Abandoned Sites Are Teaching AI Search the Wrong Facts
The answer came from a site you forgot
A prospective student asks an AI assistant about application deadlines. The answer cites a program microsite from four years ago. A resident asks about trash pickup and gets the schedule from a pilot program site that ended. Nobody in web services knew either site was still online.
Answer engines do not know which of your sites you consider current. They read what is on your domain and summarize it.
Why sprawl hurts answers
- Conflicting facts. Two sites on your domain give two different dates, and the engine picks one.
- Old authority. An older site with links can look more authoritative than the new page.
- No owner to fix it. When nobody owns a site, nobody notices it is wrong.
Regulatory dates make this worse. When the DOJ extended the Title II web dates in April 2026, every page with the old dates became wrong at once. Source: the DOJ interim final rule extending the Title II compliance dates (91 FR 20902). Sites nobody maintains still show the old ones.
Clean up the estate, then the answers
Find every site
You cannot correct a site you do not know exists. Start with a full inventory of your domains.
Retire what should not be live
Expired programs and finished projects should be archived and redirected to a current page, or removed.
Fix facts at the source
Keep one approved version of key facts, such as dates, program names, and contacts, and make every site match it.
Write quotable passages
On the pages that should answer, put a short, clear definition near the top and real FAQ entries. Mark up FAQs with the schema.org FAQPage type when the visible text matches.
Where Accorbis fits
Accorbis is a standalone web governance platform. It gives one dashboard across the WordPress, Drupal, and custom sites an organization already runs. It is not a CMS and not a hosting company.
Discovery finds the sites on your domains, and the dashboard shows each site’s owner, platform, status, and update status, so neglected properties surface. It does not rewrite content. Your teams decide what to fix, merge, or retire.
A hypothetical: two pages, two dates
Imagine a public university with an accessibility page that was updated after the April 2026 rule and an older department site that still says April 2026. Both are on the university’s domain. An assistant asked about the university’s accessibility deadline might cite either. The current page is correct, but the old one states its date in a single clear sentence near the top, which makes it easy to quote.
The fix is to find the old page, correct or retire it, and make the current page just as clear.
A cleanup checklist
- List the questions people ask about you most often.
- Ask several AI assistants those questions and record the sources they cite.
- Find every page on your domains that answers each question.
- Mark which page should own each answer.
- Update, redirect, or retire the others.
- Add short, clear answers near the top of the owning pages.
- Repeat the assistant check after cleanup.
Decision criteria for stale sites
- Wrong facts and an active audience: correct it now and add a dated update note.
- Duplicated by a current page: redirect to that page.
- Wrong facts, no audience, no owner: retire it.
- Correct but unmaintained: assign an owner or plan to merge it.
Who should own this work
Cleanup sits between teams. The web or digital team owns the inventory and the decision to retire. Content owners own the facts. Communications often owns the key questions. Name one person to coordinate, and give them a short list of approved facts to check pages against.
Common mistakes
- Fixing the main site only. The correct page exists, but older sites on the same domain still contradict it.
- Blocking instead of fixing. Hiding pages from crawlers leaves wrong information in front of human visitors.
- Deleting without redirects. Links and bookmarks break, and visitors land nowhere.
- One-time cleanup. Without an ongoing inventory, new stale sites appear.
Frequently asked questions
Can we stop AI tools from reading old sites?
You can block crawlers, but the better fix is to retire or correct the content so every reader gets the right answer.
Does llms.txt fix this?
The llms.txt proposal can point models to your best pages. It does not remove wrong information elsewhere on your domain.
How do we know what AI tools say about us?
Ask several assistants your top questions and note the sources they cite. Repeat after cleanup.
Is this an SEO project or a governance project?
Both. Governance finds and retires the sites. Content work fixes the answers.
How often should we check what assistants say about us?
Quarterly is a reasonable start, and after any change to key facts such as dates, programs, or contacts.
Does retiring a site hurt search visibility?
Redirecting retired pages to the closest current page usually keeps visitors and links pointed somewhere useful.
What facts should we standardize first?
The ones people ask about most and that change: deadlines, program names, contact details, and policy dates.
Does Accorbis tell us which facts are wrong?
No. It shows which sites exist, who owns them, and their status. Your content owners decide whether the facts on each site are current.
Next step
Schedule a 30-minute working session and see what is still live on your domains.
