I’ve been tinkering with AI and automation since 2019, first as a developer, and then as a marketer. At the start, AI felt similar to what I’d already known as a developer. Data in, with output based on loosely defined instructions and rules.
However, in the last couple of years, we’ve seen a whole different story. AI-generated lifelike avatars, personalized launch sequences, AI-led animated games, and more. For example, OpenAI released a complete launch video, made fully with Sora, generating over 15M views in under 48 hours. It was an excellent way to showcase their text-to-video engine product, and also reminds us how far we’ve come with AI.
But how does AI work? Why have AI models gotten so much better over time? How can marketers use AI to their advantage?
In this post, I answer all of these questions based on my six years of hands-on AI experience. I’ll also help you build out a usable AI workflow for just about any marketing use case.
Table of Contents
- What is AI?
- The Benefits of AI
- How does AI work?
- The Four Types of AI
- How I’d Build an AI Workflow
- AI Use Cases for Marketers
The State of Artificial Intelligence in 2025
New research into how marketers are using AI and key insights into the future of marketing.
- Marketing AI Tools
- Practical Tips
- Trends and Statistics
- And More!
Download Free
All fields are required.
What is AI?
Artificial intelligence is an advanced technology, typically run by a series of algorithms, computers, or robots, that uses real-time data to simulate human intelligence.
AI can replicate human discernment and make real-time decisions. In other words, artificial intelligence is programmed to think, act, and respond just like a real, live human.
AI is not to be confused with automation. Although both automation and AI use real-time data to perform a function, the mechanics and output are vastly different.
For example, automation requires manual data input to perform a certain task. Using an algorithm, that task will repeat, regardless of what the data says or if there’s an error.
AI, on the other hand, is machine learning. Meaning it requires an input of data. As it processes the data, AI can recognize behavior patterns and errors, then adjust its functions and algorithms as needed.
AI is growing in popularity and can be used across a variety of industries. Let’s take a look at the benefits of using it.
The Benefits of AI
AI models have improved dramatically over the last couple of years. Sometimes I see those Instagram reels that ask you to spot what‘s AI and what’s real, and I genuinely can't tell anymore. However, this also means you can rely a lot more on AI for your marketing workflow and see tangible benefits. Let’s look at some of the key ones.
1. It performs repetitive tasks.
Marketing involves a lot of moving pieces, like coordinating with stakeholders, cleaning spreadsheets, drawing up campaign reports, etc. At scale, all of these can add up and get quite repetitive. Before AI, I’d either batch these tasks and power through them or delegate them to someone else on the team.
However, one of the main benefits of AI is that it can handle this grunt work for you. In HubSpot’s State of AI report, 95% of respondents said those tools help them spend less time on manual tasks and more time connecting with customers.
By implementing AI to handle these high-volume, repetitive processes, you can help your team work smarter, not harder.
2. It helps with research and data analysis.
AI can handle massive amounts of information with ease. Given that marketing is now becoming heavily data-driven, that’s a huge plus. Of marketers using AI, 47% say they use it for research, summarization, or data gathering.
Here are some ways I’ve seen marketers use AI for data analysis:
- Sharpen brand positioning and messaging by combing through industry reports.
- Understanding which channels and activities are driving customer revenue.
- AI-powered predictive analysis for sales forecasting.
- Risk mitigation through data monitoring.
I personally love using ChatGPT’s “Deep research mode” to find the right stats for my research articles.
3. It can make unbiased, smart decisions.
Marketing decisions often get made based on gut feeling, opinions, or what worked once before. I’ve sat in meetings where decisions were finalized because “that headline felt stronger” or “this format worked for us last year.” The results were usually hit or miss.
That’s where AI has been surprisingly helpful. AI brings the focus back to the data and combs through historical patterns to uncover real insights. Not just what sounds good or what the highest-paid person in the room prefers.
Of course, it depends on the quality of the input. If the data is biased or incomplete, the results can reflect that. Your data needs to be clean for AI to be useful.
While AI doesn’t replace intuition, it challenges common biases and assumptions. All of this makes it a really useful checkpoint in the creative process.
4. It reduces costs and multiplies revenue.
All of the benefits we discussed above lead to one thing: bottom-line revenue growth. AI can help you run faster and leaner teams, while keeping employees motivated and happy.
Companies like IBM have already unlocked $3.5 billion in cost savings and a 50% increase in productivity through AI-led transformation initiatives.
How does AI work?
