These subjects were considered essential for a classical education and were established as the core curriculum in ancient Greece, with their tradition being further developed and formalized during the Middle 

Grammar is the first of the three arts of the Trivium, focusing on the mechanics of language. It teaches students the rules and structure of language, enabling them to define and categorize the information they perceive through their senses. This foundational stage is about absorbing a vast amount of information on various subjects, including the basic vocabulary and ideas that constitute the "grammar" of each subject area.

Logic, or dialectic, is the second art, emphasizing the process of reasoning and analysis. It involves the organization of the information learned during the grammar stage into coherent arguments, teaching students to think critically and analytically. This stage encourages students to ask questions and engage in discussions to deepen their understanding of various subjects.

The third and final art of the Trivium is rhetoric, which focuses on effective communication. This stage builds upon the knowledge and analytical skills acquired in the grammar and logic stages, teaching students how to express their ideas and arguments persuasively and eloquently. Rhetoric involves mastering the art of writing and speaking in a way that influences and engages others.

The Trivium is not only a method of imparting knowledge but also a way of developing critical thinking and communication skills.

In other words, your mastery of prompting is directly proportional to your ability and skills with grammar, logic, and rhetoric. 

I haven’t heard or seen anyone else connect prompting with The Trivium, at least not publicly, so I’m probably the first person to do so—and this might be the first time you hear about it. 

Or, as I’ve said many times before and maybe now you understand better: 

Good prompting is good communication. If your prompts are bad, practice your communication skills, please. Don’t buy more silly prompt packs. 

Anyway, as you’ll see, The Trivium underlies every new prompt “discovery”, including Chain-of-Thought and strategies like: 

  • Zero-Shot Chain-of-Thought (Zero-shot-CoT)

  • Automatic-Chain-of-Thought (Auto-CoT)

  • Program-of-Thoughts Prompting (PoT)

  • Multimodal Chain-of-Thought Reasoning (Multimodal-CoT)

  • Tree-of-Thoughts (ToT)

  • Graph-of-Thoughts (GoT)

  • Algorithm-of-Thoughts (AoT)

  • Skeleton-of-Thought (SoT)

I’ll cover some of these in this issue as they relate to marketing. 

Most likely, I will teach a masterclass on prompting and The Trivium soon if enough people are interested (reply and let me know if you want the invite when it’s ready). 

It’s not for the jokers and opportunity seekers. It’s for those serious about mastering prompting in the context of their lives, business, and marketing. 

Anyway, let’s move forward. 

Some notes on the fundamentals

Before I get into all these other strategies, let’s make sure we’re on the same page with a few prompt fundamentals.

Zero-shot prompting
The term "zero-shot prompting" comes from the concept of zero-shot learning.

(Zero-shot learning is a model's ability to complete a task without having received or used any training examples).

Zero-shot prompting is when your prompt doesn’t contain any additional information for the model or provide any examples. In other words, the model can only use the knowledge it acquired during its training to produce the output.

As with everything that becomes hyped up, you’ll also find a lot of newly minted AI gurus sharing their “super”, “mega”, or “massive” prompts. These are single prompts clocking in around a few hundred or a few thousand words. 

The idea is that a single, big prompt will generate the kind of output you’re looking for. This is another version of “Zero-shot prompting”, where you go in cold, send a single prompt, and hope for the best. 

It doesn’t work that well. And it has a few serious flaws to work consistently. The first being the obvious: most models have a limited context window and if your prompt takes up too much of that window, it’s rendered useless as your output will be cut off. Who cares if your prompt is 2,000 words long, when that only leaves another 1,000 words for the output? 

Secondly, even with a much larger context window, models do not look at your total prompt. They prioritize the beginning and the end. It might turn out that only around 500 words out of your 2,000 word prompt actually affect the output. The rest is “lost” in translation to tokens. 

Thirdly, models are continually updated, filtered, and adjusted. A prompt that worked last week might not produce very good output this week. 

Fourth, the only reason you need a lot of examples or guidance in a prompt, is not because you’re a prompt genius. It’s because you’re following best practices of providing examples and guidance when prompting an LLM. There’s nothing genius about your prompt. You’re just doing the bare minimum needed.

Aside from these obvious problems, time and time again, in paper after paper—and in the real world application of prompting—having a conversation with the LLM (which involves more than one single prompt) produces better output, every time. 

