If your content strategy depends on publishing AI articles at scale, August gave you a reason to reconsider it. Google completed another spam update, with subsequent reports pointing to losses among sites publishing mass-generated AI content for search rankings. Meanwhile, AI providers have been setting out how their tools will watermark generated content. These developments deserve a place in your next marketing conversation.
I think the mistake would be to treat this as a problem you can fix by making the writing sound slightly more human. You need to look at what you’re publishing, why it exists, and what it contributes.
Google’s August update has already happened
Google’s August 2026 spam update rolled out between 18 and 21 August. Reports afterwards described ranking losses among sites using AI to mass-produce content for SEO.
Search Engine Journal’s coverage points to automated publishing as a possible focus, while acknowledging that the evidence is observational. Some publishers reported losses; others said their AI-generated content improved. Google hasn’t confirmed that watermark detection drove the update. The reporting makes those limits clear.
Those qualifications matter, but they aren’t a reason to ignore what happened. If you’re paying for a growing library of articles built around keyword variations, this is already a commercial risk worth examining.
The question I would ask your agency is straightforward. What does each of those pages give your prospective customer that they couldn’t get from the other pages covering the same subject?
A different introduction won’t be a particularly convincing answer.
What AI watermarking actually means
An AI watermark is a signal embedded in generated content that a compatible detection system can look for. It doesn’t have to be a visible stamp across an image. In text, it can be a statistical pattern in the words themselves.
Anthropic explained its approach on 14 August. Claude’s text watermark uses patterns in word choices, based on Google DeepMind’s SynthID-Text method. There are no hidden characters you can remove from a WordPress draft.
The watermark indicates likely Claude involvement. It doesn’t identify the person using it or establish whether Claude wrote the original draft or substantially edited someone else’s work. Detection also has limitations, particularly with short passages and light editing.
Anthropic describes a global rollout, with older models being brought into coverage over time. Its September update also describes detection access initially limited to eligible organisations. Its explanation is specific about what the technology can establish.
For you, the implication is simple. A publishing process built around nobody noticing AI involvement is becoming a poor bet.
OpenAI and Google are moving in the same direction
The providers haven’t all introduced identical systems.
OpenAI’s current documentation describes SynthID watermarking for supported images and audio, with Content Credentials also attached to supported images. Those credentials record information about a file’s origin and are separate from an embedded watermark. Extending provenance signals to text remains a stated goal, rather than confirmation that every ChatGPT response is already watermarked. OpenAI sets out that coverage here.
Google signed the EU’s transparency code on 24 July, ahead of the August deadline. Its work builds on existing SynthID watermarking technology and Content Credentials, including partnerships with other providers. The technology predates August; the regulatory deadline gives its adoption more immediate significance. Google’s announcement explains that commitment.
Switching between tools won’t give you a durable content strategy. I would assume watermarking and detection will continue developing and make publishing decisions I could defend regardless.
What the EU rules mean for your website
The EU AI Act’s Article 50 transparency obligations began applying on 2 August 2026. They include requirements for generative AI providers to make their output machine-readable and detectable as artificially generated or manipulated. Watermarking is one way providers are implementing those requirements.
Systems already on the market before August have a limited transition period until 2 December for those marking and detection obligations.
That doesn’t mean every AI-assisted article on your website automatically needs a visible warning. Provider marking requirements and publisher disclosure requirements are different. Disclosure rules cover deepfakes and certain AI-generated text on matters of public interest. For that text, substantive human review or editorial control, with editorial responsibility, can provide an exemption. The European Commission explains those distinctions.
Your location also won’t necessarily insulate your workflow from changes. Anthropic’s global approach means a UK user can encounter watermarking introduced in response to EU requirements.
I would make sure someone owns the publishing decision. “The tool produced it” explains the process. It doesn’t account for the accuracy of what you publish.
Watermarking and ranking losses need a careful connection
The spam update and watermarking developments happened close together. That makes their relationship worth discussing. It doesn’t prove Google detected a Claude watermark and reduced a page’s rankings because of it.
I wouldn’t build a marketing recommendation around that unsupported connection.
The defensible concern is substantial enough. Publishers are reporting losses associated with mass-generated SEO content while providers are making AI involvement more identifiable. If your strategy relies on automated volume and offers little original value, both developments should make you uncomfortable.
Trying to remove a watermark leaves the underlying weakness untouched. You still have pages competing to say the same things.
AEO won’t rescue interchangeable content
Answer engine optimisation, or AEO, is the work of helping your company and content appear in answers generated by services such as ChatGPT and Google’s AI search features.
I would be wary of anyone proposing that you solve an SEO problem by commissioning more of the same material under an AEO label.
An answer engine can already produce a general explanation of most services. Your opportunity is to contribute something specific enough to make your company worth referencing.
That might be a documented comparison, an explanation of a difficult buying decision, or evidence from work you’ve actually done. It could be an honest account of when your service is unsuitable. The qualifications often tell a prospective customer more than the headline claim.
My expectation is that verifiable expertise will become more valuable as generated material becomes easier to produce and identify. That is a strategic judgement, rather than a confirmed rule about how every answer engine uses watermarks or selects its sources.
It is also a more useful investment than finding another way to rephrase information everyone already has.
What I would ask your marketing team to do
Start with performance. Compare organic traffic and rankings before the August rollout with the period after its completion. Look for patterns across groups of pages, particularly those produced through the same automated process. Account for other changes before attributing a decline to the update.
Then inspect the content itself. Choose the pages that matter commercially and ask:
- What customer decision does this page help with?
- What original knowledge or evidence does it contain?
- Who checked the substance before publication?
- Does it attract suitable enquiries or support sales conversations?
You don’t need to remove every page that involved AI. You do need to know whether your investment has produced useful assets or an expensive collection of summaries.
For new work, bring expertise into the brief before generating the draft. Get your delivery team’s explanation of the trade-offs. Capture the questions your sales team keeps answering. Verify the examples and claims.
AI can help organise that material. The substance needs an accountable source.
Take your next content review beyond how many articles were delivered. Ask what those articles add, what they achieve, and which ones your company would confidently stand behind.
That answer matters more than whether the writing carries an AI watermark.