At its core, AI works on a feedback loop: learn → act → adjust → repeat. Essentially, AI looks at patterns, learns from new inputs, and adjusts based on feedback. Here’s a step-by-step breakdown.
Input
Everything starts with data. In this step, you train the model with huge datasets. For companies, that means collecting data across their business functions or using a CRM like HubSpot to input data to their AI model. You can also use data marketplaces like Kaggle to source ethically collected data and supplement your overall dataset. Training data typically is a collection of text, images, and speech.
After training your model, the second input you give AI models is prompts. Prompts are essentially written text or voice notes that contain clear instructions for the AI model to process.
Processing
Once the input is in, the AI kicks in. Your prompt is broken down into tokens, which can be fed to the AI model. These tokens can be words, phrases, or even parts of a word.
Using these tokens, the model determines the relationship between the words in the sentence based on their position. For example, “You can only see this” and “You can see this only” have two different meanings.
Once the model does the positional analysis, it starts identifying the key entities in the prompt, based on the training data we provided earlier. All of this helps it understand the core intent of the ask and answer accurately.
Prediction
After understanding the intent, the model starts to form output options. Essentially, it predicts the next most likely word based on previous patterns to frame different responses. For example, if you ask ChatGPT to write a product headline, it pulls from everything it’s learned to predict what a successful one looks like.
After it has a couple of options to choose from, it runs the response candidates through different quality controls and safety controls. Based on the scores, the best one is presented as the final output, and the others are discarded.
Feedback
There are two types of feedback loops for AI: model-level training and prompt refining. Model-level training is when the model re-trains itself based on newly available data sources. This is called “fine-tuning” where engineers deliberately retrain the model on specific datasets to make it better at targeted tasks.
External feedback is when you actively train the model with prompts. Once you receive the output, you can go back and tweak the prompt or add new examples. This back-and-forth is the feedback loop we discussed earlier, and it’s the most effective way to make AI more reliable over time. The more you refine it, the better it supports your style and goals.
The State of Artificial Intelligence in 2025
New research into how marketers are using AI and key insights into the future of marketing.
- Marketing AI Tools
- Practical Tips
- Trends and Statistics
- And More!
Download Free
All fields are required.
The Four Types of AI
Not all AI is the same. Understanding the differences between various AI types can help set the right expectations for which one is better suited for your business.
Here, I break down the different types of AI, from the most basic to the purely hypothetical ones. Most of what we use today falls under category two.
1. Reactive Machine
These are the simplest types of AI. They respond to whatever input you give them, but that’s it. No memory, no learning, no context.
Think of a rule-based chatbot that gives a fixed response when someone clicks a certain button. Another one is a spam filter that marks an email based on a known keyword.
This type of AI was everywhere a few years ago. It handled basic, repetitive tasks just fine, but never got any smarter. You can still sometimes bump into it inside older automation flows. For instance, a template-driven bot that collects newsletter sign-ups and always sends the exact same “Thanks for subscribing” email, no matter who signed up or what page they came from.
2. Limited Memory
Most AI tools we use today sit in the limited-memory bucket. They hold on to recent context, then use that short-term history to give smarter answers.
ChatGPT is a good example. It can keep track of a few thousand words in a single chat, so it remembers your brand voice or prompt structure while you stay in the same thread.
Recommendation engines in Netflix or Spotify work the same way. They look at what you just watched, skipped, or saved and tweak suggestions on the fly.
3. Theory of Mind
Theory of Mind refers to AI that can actually understand human emotions, intent, and nuance — such as recognizing when a customer is frustrated versus merely asking a question, and adjusting the tone or response accordingly.
If this becomes real, it could change how we build chatbots, support agents, and even content tools. But for now, most AI still can’t truly “understand” how we feel. It’s just guessing based on patterns in the data. However, it can sometimes get eerily close.
For instance, scientists are now finding that AI can sometimes be more persuasive than humans in debates. We’re also seeing AI in mental health circles to administer therapy under the supervision of a therapist.
4. Self-Aware
I’ve spent a lot of time thinking about where “self-aware AI” lives.
Right now, we don’t have systems that genuinely form opinions or make decisions entirely on their own. Instead, we’re seeing tools that act more like assistants with autonomy.
Take ChatGPT agents or AI agents in general. They’re designed to perceive, take action, and adapt within boundaries. Advanced systems like multi-agent workflows simulate a group of specialists. Agents can talk to each other (for example, a content agent handing off to a data agent) to coordinate complex tasks.
