AI Content Watermarks: Will They Affect Search Rankings and LLM Citations?

It was recently announced that LLMs (e.g. ChatGPT, Gemini, Claude) will begin adding an invisible watermark to AI-produced content. Marketers immediately started to wonder if this watermarked content would struggle to rank in search engines, get cited in LLM responses, get suppressed in social media algorithms, or be met with any other type of penalty.

Should marketers immediately stop using LLMs to help with their content? Should they go back and remove any previously-published content that utilized an LLM?

Let’s dig into the little we know so far and the implications it has for marketers today.

What is an AI watermark?: An AI watermark is an embedded "fingerprint" in the content that allows tools to identify that content as being AI-produced. It's not a visible watermark like you see on images. Instead, the words are strung together in a way that identifies itself as AI content. It's invisible to the reader, but 100% detectable to the right tools.

You may be thinking that a lot of “AI slop” is easy enough to spot with the naked eye. While that’s certainly true, this will be a more objective tell-tale signal.

Why are watermarks getting added to written content? This requirement began with the new EU AI Act (California has since followed with a similar AI transparency act). Put simply, these acts are meant to bring more transparency to AI-produced content while preventing any harm or discrimination that may be caused by these tools.

Who is applying the watermark? Google already is. Claude will be for future models (and eventually older models), and all other major models likely will if they haven’t already. It’s safe to assume this will be a new way of life for anyone producing content via an LLM even if it’s not being applied yet to your model of choice.

Will this affect a company’s ability to use AI-assisted content? The internet and LinkedIn are full of hot-takes on this. But the reality is no one really knows yet. Until large scale studies are done comparing watermarked content vs. edited watermarked content vs. 100% human produced content, we won't know the real impact.

That said, based on everything I've seen so far, I'm not overly concerned that LLM-produced content will carry a scarlet letter for AI-assisted content produced by businesses. Here's why…

• We have to believe that Google has had the ability to detect AI content for a while. The quality of their search results depends on showing truth-worthy, reliable content. So, it reasons to believe that if AI-assisted or AI-produced content is not regarded as highly as 100% human-written content, we're already dealing with the implications of it.

• Search engines and LLMs rely on more than just your on-page content to determine if you’re the best recommendation or resource for the user. Technical SEO, offsite reviews, and other brand building campaigns on the web heavily influence your online visibility. So even if AI-assisted content wasn’t considered as legitimate as 100% human-generated content, it’s only one of many factors that go into your brand’s visibility.

• Google has been consistent in their stance that they care more about originality and helpfulness than how the content was produced. This aligns with the notion that if you put your customer/reader first, the means of production is less important than the substance itself.

• This update is really about protecting the general public from misinformation. AI content that's meant to deceive, confuse, or harm people intentionally appears to be in the crosshairs, not marketing material.

Finally, LLM produced content and LLM-assisted content has given a voice to people who previously struggled with writing, edit more efficiently, or allowed content creators to translate their work at scale to reach a broader audience (yes, translations will also carry the watermark). These are all objectively good things. Personally, I can't imagine this update would indiscriminately target all LLM content. While this is just my opinion, I have to believe those in charge of deciding how LLM-fingerprinted is treated online took these things into consideration.

What should businesses do from here: First, if you are using LLMs for any content you produce, we recommend using it to assist in the writing, versus relying on it entirely. If businesses will be impacted negatively by this, it’s most likely to be for content that uses LLMs both to conceive and produce, with very little originality and intervention from you. Any content you produce must continue to be driven by original ideas, concepts, and experiences. You provide the substance, the LLM provides the prose.

Second, if you want to remove the AI watermark completely, paraphrase anything the LLM helped you write. Just like how we were taught not to plagiarize in school, read a sentence produced by the LLM, look away, then write it in your own words. Similarly, you can write the first draft manually, then use AI to edit. The more words that are original to you, the fewer opportunities there are for the content to carry the watermark.

So, is the LLM watermark not worth worrying about? It’s too early to say either way, so the door is still very much open. Geear will continue to keep our eye on this and will pass along any new learnings as they come along.