I recently received a cold email from a virtual AI agent. I usually ignore these. I get enough sales emails as it is, and the fact that an AI agent can now send me a personalized pitch doesn’t make me any more interested in receiving one. But this particular email came from a pretty big technology company, one we actually use at Dash Logistics Systems, so I figured I’d give it a read.
It was immediately clear that the system knew I worked for Dash Logistics. Unfortunately, that’s about where the accuracy ended.
The email assumed Dash Logistics was an Industrial IoT company. I have worked in Industrial IoT, but that was several years and several jobs ago, and it has nothing to do with what I do today. It then complimented my background in analytics before asking if I needed help with analytics. It had gathered several accurate facts about me, stitched them into something that sounded personalized, and completely missed the context that would have made any of it relevant.
I expect better.
And to be clear, I’m not opposed to the idea of AI assembling information about a prospect. That’s basically what good sales research has always been. A salesperson looks at your company, your role, your background, and what your organization is doing, then figures out whether there’s a legitimate reason to have a conversation. AI should be able to do that faster and at a much larger scale. That’s the opportunity.
The problem is when we confuse data assembly with understanding.
There was nothing wrong with the individual facts in that email. I really do work at Dash Logistics. I really do have an analytics background. I really did spend time in Industrial IoT. The problem was that the AI assembled those facts into a story that wasn’t true. That’s not a failure of personalization. It’s a failure of quality control, and when you’re doing cold outreach, quality control is the whole job.
You probably get one chance with a prospect. If I don’t know you, I don’t owe you my time. You have a few seconds to convince me there’s a reason to keep reading, and if you spend those seconds demonstrating that you don’t understand my company or my role, you’ve lost the opportunity before you’ve told me what you’re selling. Don’t ruin that by being lazy.
There’s already a name for this. AI slop is the same idea as those AI generated images with three hands or fifteen fingers: something that was clearly produced by a machine, and clearly never looked at by a person before it went out into the world. We’ve reached a point where it’s incredibly easy to produce something that looks like good work. The email is polished, the grammar is perfect, it mentions my name and my company, it references something from my background. It sounds professional. But when you actually read it, the thought isn’t there. We’ve all seen the same thing on LinkedIn: posts that are perfectly structured but completely generic, content that’s clearly been optimized to look thoughtful rather than actually being thoughtful. The problem isn’t that AI wrote it. The problem is that nobody bothered to check whether it was any good.
There’s also a specific irony to this particular email that’s worth sitting with. The company sending it sells AI. This wasn’t some unrelated vendor dabbling in automation. Their AI email, pitching their AI product, was itself a bad advertisement for what that product could do. If your outreach is the first thing a prospect experiences from your company, and that outreach is generated by the very technology you’re trying to sell, you’d better make sure it’s a good demo. This one wasn’t. It undercut its own pitch before it got to make one.
Which is what makes the contrast so sharp when you look at what actually builds trust with a customer. A few weeks ago I had a run of small interactions at different stores that had nothing to do with AI and everything to do with the kind of thing that quietly earns repeat business. At the grocery store, I’d picked up too many tomatoes at self-checkout, and instead of asking me to wait for someone, the cashier walked off, found the manager himself, brought her back to the register, and had the line item fixed before I’d even finished bagging everything else. At another store, an employee took a few seconds to give my three year old daughter a high five and ask her about her day, which made her whole afternoon and mine along with it. At a third, one employee was running an entire sales floor by herself during a busy weekend and somehow kept it feeling calm and unhurried the entire time.
None of those people had access to my LinkedIn profile. None of them ran a report on my purchase history or generated anything personalized ahead of time. They just paid attention to what was actually happening in front of them and did something small and genuine about it, on the spot. That’s the whole difference between real personalization and the AI slop version of it. It doesn’t take more information. It takes attention, and a willingness to actually notice the person in front of you instead of running a script at them. Those interactions cost the stores nothing but a few seconds of effort, and they’re a big part of why I keep going back.
That’s the bar. Not more data, not a faster pitch, just someone paying enough attention to get it right. And that distinction matters, because I’m actually pretty bullish on AI. Used well, it’s a tremendous force multiplier. It can help a product manager synthesize customer feedback, help an analyst find patterns in data, help a developer get through the repetitive parts of a job so they can spend more time on the parts that require judgment. It takes a capable person and gives them leverage they didn’t have before. That’s where this technology gets genuinely exciting.
But the reverse is just as true. Good AI is a force multiplier. Bad AI is a force de-multiplier.
If AI gives me a solid first draft, I can make it better. If it saves me an hour of research, that’s real value. If it produces something wrong or irrelevant, now I have to spend time figuring out what went wrong and fixing it, which means the technology hasn’t saved me anything. It’s created a new task disguised as a finished one. That’s essentially what happened with this email. It probably saved someone at that company from doing the research by hand. It didn’t save me any time, and it didn’t do their brand any favors either. If anything, it made me wonder how much of the rest of their sales process is built on the same shortcut: gather information, run it through a system, generate something that sounds good, move to the next prospect.
Curious whether I was just being cranky about it, I ran a completely unscientific poll on LinkedIn. I asked how obvious or inaccurate AI outreach affects people’s response to it. Eight people voted, which is not a sample size that belongs anywhere near a research paper, but the results were still telling: 75% called it an immediate turn-off, 13% a minor annoyance, 13% picked “other,” and nobody said it was no factor at all.
Eight votes prove nothing on their own. But I don’t think the underlying idea is surprising to anyone who’s spent time online lately. We’re getting better at recognizing AI slop, and the better we get at spotting it, the worse it performs. What might have looked like impressive personalization a year ago now reads as a machine that pulled a few facts off my LinkedIn profile and stitched them into a paragraph. That’s not personalization. That’s data assembly with no editor.
None of this means we should stop automating outreach. It means we need to be more deliberate about how. Just because AI can research a prospect, draft the email, and hit send in seconds doesn’t mean it should be allowed to do all three unsupervised. AI is a tool, not a replacement, and tools still need someone operating them. Someone still has to ask one simple question before that email goes out: is this actually true? Does it describe the company correctly? Is this experience relevant to the person’s current role? Is there a real reason for this prospect to care about what we’re selling? If the answer is no, don’t send it. That doesn’t mean writing every email by hand again. It means caring about the output, not just the volume.
That’s the part scale changes. One salesperson sending a bad email is a bad email. An AI agent sending ten thousand bad emails is a strategy, and if those emails are inaccurate or obviously manufactured, that strategy just automated a bad first impression across thousands of potential customers at once.
Which is really the whole lesson. AI can draft the email, research the account, and assemble the facts. What it can’t do, at least not yet, is know whether the story it just told is actually true. That step still belongs to a person, and skipping it is exactly what turns a force multiplier into ten thousand bad first impressions.
I’d rather get a simple, honest email that says “here’s what we do and here’s why I think it’s relevant to you” than a beautifully written paragraph that proves, in the first ten seconds, that the sender doesn’t understand my business. A generic email is easy to dismiss. It tells me you don’t know me. Bad AI personalization is worse, because it tries to convince me you do, and then demonstrates that you don’t.
Data assembly is fine. Bad data assembly isn’t. You get one chance with a prospect, the same way that cashier had one chance to fix my checkout line without making me wait. He didn’t need an algorithm to figure out the right move. He just paid attention and acted on it. Don’t ruin your one chance by being lazy, and don’t assume the machine will catch what only a person would notice.