In other words, you might be drooling over your 2,000 word prompt, and mistakenly believe you’ve cracked the prompting code—but you’re likely getting subpar output compared to what you could receive, if you’d have a conversation with an LLM. 

A friend of mine, who wants to remain anonymous, shared with me recently that when he rigorously tested his “mega prompts”. It turns out that he could remove at least half of the words (out of 1,751 words) and he’d get even better outputs than with the whole prompt. 

Now, let’s move on to better prompting habits.

Few-shot prompting
Similar to zero-shot, few-shot prompting also finds its roots in a similarly named learning approach: few-shot learning.

Few-shot learning is the process when a pre-trained model is given only a few examples to learn about the new, previously unseen category of data.

Few-shot prompting can be used as a technique to enable in-context learning where you provide demonstrations in the prompt to steer the model to better outputs. The examples serve as a kind of conditioning for the kind of responses you’re looking for.

(In-context learning (ICL) is a specific method of prompting where demonstrations of the task are provided to the model as part of the prompt).

Besides zero-shot and few-shot prompting, there are several other techniques and methodologies. I’ll get into a few of them that matter for marketing, but to give you an idea: 

Maieutic Prompting: This technique involves prompting the model to answer a question with an explanation, then prompting it to explain parts of the explanation further. This recursive process helps in refining the model's outputs and enhancing its ability to handle complex reasoning tasks.

Directional-Stimulus Prompting: This involves providing hints or cues within the prompt to guide the model toward a desired output. For example, including specific keywords or themes that the model should focus on in its response.

Self-Refinement Prompting: In this method, the model is prompted to solve a problem, critique its solution, and then solve the problem again considering its own critique. This iterative process can lead to improved solutions and deeper understanding by the model.

As you can see, there’s a lot more to prompting than copy-pasting “mega prompts”. It goes beyond simple zero-shot and even few-shot. 

The better you understand the principles of good prompting, the sooner you’ll be able to prompt for anything you want—without paying $27 for the latest “mega prompt” pack. 

Good prompting is a manifestation of clear thinking and good communication habits. That’s it. 

If you practice your thinking and communication skills, you win. 

Now, let’s get into some more specific prompting strategies. 

As you’ll see (and as I noted above), it’ll become obvious that people are rediscovering and reinventing The Trivium: the liberal arts of Logic, Grammar and Rhetoric. 

They’re “discovering” that language is flexible, hackable, and powerful in and of itself.

Chain-of-Thought (CoT) Prompting

For outputs demanding reasoning and sequential thinking, providing a few examples is not enough. To solve for this, researchers came up with a strategy they called Chain-of-Thought.

merely providing a few examples proves to be insufficient. To address this, researchers suggested the use of a new technique called chain-of-thought prompting.

This new technique consists of modifying the original few-shot prompting by adding examples of problems and their solutions and a detailed description of intermediate reasoning steps while describing the solution. 

Here’s an example from the original paper:

This strategy shows how complex reasoning abilities emerge naturally in sufficiently large LMs via a chain-of-thought prompting. 

A series of intermediate reasoning steps, for a given task, improves the ability of LLMs to perform complex reasoning.

The simplest application of Chain-of-Thought is to include this in your prompt:

“Let’s think step-by-step.”

As a marketer, when would you use this? 

When you need more detailed, reasoned, and contextually relevant responses. Here are a few examples:

1. Content creation: When generating blog posts, articles, or social media content, using chain-of-thought prompting can help the AI produce more coherent and well-structured content that flows logically from one point to another.

2. Brainstorming ideas: If you're looking for creative campaign ideas or strategies, chain-of-thought prompting can encourage the LLM to explore multiple angles and provide more suggestions by walking through the thought process step by step.

3. Audience targeting: When developing buyer personas or defining target audiences, chain-of-thought prompting can assist in creating more detailed and nuanced profiles by considering various attributes, psychographics, and other factors and their implications.

4. Competitive analysis: By using chain-of-thought prompts, you can guide the AI to conduct a more thorough competitive analysis, considering multiple aspects such as market positioning, unique selling propositions, and potential vulnerabilities.

5. Problem-solving: If you're facing a specific marketing challenge, such as low conversion rates or high customer churn, chain-of-thought prompting can help break down the problem and explore potential solutions in a structured way.

