Is AI Content **Really** Detectable?

This little marketing nugget was in a popular newsletter a few weeks back and I feel the need to set the record straight. It is a pet peeve of mine when folks take advantage of busy, hard working creators. Here’s the snippet I’m referring to:

AI tools often embed hidden “markers,” like unusual whitespace or special Unicode characters (e.g., zero-width space), that can be identified both by automated scanners and forensic linguistics methods

Most things are easy to argue with if the answer is at all nuanced. Someone says it is black the other says it is white and arguments ensue. In reality, most of the time it is nuanced. However – in this case? There is very little nuance to be had.

“Current detectors of AI aren’t reliable in practical scenarios,”
said Soheil Feizi, an assistant professor of computer science at UMD.

TL;DR: There are no ‘markers’ hidden in AI produced text-based content.

Why do folks think there are sneaky signals everywhere?

In all the main, publicly available content-creating AI’s, there are no hidden signals that ‘this is AI produced’… There is discussion about whether it is happening, who is doing it, how it is being done, etc. And then there’s the obvious em-dash.

Did you know that em-dashes and the like are known as Statistical Text Watermarks?

Here’s the summary of what I found and how I came to the conclusion that the newsletter cited above, is incorrect. These are the techniques that are used (as of September 2025), where it is used and if it can be removed.

Technique	Where it’s used	How it works	Can it be stripped?
C2PA metadata	Adobe Firefly, DALL·E 3. Content creators can verify their creations in the cloud to create “Content Credentials”.	Tamper-evident metadata package attached to the file	Yes—many social-media sites wipe metadata automatically. It can be recovered if you publish to the Cloud first see: C2PA
SynthID watermark	In some video creation: ie: Google’s Imagen, Veo, Magic Editor	An imperceptible, digital watermark placed directly into some products of Magic Editor (Google) since Feb 2025	The watermark can be removed.
Statistical text watermarks	These are things like the em-dash. Basically a pattern that AI uses (at least currently) that other AI can detect.	Other AI’s can detect and remove these.	Can be removed.
Invisible Unicode characters	Rumors exist that ChatGPT “hides” zero-width spaces in text.	Add imperceptible characters to text.	Doesn’t exist.

The companies like OpenAI and Google are developing these tools to encourage ‘ethical use’ of AI – and I’m sure, to placate angry content creators and artists. And also to maintain trustworthiness among users – if users can tell something is AI-created, then they can give it the trust-weight that it deserves, if any.

The question most of us are asking is, Are we going to get in trouble for using AI on our blogs?

If you’ve been following our advice on WPBarista, then you’ll know that using AI to increase productivity is an amazing gift that can keep you competitive. If you use it to help you maintain your community and keep your trustworthiness front and center than it is nothing more than a tool.

If on the other hand, you use it to spam the internet with fake sites designed to attract clicks and get ad-pay for as long as it isn’t de-indexed… well, that isn’t you.

AI Advice as of September, 2025 for Content Creators

Our advice has been, and continues to be as of today:

1. Open your blog to LLM crawling – do not block bots.
2. Use branding, and lots of mediums for communication within the same page
3. Use proper schema when creating articles, videos, round-ups, sponsored posts, reviews, FAQs and research.
4. Create for the readers with stories – not just facts. Facts can be produced by AI.
5. Use PR and more resources for disseminating your content – AI is looking for sources that are widely distributed.

Ready to level up and learn your SEO and writing secrets? Sign up to the newsletter to get our block classes as they happen!