Red flags everywhere: why AI detectors keep getting it wrong

The AI Paradox

Just when you think you’ve reached peak AI paradox, another puzzler appears out of the ether. This time, we’re talking about the simmering tension between content authenticity and the potential for false assurances from AI detection tools.  

A successful content strategy relies on effective and trusting collaboration with your writers. 

While AI detection tools might have once provided reassurance, their flawed results have ushered in a world of confusion, complicating processes and frustrating content creators. Here's why it's time to rethink what “authentic content” really means.

How AI detectors work (and don’t)

AI detectors use ML models to identify content that isn’t human-created. They look at how something is written; the patterns, structures and so forth. Not what it is actually saying; the meaning and intent.

As with the AI content generators, they’re trained on dizzying amounts of data from humans and AIs, which helps them detect linguistic patterns.

These include: 

‍

1. Burstiness

We humans tend to write in different styles, or ‘bursts’, which vary in length, structure and complexity. AI-generated text is often constructed in uniform (read: dull) patterns that lack character and nuance.

Detectors use this to guess a text’s origin. Usually, quality content writers smooth out these bursts, creating a more cohesive flow. Ironically, this can make it more likely to be flagged by AI detection tools. 

‍

2. Perplexity

Perplexity indicates how predictable a text is. Lower perplexity = more predictable = more likely to be AI-generated. Human writing is inherently more varied than AI, so AI tools evaluate how "surprising" each word or phrase is.

The problem is, if you’re writing well (clearly and concisely), your content might score too low on perplexity and get flagged unfairly.

‍

3. Linguistic patterning

This refers to structural and stylistic fingerprints such as how sentences are formed, how often words appear, the use of idioms, etc. When training detection tools on these deep wells of internet text, they’ll flag familiar patterns, especially in marketing copy shaped by templates or Grammarly-esque tools.

So, if you’ve used a content marketing tool to up your user experience game, your copy might be regarded as AI-generated by the detection tools. 

‍

In a nutshell, AI detection tools can analyse:

• Sentence structure and syntax consistency
• Predictability of word usage (perplexity)
• Use of filler phrases or overused transitions (does “navigating the challenges of today’s competitive business landscape” sound familiar?) 
• Surface-level text features like tone, fluency and repetition

‍

They can’t analyse:

• Original insight or first-hand experience
• Author’s intent, reasoning or critical thinking
• Factual accuracy
• Research depth
• Creative framing or brand voice nuances

Need more info? Grammarly’s blog on how AI detectors work has you covered. 
‍

Why AI detectors get it wrong

Sara Stella Lattanzio, Head of Marketing at Stryber wrote on LinkedIn,

Can we just spare ourselves the time and only check for plagiarism like in good old days? As for the rest, using our best judgement to decide if it’s well written or not should suffice.

AI’s tendency towards bias has been well-reported. Here are a few other reasons why AI detectors get it wrong: 

• False positives from overly polished or formulaic content 
• Reliance on statistical scoring models
• Lack of scoring consistency
• Inbuilt biases toward simple vs. nuanced language
• Changes/improvements in AI text generators

The tool glitches are real, as strategist Shari Berg reveals:

If I cut and paste my entire article into Copyleaks, it says 100% AI. But paste it in smaller chunks? 100% human. Every. Single. Time.

And as Andrew Holland, Director of SEO, recently showed,
‍

So instead of asking 'Was this written by AI?' maybe it's time to start asking: 'Is this any good?'

Need to produce top-tier content at scale but afraid of missing a vital component? Take a peek at our Content Quality Checklist.
‍

Ripple effects on the content ecosystem

This over-reliance can even lead us astray from trusting our instincts—a trap that many of us, including those with years of experience, are falling into.

As Dave Gerhardt reflected in a recent post,

I’ve been relying too much on AI to give me the answers vs. relying on my gut, my creativity, my weird ideas.

In a sea of generated content, human connection—and good old-fashioned discernment—is more critical than ever.
‍

So, how does the AI paradox impact your content marketing strategy? 

• Clients feel unsure of whom or what to trust
• Editors stall production or second-guess decisions
• Freelancers feel demoralised or write more cautiously
• Process bottlenecks occur, resulting in unnecessary rework

‍

The real issue isn’t the AI tools.

It’s misplaced trust in the wrong parts of the process, as our Head of CSM, Fauve, explains:

Olivier, our CPO, echoes this sentiment with a warning about over-reliance on flawed tech: 

There are two main ways to spot AI-written text. One uses math to analyse patterns—it's not perfect, but at least it's clear. The other just asks ChatGPT, which often agrees with guesses, even without proof. At Contentoo, we trust context. Good tools can help, but ChatGPT’s guesswork isn’t reliable.

Getting the balance between AI tools and human intervention is key. Find out more about how to effectively leverage AI as part of a balanced content diet. 

‍

What editors should know

AI detection tools are merely flagging an issue. They’re not a final judgment. Instead of rejecting content, talk to your writer(s) about their process and if/how they use AI tools.

Great content creators can offer significant long-term value to your business and your team. So, why not build trust by asking the right questions and understanding their workflows?

As Thomas, our Head of Marketing, reasons, “If it performs well, does what it’s supposed to do and AI detection tools flag it, does it matter?”

Looking for freelancers you can trust? Work with our vetted content experts at Contentoo

‍

What can writers do to avoid false flags?

Our top tip for content writers is to retain your voice.

Make sure you include plenty of colour, first-hand insights, specifics and real-life examples. Even if the copy brief is on the formulaic side, try to use unique structures and incorporate original perspectives. Add a little dash of the unexpected.

Include a compelling story or an appropriate joke. Ditch those generic intros and cliches and, when in doubt, read your content out loud. Any clangers or boilerplate bores will stick out like a sore thumb. 

Need more tips? Discover eight unique ways to use AI for content

‍

Our role in ensuring quality 

Our role at Contentoo means we have responsibilities to both our clients and freelancers when it comes to ensuring quality and creating the best content partnerships possible. We work closely with both parties throughout the relationship.

Here’s how:

• We handpick and thoroughly vet freelancers to identify those who are a strong match for your brand and your objectives
• We provide editorial insight and support whenever it’s needed
• We’ve developed and implemented content briefing and QA processes that provide clarity for all stakeholders and help ensure smooth workflows and successful outcomes

The final word: keep it human 

Remember the Voight-Kampff test, so memorably used in Blade Runner to work out who’s human and who’s a replicant? Turns out it was flawed. Was Harrison Ford’s Deckard a replicant? We’ll never know. Is AI the enemy? That’s an unequivocal ‘nope’. 

AI can be your friend, and the best detection tools deserve their spot in your content arsenal, though not at the expense of valuable, creative work or when building that all-important trust. Content that’s imbued with human insight, an authentic voice and plenty of originality gives it power and makes it stick. 

The right tools can be super-useful, provided you set the appropriate boundaries, as Penny argues. “AI isn’t the disruptor anymore, it’s the baseline. Whether you like it or not, it’s already baked into the stack. The real differentiator is how well you use it without losing the human edge that makes content work.”

Want to create content that balances quality, originality and trust, without second-guessing your freelancers? Talk one of our strategists today.