By AJ Coutts, Founder, The Coutts Agency. Last updated: August 31, 2026

A person on our team writes your content. We use AI for outlining, research, proofreading, formatting, and website code. On August 2, 2026, Anthropic began adding an invisible watermark to output from its Claude models. If someone scans your site with an AI detector, some pages may flag. The mark shows Claude handled the text. It doesn’t prove Claude wrote it or change how Google ranks the page.

Key Takeaways

  • Anthropic began marking Claude output on August 2, 2026, as part of its response to Article 50 of the EU AI Act.
  • The mark can travel with copied text. It shows Claude processed the text, not who originally wrote it.
  • Google ranks content based on usefulness and quality. A watermark itself isn’t a ranking penalty.
  • We use AI for planning, research support, proofreading, formatting, code, and some metadata. People write the actual copy.
  • A hand-written article can carry a mark simply because Claude converted the finished piece into HTML or Markdown.

What changed on August 2, 2026?

Anthropic, the company behind Claude, began adding a hidden signal to text produced by newer versions of its models. You can’t see the signal while reading, and it can stay with the text after copying, pasting, or light editing. Images and SVG files use signed metadata instead, which works differently but serves the same basic purpose. The change came from regulation, not a ranking policy. Article 50 of the EU AI Act became enforceable on August 2 and requires AI providers to mark generated output in a machine-readable way.

Anthropic chose to apply its marking system broadly instead of limiting it to Europe. OpenAI and Google have made similar commitments around AI transparency. The headlines have made this sound more dramatic than it is. Anthropic itself says a detected mark is “not fully conclusive.” Its documentation specifically lists proofreading, translation, summarizing, and file conversion as cases where Claude may have touched text without being its original author. Very short passages may not contain enough text for the signal to register.

Does the watermark hurt my search rankings?

No. Google has been clear about this for years. Danny Sullivan, Google’s Search Liaison, summed up the company’s position in 2023:

“We focus on the quality of content, not how content is produced.”

Google’s guidance on AI-generated content says the same thing. Google acts against low-value pages produced at scale to manipulate search results. It can take action against that kind of content whether it came from a person, an AI tool, or a mix of both. There’s also a practical reason the watermark would make a poor ranking signal. It can’t tell the difference between an article Claude drafted and an article Claude simply spell-checked or reformatted. That signal tells a search engine very little about page quality.

The short version is simple: rankings, AI citations, and brand mentions still depend on useful information, clear structure, real expertise, and a website people trust. For clients, that’s a distinction that needs mentioning because a detector can report AI involvement without telling you whether AI wrote, edited, formatted, or simply handled the text before publication on your site. For us, we’ve prided ourselves on being great marketers long before AI became a prevalent asset in our industry. And our stance has always been one of humans first. It is the very essence of this industry, from our team, to you, to your customers. Nothing works without humans at the center of our business.

Why would our pages carry a mark if a person wrote them?

Because the mark records handling, not authorship. Think of it more like a shipping label than a signature. It tells you something passed through a system, but it does not tell you who made what was inside.

Here is how it happens. Your writer drafts the piece, filters using AI for strict adherence to pre-determined formatting and SEO/AEO readability, an editor reviews it, facts are checked, and then you approve the final copy. Then someone asks Claude to convert the approved text into clean HTML or Markdown for the website CMS.

Claude returns the same copy in a different format, and that handling can leave the mark. Run a detector on the finished page and you may get a hit. The detector sees something real, but it says almost nothing about authorship. We would rather explain that now than have a client run a scan three months from now and assume the worst.

One of our agency’s core values is: People First. We put humans at the center of everything we do, valuing the lives and well-being of our team, respecting our partners, and crafting marketing that connects with real people, not just algorithms.

Where we actually use AI

Specifics matter more than a vague promise, so here is where AI fits into our workflow.

Outlining and ideation. Before anyone writes, we decide what questions the piece has to answer, what competing pages already cover, and how the article should be formatted and structured. These have been standard best practices for SEO long before AEO & AI came onto the scene. AI can then test the outline, find gaps, and challenge the order before the writing even starts. This positions our team to produce the most relevant content for your brand in very specific and technical ways that would’ve taken much longer to accomplish before AI conducted these scans.

Research support. We use it to surface common questions, organize search terms and keywords, and summarize long reports so the writer can decide what deserves a full read. Any claim that makes it into published copy still gets checked against the original source.

