The Reason Your Engagement Is Slipping Has a Name. And It’s Not the Algorithm.

I scroll social media every day and I cringe.

The captions, the blog posts, the carousels, even the BOOKS now. Creators and industry leaders I genuinely respect, people whose work I used to read closely, are publishing content that screams ChatGPT-with-no-guardrails. I get it. I understand what they’re trying to do. They’re learning the tool. They’re trying to save time so they can build and scale. I’ve been there. I’m STILL there, every day, trying to figure out where AI helps and where it ruins what I’m doing.

But the gap between “using AI” and “publishing AI’s default output” is the thing that’s costing them right now and they don’t see it yet. And that gap just got measured.

What Sprout Social Just Confirmed

In May, Sprout Social published findings from their Q1 2026 Pulse Survey of more than 2,000 social media users across the US, UK, and Australia. Buried in the section on AI-generated content was this sentence:

“As consumers have gotten savvier, key ‘tells’ of AI-generated content have emerged, including the overuse of em dashes, sentences that are all the same length, three-point lists, and ‘it’s not X, it’s Y’ statements.”

Read that again. A major social media research firm just published the exact patterns I’ve been documenting in my own writing guardrails for over a year. Em dashes. Cadence uniformity. Three-point lists. Mirrored contrast. THE EXACT TELLS.

The data backs it. Sprout found that 28% of social users now name unlabeled AI content as their #1 brand turn-off, beating engagement bait by five points. 88% report declining trust in social media news because of AI content. 66% are more selective about what they engage with than they were a year ago. Half of Gen Z will block or unfollow an account for posting AI slop.

Six in ten consumers report being less likely to engage with brand content in the current AI atmosphere. That means creators are losing reach right now, and most of them haven’t figured out this is why.

Scrolling, Cringing, Watching It Happen

The creators getting hit hardest right now are the ones who saw early gains from AI-assisted content. They built audiences on their pre-AI voice. They added AI in 2024 or 2025 to scale. Their content got more polished and more frequent. And somewhere in the last twelve months, the engagement started slipping.

Here’s what’s actually happening. Their audience trained on their original voice, noticed the shift, and started disengaging. Not consciously. Most readers can’t name an em dash or describe what mirrored contrast does. They just feel something is off, scroll past faster, and stop opening the emails.

We don’t have public creator-level proof of this yet. Nobody has published the data showing “this Substack writer started using AI in March and lost 40% of their open rate by July.” But the consumer survey data points exactly where you’d expect it to if this mechanism is real. When 60% of consumers say they’re less likely to engage with brand content because of AI, that engagement has to be deducting from somewhere. It’s leaving the accounts whose content used to work and is now getting scrolled past.

The Pattern Goes Beyond Writing

The same convergence is happening across every surface AI touches, not just captions and blogs.

Websites built with AI tools are starting to look identical. Click on five different AI-generated landing pages and you’ll see the same blue and purple gradient at the top. The same rounded boxes stacked down the page. The same big bold headline. The same three-column features section. The same testimonial carousel. The same call-to-action button in the same shade of indigo. Different tools, different prompts, same result. Designer Michal Malewicz started a movement called SLOPLESS in April specifically to push back against this. Designers are publicly calling it out because they can already see what your audience is starting to feel.

ChatGPT images have their own tell. It’s called the “yellow tint” (or, less politely, the “piss filter”), and it went viral in March 2025 with its own Know Your Meme page. ChatGPT defaults to generating images with a sepia cast because DALL-E 3 uses upstream prompt rewriting that keeps injecting “warm cinematic lighting” and “golden hour” into every request. Users can’t disable it. Tools called UnYellowGPT and Yellowtint exist specifically to fix it after the fact. Once you see the yellow, you can’t unsee it. And consumers have seen it. A LOT.

Carousels and ads built by AI have the same problem. Scroll Instagram or Facebook for ten minutes and you’ll spot AI-built ads before you read the copy. Same layouts. Same stock-style faces. Same color schemes. Same headline structure. The carousel templates everyone is using produce slides that could be swapped between accounts without anyone noticing.

Audiences are starting to feel “this is AI produced” across every medium at once, and they’re reacting to it across the board.

