My team publishes a newsletter every Monday. Three blocks: a case study, an insight, and one actionable idea. This week, the insight block stopped me mid-read.
The writer had pulled research from a Wharton professor’s February 2026 article in Harvard Business Review, added three academic studies, and built an argument I have been turning over in my head ever since. Not because it was alarming (though parts of it are), but because it reframed something I work with every day in a way I had not articulated clearly before.
Here is the short version: two simultaneous revolutions are reshaping how brands reach people and how purchase decisions actually get made. And email, the channel most people have been writing obituaries for since 2002, is structurally better positioned to survive both than almost anything else in the digital marketing stack.
Let me explain why I think that, and where it gets complicated.
Email Marketing AI Disruption Happening
The first revolution is already happening
For two decades, the standard model of digital marketing ran like this.
Someone wants a product. They type a query into Google. They get a list of links. They click through to a website or two, compare options, and eventually buy. Brands competed furiously for visibility in that list, spending billions on SEO, content, and page optimisation to be one of the links that got clicked.
That model is breaking down faster than most marketing teams have noticed.
Research from London Business School and UCLA (link above), tracking over a million users via ComScore browsing data, found that online searches drop by roughly 20% after people adopt ChatGPT.
The customers driving that drop are not casual browsers, but high-retail-activity users. Basically, the people who transact frequently and spend more. The most valuable customers are changing their behaviour the fastest.
When someone asks a chatbot for a product recommendation, they get a synthesised answer. Two or three options, curated by an algorithm, with no list of ten blue links to scroll through. Brands that appear in that synthesis get disproportionate visibility. Brands that don’t might as well not exist. 🤷♂️
Google’s own response made things worse. AI Overviews now answer queries at the top of search results before users ever reach the organic listings that brands spent years and significant budget building. By early 2026, nearly 60% of all US Google searches ended without a single click to any external website. Users got an answer and moved on. No page visit, no session, no conversion opportunity.
For brands whose acquisition strategy depends heavily on people clicking through from search, this is not merely a theoretical future risk, but a present operational problem. 😬
The Stack Overflow signal that matters for everyone
One of the studies Puntoni cites tracked what happened to Stack Overflow after ChatGPT launched. Traffic collapsed, and it was not because the platform declined in quality, but because developers could get factual answers from a chatbot and no longer needed to ask other people.
Then the researchers looked at Reddit. Developer communities there saw no comparable decline.
The difference is purpose. Stack Overflow exists to deliver information. ChatGPT can now deliver that information faster. Reddit exists for community, debate, and the experience of talking to other humans. A language model cannot replicate that, however fluent it becomes.
The strategic signal here applies well beyond developer platforms. Brands built primarily on information delivery, such as comparison sites, guides, how-to content, and directories, face genuine near-term structural risk. Brands built on community, emotional connection, and experiences that cannot be summarised in a paragraph have more protection.
This is why newsletters with a genuine editorial voice are holding up.
It’s not because email is immune to AI disruption, but because the relationship between a subscriber and a trusted sender is not a delivery mechanism for information. It is something else… harder to name, harder to replicate, and apparently quite hard to automate away. 🙌
The second revolution has barely started
The first revolution is about how people discover products. The second is about who makes the purchase decision, and this is where it gets genuinely strange.
Marketing has always drawn on a distinction between the customer (the person making the purchase decision) and the consumer (the person using the product). Parents buy for children. Procurement managers buy for departments. These are not always the same person, and good marketing has always had to think about both.
AI agents are beginning to occupy the customer role autonomously.
Imagine a busy professional who tells their AI assistant: find me a project management tool for a team of twelve, budget under €200 per month, prioritise integrations with Slack and Google Workspace.
The agent researches options, compares specifications, reads pricing pages, and presents a recommendation or executes a purchase. No human ever visited a product page. The entire customer journey happened inside an algorithm.
This is already technically possible. What is missing is infrastructure, trust, and widespread adoption.
But Amazon has already moved on this. In late July 2025, the company pulled its entire product catalogue from Google Shopping, voluntarily disappearing from external search results for 31 days.
Nine days before that withdrawal, Amazon had rolled out its AI shopping assistant Rufus to all US customers. By autumn 2025, Rufus was driving over $10 billion in additional annual sales, with users who engaged with the assistant completing purchases at a 60% higher rate than those using traditional browsers. 🤯
Amazon did not pull from Google to save on commissions. It pulled because it was building for the customer of tomorrow, not the customer of today. When your internal AI can do a better job of converting intent into transaction than external search can, external search becomes an unnecessary middleman.
The two-audience problem nobody is designing for
Here is where I think most marketing teams are underestimating the shift.
Within the next two to three years, a meaningful proportion of purchasing decisions in B2B and high-consideration B2C will involve an AI agent doing the research and either recommending or executing the transaction. That agent will evaluate your brand by parsing structured data, reading API outputs, and processing machine-readable content.
It will not be moved by your brand story. It will not respond to emotional copywriting. It will not notice that your website is beautifully designed. It will look at whether your product specifications are clear, your pricing is structured legibly, your categorisation is accurate, and your claims are verifiable.
At the same time, the human who instructed the agent still needs to trust you. That trust is built through the kind of content that AI cannot replicate, such as editorial voice, consistent perspective, demonstrated expertise over time, and community.
