An old version of the Virtual Assistants Agency site was still reachable alongside the new one, with retired prices, a retired phone number and a working checkout for an offer that no longer exists. AI search engines could quote all of it. Here's how it came down in six approved steps, with no page deleted.
The brief said the 2023 site was still published on virtualassistantsagency.com, with retired prices and live checkout buttons. A read-only inventory came first, covering every listed path and every website and funnel in the account. It showed the main domain had mostly been cleaned up already. The real exposure had moved to places nobody was looking.
For a business that helps clients show up in AI search, this mattered twice over. ChatGPT, Perplexity and Bing Copilot can quote any public page, and they had two conflicting versions of the business to choose from.
An AI assistant (Claude) did the work inside the website platform. I set the rules first:
Every path was traced to its website or funnel and tested from outside, recording the first response code and the final page. That's how the live subdomains turned up.
The live checkout on the subdomain was taken down first. The retired products were checked for active subscriptions or payment links, found unused, and later deleted.
The booking pages were unpublished with a 301 to the live booking section. Both subdomains were removed from their funnels and given whole-domain 301 redirects. All 25 old pages on the second website were unpublished.
The empty sitemap now lists only the three live pages: the homepage, the privacy policy and the terms.
Only the keywords were cleared. The title, description, social image and canonical URL stayed exactly as they were, and a live check confirmed it.
robots.txt blocks no crawler, including Bingbot, OAI-SearchBot and GPTBot. The structured data was reviewed. All 16 old URLs were requested again and followed to their final page.
| Item | Before | After | Result |
|---|---|---|---|
| Subdomain checkout funnel | Live order form charging $797 and $400 | Domain removed; whole-domain 301 to the main site | PASS |
| Subdomain promo funnel | Live "40% off" page, retired phone number | Domain removed; whole-domain 301 to the main site | PASS |
| Two old booking pages | Live, retired copy | Unpublished; 301 to the live booking section | PASS |
| Second website (25 pages) | Published | All unpublished, none deleted | PASS |
| Main-domain old pages and checkouts | Already redirected (earlier cleanup) | Re-verified | PASS |
| Sitemap | Empty | Three live pages | PASS |
| Homepage meta keywords | 7 terms, several not on the page | Removed; nothing else changed | PASS |
| robots.txt | Not checked | Blocks no crawler | PASS |
| Retired VA products | Active | Checked for use, then deleted | PASS |
| All 16 old URLs, retested | Old content in four places | Every one returns 301 and ends on the current site | PASS |
The platform refused a redirect on any path a live page still used. The fix stayed inside the platform: its unpublish dialog has a redirect option that produces the same 301.
Removing a funnel's domain and deleting products are guarded steps. The assistant's safety controls stopped it, so I did those clicks myself, and the assistant checked the result.
The brief pointed at pages that were already redirected. The live checkout was on a subdomain nobody mentioned. A read-only pass found it before anything changed.
A 301 keeps every old link and bookmark working and sends it to the current answer. Unpublishing keeps the work recoverable.
The assistant did the clicking and the checking. Every live change waited for my yes, and the most guarded steps were mine.









Old pages, old prices and old phone numbers keep getting quoted by Google and AI assistants until someone finds them. Book a free 30-minute call and we'll look for yours together.