6. Decision-making: When evaluating different marketing strategies or campaign ideas, chain-of-thought prompting can help you weigh the pros and cons of each option and make more informed decisions based on the AI's step-by-step reasoning.

Multimodal Chain-of-Thought (M-CoT) Prompting

Not long after the Chain-of-Thought paper, the same fundamental reasoning ability was applied in a multimodal fashion (not just language input, but visuals, auditory, etc.)

This paper introduces Multimodal-CoT, focusing primarily on the interplay between vision and language.

Multimodal-CoT operates in two distinct stages:

  • Rationale Generation: In this initial stage, the model is fed with both language and visual inputs for ‘Rationale Generation’.

  • Answer Inference: In the answer inference stage, the rationale generated in the first stage is added to the original language input. This modified language input, along with the original visual input, is then fed back into the model to get the final output. In all my testing, this combination improves the quality of output.

What does this mean for an online marketer? 

You can leverage the strategy of Multimodal-CoT by combining visual and textual data. 

Here are some potential applications:

1. Product descriptions: When creating product descriptions, you can use Multimodal-CoT to generate more informative and persuasive content. By providing the model with both images and key features of the product, it can generate rationales that highlight the product's benefits and unique selling points, which can then be used to create compelling descriptions.

2. Ad creation: Multimodal-CoT can assist in generating ad copy and visuals that work together seamlessly. By inputting the desired message and target audience information, along with relevant images, the model can generate rationales that guide the creation of effective ad combinations.

3. Social media content: When planning social media posts, Multimodal-CoT can help generate engaging captions and descriptions that align with the accompanying visuals. By providing the model with the image and key messaging points, it can produce rationales that inform the creation of compelling posts.

4. Email marketing: Multimodal-CoT can be applied to optimize email marketing campaigns by ensuring that the text and visual elements work together effectively. By feeding the model with the email's main message and relevant images, it can generate rationales that guide the placement and integration of visual elements to maximize impact.

5. Landing page optimization: When designing landing pages, Multimodal-CoT can help ensure that the page's text and visual content are cohesive and persuasive. By providing the model with the page's goal, target audience, and relevant images, it can generate rationales that inform the page layout and content structure.

6. User experience enhancement: Multimodal-CoT can be used to analyze user behavior data, such as heatmaps and clickstream data, alongside user feedback and reviews. By processing these multimodal inputs, the model can generate rationales that provide insights into user preferences and pain points, guiding targeted improvements to the website or app's user experience.

By using a Multimodal-CoT strategy, you can create more cohesive, persuasive, and user-centric content that combines text and visuals to increase engagement and conversions.

Tree-of-Thoughts (ToT) Prompting

The next strategy, Tree-of-Thoughts (ToT), introduces a novel, structured approach to problem-solving

Researchers observed that human cognition often navigates through a combinatorial problem space, using a series of heuristics to guide decision-making.

ToT mimics this to a degree, and adopts a more human-like approach to problem-solving by framing each task as a search across a tree of possibilities. Each node in this tree represents a partial solution. 

The core of ToT can be distilled into answering four essential questions, as explained in the paper:

  • Thought Decomposition: ToT utilizes problem characteristics to segment the process into distinct thought steps.

  • Thought Generation: This phase leverages two strategies—either sampling independently identically distributed (i.i.d.) thoughts from a CoT prompt, which is ideal for problems with expansive thought spaces, or sequentially proposing thoughts using a "propose prompt," best suited for problems with more constrained thought spaces.

  • State Evaluation: At this juncture, the ToT framework uses heuristic methods to evaluate states. There are two strategies under consideration: one that values each state independently and another that casts a vote across multiple states.

  • Search Algorithm: Depending on the structure of the problem tree, different search algorithms like Breadth-First Search (BFS) and Depth-First Search (DFS) can be deployed.

How can you use this in marketing? 

1. Campaign planning: Use ToT to break down your campaign planning process into distinct steps, such as defining objectives, identifying target audiences, selecting channels, creating content, and measuring results. At each step, generate and evaluate multiple ideas or strategies before proceeding to the next step.