If the agent finds your product and the human has never heard of you or has no reason to trust you, the transaction may not complete even after the agent recommends you. 😉
The brands that will hold ground through both revolutions are the ones that can simultaneously satisfy two entirely different audiences with entirely different needs. Machine-readable, structured, authoritative content for the agent, and (at the same time) a genuine human connection for the person.
Research from the Columbia and Yale sandbox study adds a layer of complexity that should make every digital strategist uncomfortable. AI agents have positional biases that vary across models. Yes, you read this right.
When tested on e-commerce platforms, GPT strongly favoured the first product in a row, Claude preferred the middle, and Gemini preferred the right side. These are not rational quality assessments. They are artefacts of how different models process sequential data. Optimising your digital presence for one model’s quirks may actively work against you with another.
The agents also penalise “sponsored” labels and respond to endorsements, behaviours that partially mirror human psychology but arise from entirely different mechanisms. And researchers found what they term AI-AI bias: AI systems rate AI-generated content more favourably than human-created content.
In a world where algorithms evaluate your marketing, the humans who wrote your best copy may lose out not because the writing is worse but because of structural biases in how the evaluation happens. 🫣
Nobody has fully solved this. The rules of what researchers are calling generative engine optimisation (GEO rather than SEO) are still being written. But the direction of travel is clear enough to act on now.
Why email sits at the intersection of both
Here is the argument my team’s newsletter piece made that I keep returning to.
Email bypasses the first revolution almost entirely. It does not compete for chatbot visibility. It does not depend on AI Overviews, leaving space for organic links. It does not need a human to search for you in order to reach them. It lands directly in an inbox, in a space the subscriber opted into, at a time they choose to engage with it.
That direct delivery model is worth more now than it was two years ago, because every other acquisition channel that depends on people discovering you through search is becoming structurally less reliable.
But email’s role in the second revolution is more interesting. A subscriber who receives your newsletter is building a relationship with your editorial voice, your perspective, your demonstrated expertise.
That is exactly the kind of trust that makes a human inclined to act on an AI agent’s recommendation rather than dismiss it. The agent might surface your product. Whether the human converts on that recommendation depends significantly on whether they already know and trust your brand. Email builds that trust at a level of depth that social media posts and search impressions rarely achieve.
There is also a practical implication for what email links to. If a subscriber clicks through from your newsletter to a landing page, that page needs to work for both human readers and for AI agents that may later crawl, parse, and cite it when advising other customers. Structured data, direct answers to common queries, clear product categorisation, verifiable claims… these are not just GEO best practices, but the architecture of a page that converts both audiences.
The loyalty programmes, subscriber-exclusive content, and editorial newsletters that deepen human connection are also, as the Stack Overflow vs Reddit finding suggests, the formats most resistant to being replaced by AI summaries. You cannot summarise a community. You cannot replicate a voice that a subscriber has read every Monday for two years.
What this means in practice
The strategic audit every marketing and brand team should run right now starts with one question: what proportion of your acquisition depends on people clicking through from search?
If the answer is high, the near-term exposure is real. That does not mean abandoning SEO or organic content, but it does mean urgently diversifying toward channels that do not depend on the discovery layer remaining intact.
Email is the most obvious candidate, but the work is not simply to send more emails. It is to build the kind of subscriber relationship that creates the human trust layer your AI-mediated marketing will eventually depend on. And it is to ensure that the content your emails link to is structured for both the human reading it now and the agent that may evaluate your brand later.
A few things worth doing now:
1. Track how your brand appears in AI responses.
Ask ChatGPT, Claude, and Gemini about your product category. See if you are mentioned. See how you are characterised. This is an early signal of your GEO position before the tools to measure it properly exist. However, what comes up for you can be very different from what comes up for another person, given their own particular interests and history/context.
2. Audit your landing pages for machine legibility.
Can an AI agent parse what you offer, at what price, with what specifications, supported by what evidence? If the answer requires a human to read between the lines, the page needs work.
3. Invest in subscriber depth over subscriber volume.
A list of people who have read your newsletter for a year and trust your editorial judgement is worth more in the emerging environment than a large list of low-engagement subscribers. That depth is the human trust layer that makes AI-mediated discovery convert.
4. And take the two-audience problem seriously.
Not as a future consideration, but as a design constraint that applies to content you are producing today. The page you write this month may be evaluated by an AI agent on behalf of a potential customer within the next 18 months. Whether it converts that evaluation depends on how you built it.
The inbox has survived mobile, social, spam, GDPR, and a dozen other disruptions that were supposed to make it obsolete. The reason is straightforward: it is a direct, opted-in, human connection that no intermediary controls.
In a world where two AI revolutions are simultaneously disrupting discovery and purchase decisions, that direct connection is becoming more structurally important than it has been in years.
My team surfaced this in a newsletter insight. I am still thinking about the implications.
Sources: Stefano Puntoni, Harvard Business Review, February 2026 | McKinsey State of AI 2025 | London Business School / UCLA ComScore study (cited in HBR) | Boston University Stack Overflow study (cited in HBR) | Columbia / Yale AI agent e-commerce sandbox (cited in HBR)
Leave a Reply