However, the agentic future is far beyond that. AGI, or artificial general intelligence, is when the “thinking” can be handed off to the AI tool. They function as effectively as a senior-level assistant, executing without oversight, prompting, and adjusting their strategies.
Ben Goertzel, the father of the term AGI and co-founder of Singularity.NET, believes we could reach AGI within three to eight years, with an ambitious prediction of 2027-2030 as the likeliest window.
How I’d Build an AI Workflow
When people say “build your own AI,” it sounds intense, like you need to write thousands of lines of code. Trust me, it’s way less daunting than it sounds.
You can easily use AI to perform repetitive functions that drain your employees of their valuable time — time that could be spent strengthening client relationships or making a sale.
For me, building a simple AI system means setting up a structured workflow around tools like ChatGPT that can actually help with writing, research, or repurposing content, without needing a developer.
Here’s the framework I follow when I’m building an AI-powered assistant.
1. Start with a clear problem.
You can’t automate what you can’t articulate. Automating a process needs you to clearly define the exact problem, lay out the steps a human would take to achieve the task, and then identify areas of improvement.
For instance, does your team spend significant time sorting through data to find contact information for potential clients? Could they use their time better by speaking to potential clients and onboarding new customers?
Then, AI can help.
For my workflow, I typically identify something that’s slowing me down. This could look like:
- Too much time rewriting briefs.
- Repeating the same social post formats every week.
- Creating outlines from scratch for every blog.
Once I know what task I want to speed up or simplify, I know where AI fits in. Figure this out for your company workflows, make a list of potential areas AI could help, and then choose one of them to start with.
2. Define the outcomes.
The next step is to get super clear on what a “good result” looks like. Just like training a new hire, AI needs you to give it a proper brief and examples of what the end output looks like. When you’re training AI, take on a managerial role and guide it towards the ideal output.
Apart from defining a good result, you also want to tell AI what not to do. Build out limitations and guardrails to ensure the tool doesn’t hallucinate. For instance, if I am writing a thought leadership piece, I’ll include things like “don’t make unsubstantiated claims,” or “avoid generic or common advice and go deeper in the research stage.”
While your use case might look different from mine, the guiding principles are the same.
3. Pick the right AI tool.
Like I described earlier, there are different types of AI suited for specific tasks. You want to choose the one that is right for your budget and goals. I spoke to Forrest Webber from Trade Table, who suggests A/B testing tools before finalizing a specific one:
“We use both ChatGPT and Canva’s AI to create marketing images. It really depends on which tool gives us the better result for a specific product. I’ll usually start by writing a prompt or concept, then test it on both platforms to see which one nails the look we’re going for.”
My approach is quite similar to his. For example, I turn to Claude for creative writing tasks because it can emulate my writing style better. However, I’ve found that ChatGPT is clearly superior for data analysis and research-based tasks. If I want to try out niche tasks like graphic design or logo generation, using a tool trained with specialized data often gives me better results than a general chatbot.
4. Train the tool.
Now, it’s time to feed relevant and accurate data. Each AI model is already trained on a vast amount of data that makes it good at a particular type of task. In this step, you need to customize it to meet your specific needs.
For instance, if I want help with writing a blog, I always try to include:
- Examples of past content that worked and flopped.
- Target audience info, such as job titles, company size, and industries.
- Known pain points or motivations (or I ask AI to help me uncover them).
- Distribution channel and key KPI for the blog.
Similarly, if I want to train a tool to help with sales, I’d leverage the HubSpot custom integration with Claude to connect my company’s CRM data to it. That way, all recommendations from Claude will be heavily customized to my customer pain points and motivations.
5. Test, refine, and save what works.
Now, for the most critical part, feedback. I usually test a lot of different prompts for my use cases, adjust based on what’s off, and once I like the result, I save it. That might mean:
- Adding the prompt to my swipe file.
- Bookmarking a good ChatGPT thread.
- Updating my custom GPT instructions.
- Turning the whole thing into a repeatable process for future tasks.
Every time I do this, it gets easier to use AI again, and faster to get something I can actually publish.
The State of Artificial Intelligence in 2025
New research into how marketers are using AI and key insights into the future of marketing.
- Marketing AI Tools
- Practical Tips
- Trends and Statistics
- And More!
Download Free
All fields are required.
AI Use Cases for Marketers
Dharmesh Shah, the CTO of HubSpot, believes teams in the future will be hybrid, with a mix of humans and AI agents. We’re already seeing AI tackle a lot of different use cases for marketers and help them become more productive.
