Growth Newsletter #268

A friend and I were dissecting a LinkedIn post from someone we both know, and it reeked of pure, unadulterated AI slop.

She asked a simple but important question: “Do you keep reading once you know it’s just pasted from AI?”

My response was a hard “No.” And I’ll be less likely to read another post from that person in the future, just because they’ve shown their hand. Their six-fingered hand, that is.

This anecdote speaks to the ways people (and platforms) have quickly learned to spot what’s genuine versus generated (what a CUNY professor describes as an “incredibly banal, realistic style.”) It also encapsulates how these types of posts and people are being categorically dismissed.

Since LLMs are so accessible and widespread, let’s start there, so you can be sure your brand’s associated text outputs (captions, ad copy, scripts, post on LinkedIn or X, web copy, emails, etc.) are boosting your credibility and reach.

Eight Ways To Nuke Your Copy Cred

To prove just how predictable AI writing can be, we threw a simple prompt into ChatGPT 4o to see how many instances of AI tells show up. While there are always exceptions, let’s look at some of the common (and lesser-known) AI text transgressions that can register as robot authorship:

• “It’s not just _____. It’s ______.” Similar formulaic, context-based comparisons such as these have been a darling of LLMs. While these can make a strong point or pivot in just a couple words, strive to find another format or approach to convey the idea.

• The much-maligned em dash: This well-documented “tell” has been a major blow to the infinitely versatile punctuation mark that writers notoriously overuse (guilty). In a strange reversal, real human writers are ditching it altogether to signal “Hey, I really put time and attention into this,” as much as it might pain them to do so.

• Excessive hedging or overly formal transitions: Spotting phrases like, “It may be prudent…” or “It might appear that…” show the lack of a clear and compelling human POV. Also formal transitions like "As a result" , "In parallel" , "Ultimately" feel mechanical and overused.

• Tripping over triplets: Using a sequence of threes is a historically strong tactic for storytelling. The number three scratches some itch in our human brains, and AI knows that. So if you see a “build, launch, and scale” or a “simple, smart, seamless” you’re likely staring down the barrel of a bot.

• Grandiose descriptors and verbs: These sensationalist adjectives, adverbs, and verbs try to drum up the emotion and persuasive flair, but they usually feel like a reach. I’m talking adjectives like revolutionary, groundbreaking, jaw-dropping, next-level etc. Adverbs like radically, seamlessly, shockingly, infinitely, etc. And verbs like delve, harness, illuminate, unleash, underscore, facilitate, streamline, bolster, embark, leverage, unlock, elevate, foster, map out, etc.

• Rhetorical Questions: These faux-conversational devices pop up when AI tries to sound engaging and interactive. Instead of naturally transitioning between ideas, AI drops in questions like "What's changed?" or "Sound familiar?" to create artificial dialogue.

• Nebulous quotes & sources: If you’re using a quote from someone, triple check that it exists in a credible source outside of your LLM. Sometimes you’ll get quotes that never existed, along with cloudy takes like “Studies show…” or “Experts agree…” when there’s not hard evidence backing it up. A real study showed AI failed to produce accurate citations about 60% of the time, so proceed with extreme caution.
  ​
  ✅ 0 instances ✅
  ​
• “Tapestry.” Just the word “Tapestry.” From personal experience (and the experiences of those on a Reddit thread and Grammarly blog too), it seems like AI can’t get enough of this one. Try to resist the urge to use it, unless your content relates to an actual tapestry (not a metaphorical one).
  ​
  ✅ 0 instances ✅

The examples above will be hard to avoid completely in all of your writing. Lots of the callouts are fairly common, and some are particularly effective (especially before they became a little suspicious). If you list a series of three things or use an em dash, nobody is going to write you off. But if you copy and paste a first-draft 300-word output from your LLM, you’ll probably sound some alarm bells.

Our Recommendation: If you're going to use AI to help draft content, use Claude over ChatGPT. In our experience, it's better at writing human-sounding prose. Use a reasoning model (like Opus 4 or Sonnet 4). These models can loop through instructions to avoid AI tells and make sure they don't show up. Create a "Claude Project" with plenty of examples of your actual writing, in your tone. Use this Project to draft your content. Finally, dictate your perspective (and loads of context) into the LLM before you simply prompt it to "write me a blog post." Your unique POV is a huge component of what makes your content human.