2. Content creation: Apply ToT to content creation by decomposing the process into stages like ideation, outlining, drafting, editing, and optimization. Generate multiple ideas or outlines, evaluate them based on relevance, originality, and potential impact, and then proceed with the most promising ones.

3. A/B testing: Employ ToT to design and analyze A/B tests for landing pages, ad copy, or email subject lines. Break down the process into steps like identifying test elements, generating variations, defining success metrics, and analyzing results. Use ToT to systematically explore and evaluate different test hypotheses.

4. Customer journey optimization: Use ToT to analyze and optimize the customer journey by breaking it down into stages like awareness, consideration, purchase, and retention. At each stage, generate and evaluate multiple strategies to improve the customer experience and drive conversions.

5. Influencer marketing: Apply ToT to influencer marketing by segmenting the process into steps like defining campaign goals, identifying relevant influencers, evaluating their fit, negotiating partnerships, and measuring results. Use ToT to systematically explore and evaluate different influencer strategies.

6. Marketing automation: Use ToT to design and optimize marketing automation workflows by decomposing the process into triggers, conditions, actions, and outcomes. Generate and evaluate multiple automation scenarios to find the most effective ones.

7. Customer segmentation: Employ ToT to segment your customer base by breaking down the process into steps like defining segmentation criteria, generating customer profiles, evaluating segment viability, and creating targeted campaigns. Use ToT to systematically explore and evaluate different segmentation approaches.

8. Marketing budget allocation: Apply ToT to optimize your marketing budget allocation by segmenting the process into steps like defining objectives, identifying channels, evaluating channel effectiveness, and allocating funds. Use ToT to systematically explore and evaluate different budget allocation scenarios.

Of course, there are more ways but this should give you some good ideas to start with. 

The ToT strategy can help you break down complex problems into manageable steps, explore multiple solutions, and ultimately arrive at the most effective strategies for your specific marketing challenges.

Graph-of-Thoughts (GoT) Prompting

Anything that can be connected entities can be a tree, and can be a graph. 

Another prompting strategy is the so-called Graph of Thoughts (GoT) framework, designed to help and improve the reasoning capabilities of LLMs using a graph-based structure. 

This strategy provides a structured, extensible mechanism for thought transformations, evaluations, and rankings.

The GoT framework is built as a set of interacting modules:

  • Prompter

  • Parser

  • Scoring module

  • Controller.

Each module performs “a specialized task in the reasoning process, ranging from preparing the prompt to validating and scoring the generated thoughts”. 

The Controller “coordinates these modules and also houses two key elements: the Graph of Operations (GoO) and the Graph Reasoning State (GRS), which further assist in managing the LLM reasoning process.”

Sounds complicated? It’s not, it’s a reflection of Logic, along with Grammar and Rhetoric. 

For an online marketer like yourself, you can apply this strategy to various aspects of your work to improve your reasoning and decision-making processes:

1. Customer journey mapping: Use GoT to create a comprehensive customer journey map by breaking down the process into stages, identifying key touchpoints, and analyzing customer behavior at each stage. The framework can help you evaluate and score different strategies to optimize the customer experience.

2. Content strategy development: Apply GoT to develop a content strategy by structuring the process into steps like defining objectives, identifying target audiences, researching topics, creating content pillars, and measuring performance. The framework can help you generate and evaluate different content ideas and strategies.

3. Marketing funnel optimization: Use GoT to optimize your marketing funnel by breaking it down into stages like awareness, interest, desire, and action. The framework can help you identify bottlenecks, generate ideas for improvement, and evaluate the potential impact of different optimization strategies.

4. SEO strategy: Apply GoT to create and refine your SEO strategy by structuring the process into steps like keyword research, on-page optimization, link building, and performance tracking. The framework can help you generate and evaluate different SEO tactics and prioritize them based on their potential impact.

5. Cross-channel marketing optimization: Use GoT to optimize your cross-channel marketing efforts by breaking down the process into steps like identifying channels, defining channel roles, creating integrated campaigns, and measuring cross-channel performance. The framework can help you generate and evaluate different cross-channel strategies and allocate resources effectively.

6. Marketing technology stack optimization: Apply GoT to optimize your marketing technology stack by structuring the process into steps like identifying business needs, evaluating tools, integrating systems, and measuring ROI. The framework can help you generate and evaluate different technology combinations and prioritize investments.