Below are the use cases that have worked for me, along with the specific tools and workflows that made each one possible.
Executive Assistant
I personally use ChatGPT as a form of accountability. At the start of each week, I share my task list and campaign goals with ChatGPT. It groups tasks by priority, how long they’ll take, and what depends on what. This makes it easier to see what needs to be done first and what can be done together.
I also use it to summarize long meeting notes or Slack threads into short action points, along with reminders for follow-ups.
The best part? I can do all of this with voice notes. I just speak to AI like I would to my assistant, and it logs everything I tell it.
Apart from personal productivity, you can also use AI to handle repetitive business tasks. For example, I spoke to Cordes Owen, who runs Bake My Pies and uses AI to screen candidates.
“Our custom-built AI bot, Interview Assistant, conducts voice‑to‑voice interviews. Candidates literally ‘speak’ with an LLM that probes job requirements, soft skills, and culture fit.”
Data Analysis
As marketers, we’re used to dealing with a ton of data — campaign reports, analytics, tracking reports, and more. I used to spend a lot of time fixing formulas and cleaning up messy spreadsheets. However, AI has made my job a lot easier.
When something breaks, I copy the formula or a small part of the data and paste it into ChatGPT to ask what’s wrong. Sometimes I describe what I want the formula to do, and it gives me the correct version.
If the problem is bigger (like cleaning up a whole sheet), I upload a sample file and ask AI to write the steps or code to fix it in Excel or Google Sheets.
For example, when I was building a campaign performance report, my click-through rate formula kept showing an error. I pasted it into ChatGPT, and it found that I was dividing by the wrong column.
Personalization at Scale
We all know it: Personalized content converts far better than mass emailing your list. However, doing it manually doesn’t scale very well, especially for larger companies. This is where I’ve found AI to be quite useful.
For example, HubSpot’s team used AI to rewrite nurture-flow emails and segment sends by intent. By doing this, we received an 82% lift in conversions and 30% higher open rates. Here’s the exact process we followed:
Similarly, a good recommendation engine customized to what the user is browsing at that time can easily boost your order value. You can easily use HubSpot tools like Breeze to integrate your CRM data and content patterns to auto-personalize recommendations, emails, and landing pages at scale.
Lead Generation
Lead generation is a hectic process. You need to collect a ton of data, create customer profiles, understand their motivations, and nail the right message to get them to respond. Don’t get me started about the relentless follow-ups.
This process typically takes days (sometimes months) to complete, cutting into time that could be spent delivering value.
AI is quite strong at multiple aspects of the lead generation process. We’ve already discussed personalization at scale, which lends itself well to the cold messaging aspect of lead generation. However, it can also filter data, sort by buying intent, and find repeatable patterns to boost the conversion rates for your salespeople.
With the time saved, salespeople can better use their time by contacting qualified leads, establishing relationships with new clients, and making the all-important sale.
Streamlining Content Creation
This is where I use AI every day. Here are a few micro‑use cases:
- Ideating topics, intros, and outlines for campaigns.
- Rewriting sections for different audiences (e.g., B2B vs. B2C tone).
- Summarizing long reports or research into social posts.
- Taking one blog and spinning it into email snippets, LinkedIn carousels, and newsletter leads.
For example, I used HubSpot Breeze to create an outline + first draft, then manually refined it. It shaved several hours off the process. Similarly, HubSpot’s Case Study Generator pulls together transcripts, emails, and notes to draft case studies. The possibilities with AI-led content creation are endless.
Here is why I think every marketer should learn AI.
Already, 92% of marketers report that AI has impacted their role, with more than a third saying “very significantly.” However, I think we’re just at the tip of the iceberg. Models are only going to get better, agents are going to get more capable, and teams will adopt AI as a part of their hybrid workflow.
That’s why it’s so critical for marketers to learn how AI actually works. Invest in learning AI now, and it’ll help you keep up with the growing demand for AI in marketing.
Editor's note: This post was originally published in April 2023 and has been updated for comprehensiveness.
The State of Artificial Intelligence in 2025
New research into how marketers are using AI and key insights into the future of marketing.
- Marketing AI Tools
- Practical Tips
- Trends and Statistics
- And More!
Download Free
All fields are required.
Artificial Intelligence
![AI email subject lines that drive 3x more revenue and actually convert [+ exclusive insights]](https://53.fs1.hubspotusercontent-na1.net/hubfs/53/ai-email-optimization-1-20251014-4500151-1.webp)
-1-20250905-2237709%202.webp)