Worth Thousands of Words, Audiovisual No-No’s

When you’re going for realism, the same sloppy shortcuts that give your AI-generated copy away can taint your visual and audio assets. If you’re going for something so outlandish, unrealistic, or very clearly made with AI, people don’t judge the same way. As long as they know you’re not trying to fake them out.

But the moment you try to pass AI content off as an actual photograph, something human-made, or a real interview, it can be interpreted as deceptive and unethical. The backlash can be even stronger in certain industries like fashion, where models and photography are sacred. Last week’s public response to the AI model featured in a Guess ad in Vogue Magazine had people upset to say the least.

Product photography affected by AI can also pose an ethical dilemma. Let’s say you include a reference image of your brand’s purse in your Midjourney or Sora prompt, but the output looks more upscale.

The image looks glossy, immaculate, velvety, detailed, even though it doesn’t quite look like that in real life. Is it still honest to use that image in your ad, knowing AI dolled it up? Or have you crossed the line to misrepresentation?

We're not attorneys, but we'd urge you to research FTC regulations surrounding false advertising and deceptive claims in your industry.

Below are some practices and approaches you should avoid to ensure your audio/visual outputs aren’t seen as deceptive or jarring, especially if you are conveying realism:

• Fake humans pretending to be real ones: Passing off synthetic models or avatars as real people (whether in customer testimonials, interviews, or photoshoots) erodes credibility. Even if you disclose your AI usage or use an outright fake influencer (like the fleeting Mia Zelu of Wimbledon fame), it’s no guarantee people will like it or see it as ethical.

• Uncanny glitches in video: An unnatural movement or jarring facial expression quickly tells viewers they’re looking at AI. If the point of your video hinges on looking real, don’t use AI unless you have total mastery of the output and you disclose your usage of AI. The “Uncanny Valley” theory supports this notion that people can’t stand to look at something almost human. Act accordingly.

• Fake news-style interviews: Creating realistic yet fake footage of someone saying something they never said crosses the line (whether it’s a deepfake of a real person or a made-up character that looks like an everyday person). Unless you’re 100% transparent, or if your account is clearly for comedy/satire purposes only, stay away. This format has proven to be easy to pull off with Google Veo 3.
  ​
• Synthetic voiceovers with no context: AI narration with unnatural or flat delivery can stand out as inauthentic to some listeners (or stylistic to others). Use it to prototype or time your edits, but if you want the best shot at a voiceover that conveys human emotion (especially in high-touch, produced pieces) use real voiceover talent. However, ElevenLabs raised $180M in January at a $3B valuation, so the gap might be bridged pretty quickly if progress stays the course. And of course there are cases where AI voiceover works well.
  ​
• ChatGPT Sepia Filter: That telltale yellow-beige tint that washes over AI-generated images. It's the visual equivalent of the AI em dash. Yellowish images can still get your point across and might be better than nothing, but make sure the visual message is worth the potential credibility hit. A quick "temperature" and "tint" adjustment can fix the issue.

Platforms Prioritizing Quality

Tech platforms are making their stance clear, use AI to elevate, not automate. They want creators and users who embrace this new technology to add value. They’ve been fine-tuning their policies and algorithms to sort for quality over sheer quantity.

YouTube made a change to their monetization policy where “faceless,” repetitive, or mass-produced content (often made possible with AI) will now be deemed inauthentic and ineligible for monetization. And while remixes and AI content are still allowed, creators must add clear value or commentary for it to qualify. The bar might not be particularly high, but at least the bar has been re-set.

Google has also clarified their stance of “Rewarding high-quality content, however it is produced.” Pay special attention to that phrase: “high-quality.” They’re much less concerned with how something was made, and much more concerned with the actual merits of the site or blog. In other words, you can definitely use AI to help write your SEO-boosting blogs, but the final draft better not read like unoriginal, regurgitated goop.

And if you’re like me, you’re probably scrolling through LinkedIn like you’re walking through a minefield, cautiously approaching each post to inspect for clear and present traces of AI before devoting time to reading it.

Bottom Line: Recalibrating Our Filter

Sure, the tools are faster than ever. The slop is sloppier than ever. But we humans are already doing what we’ve always done during these eras of newfound media overload, recalibrating our filters and redefining our threshold for what’s worthy of our time and attention.

So while you can now “write” a blog post by dictating a few sentences into AI and hitting a couple buttons, remember, so can everyone else. The barriers to entry are lower, and because of that, the bar for quality has never been higher.

​Gil Templeton​
Demand Curve Staff Writer