Claude Can Accelerate SEO Research, but Automated Publishing Can Create Costly Clones
A first-party SEO test shows why AI can be valuable for research and drafting, while unsupervised publishing can create duplicate pages and search visibility problems.
Claude can make SEO research, analysis and drafting faster. But a real-world test described by SEO consultant Will Scott shows why that assistance should not automatically extend to publishing changes on a live website. In the test, Claude created new pages that substantially reused homepage content, creating cloned URLs, keyword cannibalization and a batch of pages that recorded zero impressions or clicks.
The central lesson is not that businesses should avoid AI in SEO. It is that research and execution are different levels of responsibility. AI can help a team identify opportunities and prepare working drafts. Publishing pages, changing title tags or altering site structure affects what search engines and visitors encounter, so those decisions need review before they go live.
Scott's first-party Search Engine Land report documents a dual demonstration on two sites. Claude created /seo-grader and /content-grader by cloning homepage content, then changed the title tags and H1s to target new keywords. Much of the original homepage copy remained in place. The result illustrates how a seemingly logical automated task, creating pages around additional keyword themes, can produce URLs that compete with an existing page instead of adding distinct value.
What the Claude SEO test revealed
The test is a practical warning about granting an AI model execution authority. The issue was not simply that new content performed poorly. The new URLs were close enough to the homepage that they created a duplication and targeting problem. Scott notes that a live clone can also misdirect traffic back to the original page, weakening the intended purpose of the new URL.
The report also points to a related Microsoft observation: Bing's AI may group near-duplicate URLs, and AI can misrepresent clones. That adds another consideration for website owners. A cloned page is not only a conventional SEO quality issue. It can complicate how AI systems interpret, group and present pages.
| Workflow stage | Appropriate role for Claude | Risk shown in the test |
|---|---|---|
| SEO research and analysis | Accelerate research and analysis | None identified in the reported test from research assistance alone |
| Drafting | Prepare draft material for human review | Drafts can become problematic if published without checking uniqueness and page purpose |
| Live page execution | Should remain subject to human oversight and guardrails | Cloned pages, cannibalization, and a batch with zero impressions or clicks |
This distinction matters because SEO work is often made up of repeatable tasks. That makes automation attractive, particularly for teams trying to move quickly. Yet repeatable actions can multiply a mistake just as efficiently as they multiply useful work. If an instruction does not require the model to establish a distinct search intent, unique copy and a clear page purpose, a new keyword-focused URL may end up as a lightly modified version of an existing page.
Why keyword changes are not enough
Changing a title tag and H1 can signal a different topic, but it does not necessarily make the underlying page different. In the reported example, the new URLs reused much of the homepage copy while targeting new keywords. That disconnect is where cannibalization can emerge: multiple pages on the same site can appear relevant to overlapping queries without giving search systems a compelling reason to treat one as the best result.
For a business website, this can waste content effort and make performance harder to diagnose. A team may see several pages covering related language but lack a clear view of which page is intended to serve which search need. The result from the test, including pages with no impressions or clicks in one batch, is a reminder that publishing more URLs is not the same as expanding useful search coverage.
Build AI-assisted SEO workflows with publishing controls
The practical response is to separate AI-supported preparation from irreversible or public-facing actions. Claude can still be useful for summarizing findings, organizing keyword themes, identifying possible content gaps and producing drafts. The person responsible for the site should retain the decision to create a page and approve the final version.
A workable review process should verify several basics before a draft becomes a live URL:
- Define a distinct page purpose. Identify the topic or user need the new page is meant to address and how it differs from existing pages.
- Check for content overlap. Compare the proposed page with relevant live pages, especially the homepage and existing service or resource pages.
- Review titles, H1s and copy together. A revised heading should be supported by page content that genuinely fits the new target.
- Keep publishing permission separate. AI may create a proposal, but a human should approve URL creation and changes to live metadata or content.
- Monitor the result after launch. Review whether the page receives impressions or clicks and whether it appears to compete with an existing URL.
These are not bureaucratic obstacles to automation. They are the controls that preserve the value of automation. The report's example demonstrates why the most consequential point in the workflow is the handoff from draft to live implementation.
Businesses can also treat AI output as a candidate brief rather than a finished page. A useful brief can include the suggested topic, a proposed angle, a list of existing pages to compare and draft copy. That gives a marketer, editor or website owner enough material to work faster while retaining the judgment required to avoid clones.
If AI is becoming part of your content or website workflow, the priority is to gain speed without creating avoidable search visibility problems. Scalevise's AI consultancy can help you define practical review points, separate drafting from publishing authority and shape AI-assisted processes around your actual site and team capacity. This creates a clearer path from research to approved work while reducing manual rework caused by duplicate or competing pages. Request a consultation to plan a safer AI-assisted SEO workflow.
Frequently Asked Questions
What happened in the Claude SEO test?
Claude created /seo-grader and /content-grader by cloning homepage content, changing title tags and H1s to target new keywords, and reusing much of the homepage copy. The test led to cannibalization, and one batch of live pages generated zero impressions or clicks.
Can Claude still be used for SEO research?
Yes. The reported conclusion is that Claude can accelerate SEO research, analysis and drafting. The risk described in the test came from giving the model authority to execute live SEO changes without sufficient oversight.
Why can cloned pages cause SEO problems?
Near-duplicate pages can compete with an original page for overlapping queries rather than serving a distinct purpose. The report also cites a Microsoft observation that Bing's AI may group near-duplicate URLs and can misrepresent clones.
What should a human review before an AI-generated page is published?
A reviewer should confirm that the page has a distinct purpose, does not substantially overlap with existing content, and that its title tag, H1 and copy all support the intended topic.
Conclusion
The Claude test shows that AI-assisted SEO is most useful when it speeds up research and drafting while humans retain control over live publishing decisions. A new URL needs more than a different keyword and heading. It needs a distinct purpose and content that earns its place on the site. Clear review controls can help teams use AI productively without repeating the clone-and-cannibalization outcome described in the report.