Formatting and proofreading. We use AI to convert approved copy into HTML or Markdown, catch broken links, spot duplicated words, and find small formatting issues. We also use it to create specific language that makes each piece of content easier to read from a search query perspective in Google, AI search, and other engines. This is the most common reason human-written content may carry an AI mark.

Code. Schema markup, tracking snippets, redirect maps, and small template fixes are all areas where AI can save time. A person still reviews and tests anything before it goes live on your site.

Meta titles and descriptions. We sometimes use AI to create a first pass, then our team rewrites and approves the final version. These are usually too short to trigger detection with only 50-160 characters, but they are still part of the workflow.

Where we don’t use AI

The writing itself. Your articles, product pages, landing pages, and emails are written by a person who understands your brand and has reviewed the history of your account.

We also don’t publish machine-produced copy without extensive human review, and we don’t create content at scale simply to fill a publishing calendar. Those are the habits Google’s own guidance warns about, and they are also poor marketing.

The reason is a very practical one. A model can produce a general page about your product, but it can’t explain why your engineers machine a part a certain way, what your team learned in the field last season, or which customer question comes up during every sales call. That knowledge lives inside your company, and it makes a page worth trusting and citing. We explain that approach in What Google’s July 2026 Search Change Means for Firearms & Outdoor Brands.

What this means for you, practically

Three things matter most with this topic of discussion. If you scan your site and see AI flags, compare the page against the workflow above before drawing a conclusion. Formatting, proofreading, and schema can all cause human-written content to show signs of AI handling. If you want details on a specific page, ask us which steps involve AI.

If a customer, partner, or internal team member asks whether your content is AI-written, you can answer honestly that it’s not and explain that the mark may reflect a formatting or processing step. If useful, we can document that process in writing. And if you would rather keep AI out of your content workflow entirely, tell us. Formatting everything by hand takes longer, but that is the tradeoff, and we can do it if you want us to. Our dedication is to the humans involved in this process, not a learning model or algorithm. We want the people in our business to come first, because you and your customers are who we care about most.

Frequently Asked Questions

Can I tell which parts of a page were AI-generated?

No, and neither can anyone else with certainty. The mark is a signal spread across a passage, not a sentence-by-sentence authorship label. It may show that Claude processed text somewhere in that block, but it doesn’t reveal how much originated with the model or which sentences were touched. Anthropic says this directly in their documentation. A detector that gives you a percentage is estimating, not measuring authorship.

Will the watermark stop my content from being cited in AI answers?

There is no evidence that it will, and no major search or answer platform has announced plans to use watermarks as a citation filter. ChatGPT, Perplexity, and Google’s AI Overviews select sources based on criteria like relevance, clarity, and authority. Your citation performance still depends on whether the page answers the question well and gives the system a source worth referencing.

Does this apply to content published before August 2026?

Possibly. Older model versions are being updated with the marking system, so older pages may or may not carry it depending on when and how the text was processed. Either way, the underlying content didn’t become weaker because a labeling rule changed. If a page ranked well before August, the presence of a mark doesn’t suddenly change the quality of that page.

Should we add a disclosure to pages where AI was involved?

For most firearm and outdoor brands, not by default. The EU requirement discussed here applies to AI providers such as Anthropic rather than serving as a blanket disclosure rule for every business using AI in a marketing workflow. If your company sells into the EU or works with a partner that has its own disclosure requirements, talk with your legal team and let us know so we can build those needs into the process.

How do you keep AI out of the writing if the tools are right there?

Mostly through structure. A writer starts with an approved brief that identifies the angle, the sources, the customer questions, and the firsthand information the article needs. Every finished piece of content is initially reviewed for technical adherence to the outlined criteria of the initial content brief before being reviewed and edited for substance by humans on our team. Before an article gets sent to you for your review, it’s already been checked by multiple team members. That assignment can’t be effectively completed by asking AI for a generic draft because the strongest material isn’t sitting on the public web.

Can I see this policy applied to my own account?

Yes. Ask us about any piece we’ve delivered, and we’ll walk you through the content brief, the writing and review process, as well as the steps where AI was used. We keep that workflow visible because there’s no reason to make it mysterious. Your monthly growth report can also point back to this page, so your team has the same explanation in one place.

Related guides

Have a question about how a specific page of yours was produced? Ask your Coutts account lead for the content trail.