Email marketing’s history can guide AI governance

AI is where email was at the start of the 21st century, when rapid development created the need to set limits, create rules for responsible behavior, and develop a set of generally accepted best practices.

Email marketers learned the importance of establishing their own standards as governments and technology providers developed rules for the channel. As AI adoption accelerates, those lessons can help guide the establishment of ethical standards for its use.

As I’ve explained in earlier MarTech columns, email marketers faced a familiar challenge: How to deal with new technology without a unified framework of government rules and regulations.

High-level thinking about the ethics of this new technology can be overwhelming. We’re learning to use this technology and developing its rules at the same time.

Much of the AI-related discussion in the news centers on who gets to decide what the rules should be, whether it’s the AI companies or — God forbid — the government.

Email had this same problem in the pre-permission age, thanks to unfettered email address acquisition and firehoses of messaging, which nearly sank the channel until responsible people realized we needed to govern ourselves better.

That was before Congress followed EU countries’ examples and passed the CAN-SPAM Act.

At the same time, email practitioners and industry thought leaders worked across brands, tech providers, and industry interest groups to develop a code of ethics based on what we’d learned from email marketing and previous work in marketing and technology.

It’s not written in a single codebook. But this generally accepted set of practices cascaded to companies that power email, such as ESPs and marketing technology providers.

Now we have laws, ESP and martech acceptable-use policies, blocklist rules, and organization-level codes of conduct, along with a library of books, articles, training workshops, and conference presentations. They don’t agree on the details, but they map a path.

AI is moving faster than email

At no other time in history have we gone from idea to execution and final product so quickly. We thought we moved fast with email, but AI development is outpacing email at light speed.

The email industry developed its code of conduct over two decades, and it’s still developing. ChatGPT launched in 2022, and its adoption took off immediately. Some forms of AI and machine learning existed for years, but they didn’t have the reach and budget-friendly cost.

This leads to another parallel: how AI’s growth can be good and bad at the same time.

Fast AI growth is good. I can get stuff done faster. I can come up with an idea, take the dog for a walk to think about the possibilities, then sit back at my computer, fire up my AI assistant, and boom! I have a plan.

Fast AI growth is bad. AI and how we use it are evolving faster than our understanding of the impact. If we’ve learned anything over the last 100 years or so, it’s that making big changes without considering the consequences can create disaster.

It’s not just about the burgeoning growth of data centers to power AI, although we need to think about that, too. It’s about our personal use of AI and the need to govern ourselves before someone makes those decisions for us.

Let’s organize the discussion on ethics right here

This is happening all over, but just as in email’s early days, conversations are fragmented, ad hoc, and dominated by special interests and radically opposing viewpoints, from “AI will save us!” to “AI will kill us all!”

What are we doing now to be ethical actors? I’ve always maintained that we have to govern ourselves first. Now we have to take that from governance to ethics. Let’s do that now.

The three As of ethical AI use

1. Attribution

I could fill a library with all of my blog posts, white papers, industry articles like this one, company reports, keynote speeches, customer guides, webinar presentations, PowerPoints, and who knows what else that I’ve created over the last 20-plus years. So could many of my fellow email thought leaders.

Along the way, two people coined phrases that — at least among the old guard — almost instantly became part of the lexicon:

• Loren McDonald, an early email pioneer, mashed the concepts of mobile, social, and email messaging into a single concept, “mocial,” in 2011. In 2009, he also originated the phrase “Average is the new bottom,” meaning average benchmarks are no longer acceptable.
• Kath Pay created and named “holistic email marketing,” which centers marketing strategy around customers and their needs, using email as the primary channel to engage and guide them through their brand journeys.

These phrases became inextricably intertwined with their creators. If another marketer used them without acknowledging their creators, we recognized that. “Yeah, Loren talks about that.” Or “Kath came up with that concept. Stop stealing her stuff!”

Now, go to Google and search “mocial.” Or “What is holistic email marketing?” The AI Overviews sum up the concepts accurately enough, but they don’t name the creators. You must click on the attribution links or scroll down through layers of links to find results that mention names.

Let’s say I want to write about ESPs and headless commerce. I’ll ask my good buddy Claude some questions. It’ll do the research and bring me a perspective that answers my questions. But it doesn’t tell me who came up with the ideas it found.

Some AI models, including Google Gemini, cite URLs so you can track down their sources. But are you going to do that if you want Claude to deliver a fully formed article that you can slap your name on?

The problem is that AI’s results consolidate information from many sources. It might give you the source, but not the person who originated the thinking.

If I publish that article without attributing my sources, AI could treat me as the authority on that topic. The next time someone writes about ESPs and headless commerce, their LLM will regurgitate my thoughts and perspectives without naming me. That writer will present my article as their idea. Now we’re even farther downstream in attribution.