7. Customer retention strategy: Use GoT to develop a customer retention strategy by breaking down the process into steps like analyzing churn reasons, identifying retention triggers, creating retention campaigns, and measuring retention rates. The framework can help you generate and evaluate different retention strategies and prioritize them based on their potential impact.

8. Marketing funnel conversions: By breaking down the funnel into stages, analyzing target audience segments, generating and evaluating persuasive messages, selecting appropriate communication channels, and refining based on performance data, you can systematically improve your funnel conversions.

Makes sense, right? 

Skeleton-of-Thought (SoT) Prompting

Another strategy in the Chain-of-Thought family, is Skeleton-of-Thought, which plays around with how LLMs generate text. 

Instead of building responses sequentially, SoT uses parallelism to enhance speed and accuracy. 

SoT creates a concise "skeleton" of the answer, then fills in details in parallel, mimicking the organized way humans think.

  • Skeleton Stage: Utilizing a template, the LLM crafts a skeletal response to the user's question, from which key points are extracted.

  • Point-Expanding Stage: These points are then expanded upon in parallel, leading to a more detailed final answer.

How do you apply this in online marketing?

Here are some use cases:


- Creating ad copy for search engine marketing (SEM) campaigns. Use SoT to generate a skeleton of the ad copy, including the headline, description, and call-to-action. Then, expand on each point in parallel to create multiple variations of the ad copy.

- Developing storyboards for video ads. Use SoT to create a skeleton of the video script, outlining the key scenes and messages. Then, expand on each scene in parallel, adding details like visuals, dialogue, and transitions.


- Writing product descriptions for e-commerce websites. Use SoT to generate a skeleton of the product description, including the key features, benefits, and specifications. Then, expand on each point in parallel to create a detailed and persuasive description.

- Crafting email sequences for lead nurturing. Use SoT to create a skeleton of the email sequence, outlining the key messages and calls-to-action for each email. Then, expand on each email in parallel, adding personalized content, examples, and value propositions.

Marketing Analysis:

- Creating a SWOT analysis for a marketing strategy. Use SoT to generate a skeleton of the SWOT analysis, identifying the key strengths, weaknesses, opportunities, and threats. Then, expand on each point in parallel, providing detailed insights and examples.

- Developing buyer personas for targeted marketing. Use SoT to create a skeleton of the buyer persona, outlining the key demographic, psychographic, and behavioral characteristics. Then, expand on each characteristic in parallel, adding specific details, preferences, and pain points.

Other use cases:

- Content creation: Use SoT to create blog post outlines, social media post structures, or video scripts, allowing for efficient and organized content generation.

- A/B testing: Use SoT to create a skeleton of different test variations for landing pages, ad copy, or email subject lines, then expand on each variation in parallel to create detailed test scenarios.

- Marketing strategy development: Use SoT to create a skeleton of a marketing strategy, outlining the key objectives, target audiences, channels, and tactics. Then, expand on each point in parallel to create a comprehensive and actionable strategy.

It’s a great strategy for streamlining your content creation process. 

The parallel expansion of key points allows for faster generation of ideas and content, while the skeletal structure ensures that the final output is coherent and organized.

That’s all I’ve got for this issue. 

There are new ideas and methods (such as chain-of-verification (CoVe), chain-of-density (CoD), etc.) but I’ll cover those in a different issue.

I’ve got 2 things coming up: 

Premium subscriptions: I’ll soon test out a $99/month subscription, with 2-3 issues per month that cover things like actual prompts, agents/GPTs, in-depth coverage of important papers, techniques for AI advertising video creation, and more. 

Prompt Masterclass: I touched on this at the beginning, and as you can see from this issue alone, there are many techniques and strategies for prompting that can make life easier for you, your business, and your marketing. 

I’m thinking about a paid masterclass (or perhaps a multi-day workshop) that installs a mental model operating system in your mind, so you can come up with prompts (and in turn, GPTs and agents) at will, on command, and on demand. For any area of life and business.

This would go deep and focus on the models and frameworks that matter. It won’t be something silly like “get 1,000 prompts!”

And—this would be pretty expensive and a small group. 

If either of these, or both, sound interesting to you—reply and let me know.

I’m only doing either (or both) if there’s enough interest.

Talk again soon,
Sam Woods
The Editor