NEW! Listen to article
More and more startups and other tech providers are offering brands the ability to use generative AI to create unique marketing emails tailored to individual recipients based on collected data.
Although there’s undoubtedly a range of potential uses for that kind of capability, I see two as being the most likely:
- Bulk cold email outreach
- Phishing emails
The former is a relatively benign, yet annoying, form of spam, whereas the latter is classic malicious spam. Both would benefit from generative AI in a few ways.
Unfortunate Use Cases
1. Greater Scale
Generative AI allows faster creation of unique emails based on a message prompt and the addition of key recipient details, such as the person’s name, company, title, and other details scraped from corporate websites, LinkedIn, and other social media profiles. Soon, the process will be completely automated.
2. Avoidance of Spam Filters
The “uniqueness” of a generative AI-written email has a higher chance of eluding spam filters, in part by avoiding spam fingerprinting—a spam detection technique that looks for message elements in known spam emails in incoming emails. The variations in copy that generative AI produces will make spam fingerprinting less useful in filtering spam, increasing the rate at which cold emails and phishing emails alike reach inboxes.
3. Improved Writing
Generative AI offers significantly better writing output for phishers, who often draw suspicion because of spelling and grammar mistakes in their emails. That means higher success rates for phishers and more pain for their victims.
That may seem harsh, so now let’s talk about the…
Legitimate Use Cases
As generative AI becomes more sophisticated, trainable, and specialized, it will find many solid uses in the email space, especially among time- and resource-strapped small businesses. Sending one-to-one automated support and customer service email responses seems particularly promising.
However, generative AI is highly unlikely to ever be used to create one-to-one emails in the highest frequency use case: permission-based promotional marketing emails. That’s for a few reasons.
First, email marketers don’t want unique promotional emails. Sure, they want to personalize their emails—at least, some of them. But most aren’t interested in their emails’ being unique to the degree that generative AI offers.
That’s because those senders care about their brand image, and they want to control it and keep it consistent. The idea of turning control of brand messaging completely over to even a well-trained generative AI via an API would be anathema to a brand manager.
Second, as much as we’ve been talking about hyper-personalization in recent years, greater degrees of personalization result in diminishing returns.
Although marketers have tons of criteria by which to personalize their emails (our Segmentation & Personalization Ideas checklist details more than 170 of them), most of them will be either irrelevant or ineffective for most brands. That’s the challenge with personalization: to find those customer data points that significantly—even dramatically—move the needle for your brand while avoiding those that create noise, creepiness, or worse.
Third, on a per-email basis, a generative AI-created email would likely need to cost no more than 2 cents more than a traditional, human-created message, AND produce a significant return on investment, to be worthwhile. Even in the long term, that combination is probably a stretch.
That’s because traditional AI models and machine-learning models are entirely about increasing performance, whereas generative AI models aren’t trained on any performance-based data. So, although generative AI can, for example, produce 100 subject lines for your email, it has no clue which will generate the highest clicks or conversions for your message. That also means the subject lines it creates aren’t any more likely to be better than the ones you could come up with.
Even if generative AI is trained on performance data, as some vendors are already beginning to do, getting the cost-benefit balance in a healthy place both for vendors and for senders will be hard. Vendors have been trying to perfect machine-learning subject-line optimization engines for many years now with mixed results, so there’s no reason to assume that more complex models will achieve widespread success anytime soon.
And fourth, the prescriptive and machine-learning-driven segmentation, personalization, and automation methods that are the status quo at email service providers (ESPs) today offer brands more predictable and less expensive ways to create on-brand, tailored subscriber experiences than generative AI. And most brands are nowhere near maxing out the relevance-boosting capabilities of their ESPs.
For those reasons, generative AI will remain in an assistant role in the permission-based promotional marketing email space, helping copywriters brainstorm and refine their messages and helping designers create and refine their images.
Most Likely Outcomes
As generative AI fuels a growing tidal wave of bulk cold email outreach and phishing emails, it’s likely to lead to the rise of several kinds of deterrents, which will have varying levels of effectiveness.
1. Economic Deterrents
We’re in the early, market-building part of the generative AI industry cycle. As with the e-commerce and social media industries, concerns about profits will come in a few years—probably in 2026 or 2027.
Seeing as Microsoft alone has invested more than $10 billion in generative AI development, generative AI pricing models will likely need to change significantly in the years ahead. After all, Microsoft, Google, and others will want a return on their investments.
That said, the major generative AI engines would probably need to raise the price of their API access a lot to make bulk cold email outreach unprofitable. And making generative AI email phishing unprofitable through economic deterrence alone is probably not possible. So, other deterrents will be necessary.
2. Technological Deterrence
To improve their spam filtering for the age of generative AI, inbox providers such as Gmail, Yahoo Mail, and Outlook will incorporate more AI detection tools, such as OpenAI Detector and GPTZero.
However, using generative AI alone would be unlikely to trigger junking or blocking, although it might lead to increased rates of quarantining and throttling, which would slow ((but not stop)) the delivery of an email.
The detection of generative AI would become a new component of content filtering, which is just one of seven factors that affect email deliverability. However, other considerations, such as the reputation of URLs included in emails and the sender’s IP and domain reputations, would remain much bigger determinants of whether AI-crafted emails are junked or blocked.
Authentication will likely be given significantly greater weight, because spammers almost never take the time and effort to set up SPF, DKIM, and DMARC. Although Brand Indicators for Message Identification (BIMI) isn’t factored into spam filtering, it may be soon, acting like official letterhead for corporate senders. From a consumer standpoint, receiving an email from a major brand that doesn’t have BIMI in place will be increasingly suspicious.
As an aside, one thing I don’t expect inbox providers to do is flag or tag emails that appear to consist of mostly AI-generated copy or other content. That’s because many major inbox providers are also major players in generative AI, most notably Microsoft and Google.
Although I have a lot of confidence in inbox providers to combat spam and phishing, it’s unfair that we saddle them with the full weight of the burden. Eventually, the government will have to get involved and change its role from being a passive enabler of spam to an active fighter of spam.
3. Legislative Deterrence
The US has the most antiquated privacy and anti-spam laws among modern industrialized nations. Our 20-year-old CAN-SPAM Act of 2003 codifies opt-out marketing and stands in sharp contrast to international laws, such as Canada’s anti-spam legislation of 2010 and the EU’s not just talk.
The two most likely scenarios would be an AI-focused law that includes provisions about its use in marketing communications, and a privacy-focused law that includes provisions about generative AI and permission-based marketing. And then, of course, a follow-up law a couple of years later that clarifies all the confusing and unclear elements of the first law.
Because none of those deterrents is perfect, it’s likely all three will come to pass to some degree. For B2B marketers, that means that generative AI is likely a long-term dead end for bulk cold email outreach.
Indeed, generative AI may actually be the catalyst that brings about the end of bulk cold email outreach altogether.
More Resources on AI for Email
Four Ways to Empower Your Email Marketing Strategy With AI
Email Marketing and Artificial Intelligence: What’s Coming and What You Need to Know