Aside from potentially perpetuating outdated or incorrect thinking, this raises the ethical problem of presenting someone else’s thoughts as your own — a form of plagiarism.

Attribution as a property of ethical behavior means asking our AI assistants, “Where did you get this information?”

Your information might have come from AI, or your AI buddy wrote your article for you, but that doesn’t mean AI invented it. A human had to come up with the original material.

Our responsibility as ethical people is to ask Claude, ChatGPT, or Perplexity to cite its sources and for us to acknowledge them.

My standing rule in my AI prompting process tells my AI assistant to cite sources and explain how it found them. I check every reference every time to determine whether AI made up a concept.

If you don’t do it, that verges on plagiarism, even if you change a few peripheral things to try to make the information your own.

Plagiarism itself is due for a redefinition. Classic plagiarism is passing off someone else’s content as your own. But if you copy an AI article and put your name on it, and it turns out AI lifted key passages or phrases from another source, is that still plagiarism? Can you plagiarize an AI Overview?

Transparency everywhere

Our responsibility as ethical AI users is to give due credit to our sources, both in the material we gather and in the information we share with others, whether it’s our readers, our coworkers, or our customers and clients.

We should add a disclaimer indicating the role AI played in our thought process or content creation. I do that with my client deliverables. I don’t give away all the sources I cite. I outline where AI might have influenced me.

2. Accuracy

Let’s dispel the illusions that AI is smarter than we are and that we should assume everything it tells us is true. We know AI makes up stuff, that it hallucinates or lies. Poorly worded or biased queries or prompts, which I’ve discussed in previous columns here on MarTech, often drive AI’s fairy tales.

I’ve also seen this in action. I’m building an agentic model right now. I discovered that one of the model’s agents was making up things, such as inflating the number of people we interviewed for a project and then reporting that incorrect information in a client report.

That leads me to create another rule: “Don’t put in anything that the facts don’t support.”

That’s not exactly a hallucination, more like an exaggeration. However, it’s still inaccurate. It was a weakness in the agent build, and I had to recognize the slip. It’s also a clear reminder that not everything AI generates is correct.

Assume that everything your AI agent tells you is wrong. If that sounds too extreme, ask it to prove everything it has told you is accurate.

This helps keep you ethical and preserves your brand value. Your reputation rides on everything you publish or distribute. That’s why you can’t assume AI is right.

3. Authenticity

I confess: I let AI write some of my emails. That’s because I suck at it. I hurry. I leave things out. I need to be sure my clients have the full picture and that I’ve considered everything that needs to go into that message. I’ll ask ChatGPT or Claude to clean up a message before I send it.

Now, does the email reflect my words or Claude’s? Claude knows my voice and tone because I spent weeks training it on my personal and professional voices. It also knows our previous discussion. But the email reflects Claude’s work, influenced by me. I review every AI-assisted email before it goes out to be sure it says what I intended.

If you use AI to reply to emails without overseeing them, you’re not being authentic. Authenticity is an emotion rooted in the human brain and in customers’ perceptions of your brand.

This matters because your email messages represent your company’s brand, whether you’re sending personal messages or branded campaigns. That can help or hurt your company’s perceived integrity.

It’s easier to let AI handle your email communications, but you’re not being authentic. That’s one reason many companies struggle to give AI full control over repetitive tasks like replying to emails.

Without oversight, you can end up in a special kind of email hell. This happened to me during a client email exchange with many people on it, and it became clear that the conversation had devolved into a back-and-forth between AI agents.

That’s how AI can take the human interaction out of decision-making and rob an exchange of authenticity.

Time to state your beliefs

These may seem like simple things, but they’re ethical considerations because they affect your brand.

We’re talking about how you portray yourself, whether it’s your personal image or your company.

AI is advancing much faster than we can define the ethics of using it. That happened with email 25 years ago, too, when email took off faster than we could figure out whether we were making serious mistakes with it. Nobody even thought about ethics until we had time to think about what we were doing and how it affected our customers, subscribers, marketing teams, and brands.

Much of the commonly accepted ethical code that governs email use came from those mistakes. Today, I’m seeing mistakes in how people use AI and in AI infrastructure.

I hope people are trying to be ethical in their use of AI. But now it’s time for us, the users, to take control and start writing and talking about AI ethics. We need to build in guardrails and ethics so we can develop and share an industry-wide understanding.

Suppose someone — your boss, maybe, or a VIP customer — asks you, “What’s your AI ethics policy?” Could you hand over a document that lists your code of conduct? Or would you just say something noncommittal about following industry best practices?

Now is the time to sit down with your team and create a policy that covers your ethical use of AI in all circumstances and ensures that you — not your AI assistant — remain in control.