What the Numbers Are Actually Telling You
LinkedIn now has a complaint button for AI-generated content. Users see a post, decide it’s slop, click the button, and a negative signal goes straight back to the algorithm. One click. No friction.
That button didn’t appear out of nowhere. It appeared because the platform had no choice.
Here’s the situation as of mid-2025: there’s roughly 20-25% more AI-generated content being published on LinkedIn than there was a year ago. Engagement across the platform is down about 17%. Those two numbers are not unrelated. The volume went up, the quality didn’t follow, and the algorithm which exists to surface content people actually want to read is now actively working against the content that’s dragging down the experience.
If your team has a content flow that auto-publishes, that flow is now one of the things the algorithm is trying to filter out.
This isn’t a technology problem. The tools work fine. This is what happens when organizations adopt automation faster than they build the judgment to use it well.
Why Hyper-Personalization Backfired and What Replaced It
The same pattern showed up in outbound sales, and it played out even faster.
Two years ago, tools like Clay, Instantly, and Warmly made it possible to send highly personalized cold outreach at scale scraping LinkedIn for recent posts, pulling funding announcements, referencing specific details about a prospect’s business to make the email feel like it came from someone who did their homework. For a window of time, it worked. The signal was real because it was rare.
Then everyone ran the same playbook. The signal became noise. And the personalization started going wrong in ways that were worse than generic wrong company names, hallucinated context, misattributed details that made it instantly obvious the email was generated by an agent that had scraped the wrong Kevin Williams from a data broker list.
I got one of these emails recently. The hook was accurate it referenced something real about Ascend Labs. Then it went sideways. The scraper had pulled something from a completely different context, the automated SDR had woven it into a pitch that made no sense, and nobody had read it before it went out. The personalization that was supposed to build trust destroyed it in one sentence.
What the people who actually run outbound at scale will tell you now is straightforward: stop chasing the personalization. What converts is simpler. This is who we work with. This is the problem we solve. This is why it matters for your situation. That’s it. A clear, relevant message to the right person at the right time consistently outperforms a personalized mess especially when the personalization is obviously manufactured.
The insight behind this is worth sitting with: the value was never in the personalization itself. The value was in reaching the right person at the moment they actually had the problem you solve. AI tools are genuinely good at helping identify that moment the new executive with fresh budget, the company at the inflection point where your solution becomes relevant. That’s where the leverage is. Not in faking familiarity.
The AI Slop Feedback Loop and How It Ends
Here’s a direct answer to a question a lot of marketing leaders are asking right now: Why is our content reaching fewer people even though we’re publishing more?
The answer is a feedback loop. AI-generated content at volume without human review tends to be average by design LLMs revert to the mean. Average content generates less engagement. Less engagement teaches the algorithm the content isn’t worth distributing. Lower distribution means even the good content from that account gets suppressed over time. The complaint button accelerates the cycle by adding a direct negative signal on top of the engagement signal.
Anthropics watermarking rollout now live and required under the EU AI Act adds another layer. The watermarks aren’t Unicode artifacts you can strip out. They’re probabilistic patterns baked into how the text is formed at the word level, designed to survive copying and pasting. There are already tools trying to humanize AI output past detection, and there are already detection tools training on those humanizers to close the gap. It’s an arms race, and the arms race favors the platforms, not the publishers.
This is where the organizational problem becomes visible. The CMO who told their agent to generate a month of content without review, the sales team lead who let the SDR tool run unsupervised, the candidate who submitted an AI-generated work sample without reading it they all made the same decision. They removed the human from the loop and called it efficiency.
It wasn’t efficiency. It was abdication. And the platforms are now building infrastructure to make sure it costs something.
Listen to the full conversation with Matt Graham:
What Actually Works and What to Do Monday
None of this means AI is the wrong tool for content and marketing. It means the teams using it without a judgment layer are about to have a harder time, and the teams that kept a human in the loop are about to have a relative advantage.
Matt’s team at his company runs a process worth paying attention to: they’ve built a voice model from three years of recorded content and LinkedIn posts, use it to draft, then run multiple review steps before anything goes out. The AI handles speed. The humans handle accuracy and distinctiveness. The output sounds like Matt because actual Matt is still in the chain.
That’s the model. Not AI instead of human judgment AI plus human judgment, in the right order.
For content: the differentiation that matters right now comes from having something original to say. An LLM asked to write about what’s happening in AI today will produce something that sounds like the consensus, because the consensus is what it was trained on. A conversation between two people who are actually watching things change in real time like the one that became this post produces something the LLM can’t generate on its own, because it didn’t exist yet. That’s the raw material. AI is the processing layer on top of it, not the source.
For outbound: use the tools to identify the right moment, not to manufacture the right message. When someone moves into a new role, when a company hits a growth inflection, when a specific signal suggests your solution has become relevant that’s when a clear, direct message from a real person cuts through. The personalization that matters is timing and relevance, not a reference to someone’s most recent LinkedIn post.
For your team: the question worth asking is not “are we using AI?” It’s “who is reading the output before it goes out?” If the honest answer is nobody, that’s the thing to fix. Not the tool. The accountability structure around the tool.
The window to course-correct on automated content and outbound is right now, before the algorithm and the complaint button do the correcting for you.
If you want to think through what this means for your specific team’s approach, I set aside time every week for exactly these conversations: tidycal.com/kevinwilliams
Or start with the free assessment to figure out where your biggest gaps are: https://assessment.ascendlabs.ai/
♻️ Repost this if someone in your network needs to hear it.
🔔 Follow Kevin Williams for the real side of AI adoption.