What’s Actually Costing You

In March, Google rolled out an algorithm update that targeted sites publishing AI-generated content at scale. The biggest content farms lost between 60% and 80% of their search traffic overnight. Some lost everything. That tells you what Google now considers low-quality content, and it tells you where the search algorithm is heading. A separate analysis by Graphite found that 82% of articles getting cited by ChatGPT and Perplexity right now are written by humans, only 18% by AI. The AI search engines are picking human writing too.

Substack is now explicitly marketing itself as a “sanctuary for human content.” The Emplifi 2026 Digital Authenticity report found that 85% of consumers will pay MORE for brands they perceive as authentic, and 91% expect brands to disclose AI use.

Now is the moment to address the obvious. You’re reading this on Substack. Substack is positioning itself as the place readers can come for human writing. And I’m telling you I use AI. Both things are true at the same time, and both things have to be true if my whole argument holds. Every piece I publish here starts with AI doing research, organizing the spine of the argument, and giving me a working first draft. Then I sit with the draft and edit it against my guardrails document, which is a documented list of patterns AI defaults to that I’ve decided my content will never use. Whole paragraphs get rewritten. Whole sections get cut. Sentences I would never say out loud get replaced with the version I would actually say. By the time you read it, the structure may have come from AI, but the voice is mine. That’s the only version of AI-assisted writing that belongs on a platform like this. The version everyone’s audience is starting to smell is the one that skipped the editing.

Read the trend lines together. Search is penalizing detectable AI. Engagement is collapsing across social. Consumers are paying premiums for authenticity, and premium publishers are now treating human-written content as the differentiator.

The Fix Is Not What You Think

The conclusion most people jump to is “stop using AI.” That’s the wrong call, and it’s not what the data supports.

The tool isn’t the problem. The output is.

Sprout’s own finding is precise on this. Hybrid work, where AI handles structure and the human handles voice, performs comparably to fully hand-built content for a fraction of the time. Patsy Wagner, the Associate Director of Global Content at Spotify, told Sprout: “AI content often sounds similar in cadence and punctuation. The more a brand defines its distinct tone, the more it will differentiate and connect with customers. Humans are drawn to the zags, or minor errors. AI produces textbook content, but that doesn’t get noticed.”

The fix is to keep using AI and stop SOUNDING like it. The choice is between abandoning the capacity AI has built into your work, or giving AI the specific rules it needs to stop defaulting to the same patterns as everyone else.

Specific Rules, Not Vibes

When I say specific rules, I don’t mean “write in a more conversational tone.” I don’t mean “make it sound more human.” I don’t mean pasting your bio into the system prompt and hoping for the best. Those prompts produce the exact generic output everyone’s audience is now learning to smell.

I mean a documented list of the patterns AI defaults to that you’ve decided your content will never use. Mirrored contrast. Em dashes. Eighty-plus other patterns I’ve catalogued from two years of obsessive editing sessions, organized by category, integrated with a five-pass editing process that walks the model through every pattern systematically. Plus a customization layer that captures YOUR specific vocabulary and the words you’d never use.

That’s the system. Sprout just caught up to what I’ve been trying to shout from the rooftops. What separates content the audience smells as AI from content that reads like you actually wrote it is whether the model has a documented list of patterns to avoid and patterns to use.

Every piece of content I publish runs through my guardrails. The article you’re reading right now went through five editing passes against the document before it landed in your inbox. That’s why my AI-assisted output doesn’t trigger the same recognition pattern the data is now naming. And every time I scroll past another caption from someone I respect that screams ChatGPT defaults, I think about how preventable this is. The creators losing reach right now are skilled people who just don’t have the rules yet.

Done well, AI gives you capacity. Get it wrong, and your audience walks.

If you want the system, I packaged it as AI Writing Guardrails at guardrails.aiwithleah.com. Two tiers. The $97 self-guided version gives you the full 92-pattern framework and the builder to customize it to your voice. The $297 custom build adds my personal analysis of your content and a guardrails document I review and refine before delivery.

The data caught up in Q1 2026. Your audience is already operating on it.

Get the system →

P.S. If you’re reading this on Substack and thinking “wait, does MY content sound like AI?” run a quick test. Open your last three captions or blog posts. Count the em dashes. Count the three-point lists. If the number is above zero on either, your audience already noticed. You can start fixing it this week.