If you’ve used an AI tool to help draft a service page, an FAQ or a blog post, here’s something new to know. Since August, Anthropic has been adding a hidden watermark to text from its newer Claude models, and it plans to cover its older models by December 2. Other AI providers that offer their tools in the EU are under the same new marking rules.
So the obvious question for a business owner: can Google see that mark, and will it hold your page back?
Here’s what the official sources actually say.
What the watermark is
Anthropic explained it in an August 14 post. Every time Claude writes, it’s choosing between several reasonable next words. The watermark nudges those small choices into a pattern you can’t see when you read the text, but that Anthropic can pick up with a key only it holds. Anthropic says the method builds on SynthID-Text, a technique from Google DeepMind.
A few details matter if you publish AI-assisted copy:
- According to Anthropic’s help article, Claude models launched on or after August 2, 2026 carry the mark from launch, and Anthropic is adding it to earlier models, with all of them covered by December 2, 2026. It applies wherever Claude is used, including through the API.
- The mark travels with the text when you copy and paste it. Anthropic says light editing probably won’t remove it completely. Heavy editing, paraphrasing or translation can leave it undetectable, and a full rewrite where every word changes will remove it.
- Checking for it takes Anthropic’s detection tool, which is in private preview for groups like regulators, fact-checkers, researchers and companies with their own legal duty to verify. Anthropic says it plans to open access over time.
Why it’s happening now
The push comes from Europe. Article 50 of the EU AI Act started applying on August 2, 2026. It requires companies that provide generative AI systems to add machine-readable marks so AI-generated content can be detected. Anthropic signed the EU’s code of practice on transparency for AI-generated content in July to meet that duty.
So the watermark exists to answer one question: was a model involved? It says nothing about whether a page is accurate, useful or worth reading.
What Google has said about AI content
Google’s Search documentation doesn’t tell site owners to stay away from AI. Its guidance on generative AI content puts the line at value. Using AI to churn out many pages without adding anything for users may break its spam policy on scaled content abuse.
That spam policy, last updated August 28, defines the problem by purpose. Scaled content abuse means lots of pages made mainly to game rankings instead of helping people, and Google says that applies no matter how the content was created. Using AI to generate many pages that add nothing sits on its example list, next to scraping, stitching other pages together and pages stuffed with keywords that barely make sense.
Google’s guide to helpful content asks creators to think about who made a page, how and why. On the “how”, it suggests explaining how AI or automation was used when a reader might reasonably wonder how the content was made.
None of the Google Search documentation we reviewed says watermark detection plays any part in ranking. That’s the absence of a statement, not a promise from Google, so keep it in proportion. Still, everything Google has published points one way: the page gets judged on what it gives the reader.
One timing note. Google started rolling out its September 2026 spam update on September 24 and says it may take up to two weeks. If your rankings move this week, let the update finish before you read too much into it.
Where your effort actually pays off
Nothing Google has published suggests that scrubbing a watermark out of your pages helps in search. The better use of that hour is making each page say something only your business could say.
Pull up the five pages that bring in the most calls or form fills and ask:
- Does the page name the real service area, the real process and what happens after someone reaches out?
- Is there anything on it a competitor couldn’t copy? Your own photos, the price ranges you actually quote, warranty terms, license details, the questions customers really ask on the phone.
- Did someone who does the work check every claim, number and promise?
- If you have a page for every town you serve, does each one carry real local detail, or is it the same page with the city name swapped?
That last question matters most. Google’s spam policies name this pattern directly: pages aimed at specific cities or regions that funnel people to one page count as doorway abuse, and mass-produced pages with little value can count as scaled content abuse.
If AI helped you write something, there’s nothing to hide. Where a reader could fairly ask how a page was made, a short note along the lines Google suggests is fine. On most service pages, though, the reader mostly wants to know whether you cover their area and when you can come out.
That’s also where Glory’s landing page work starts: pages on your own domain, built around the services you actually offer and connected to your CRM, so the page and what happens after the form line up.
So use AI for the first draft if it saves you time. Just treat it as a first draft. The finished page should carry the details you’d give a customer across the counter, checked by the person who does the job.
Source note: Facts about the watermark come from Anthropic’s August 14, 2026 announcement and its Claude Help Center article. The EU rule is from the European Commission’s Article 50 guidelines. Google’s position comes from its Search Central documentation and Search Status Dashboard. The page checklist is Glory Digital Solutions’ own guidance.