Key takeaways

  • A “digital ghost” is a retired, rebranded or acquired brand that AI answers keep describing as an active option after the change.
  • In our Q2 2026 Insurance benchmark, captured after Liberty Mutual confirmed Safeco’s retirement on April 25, 2026, Safeco was still mentioned, with a 43% recommendation rate and no citations connecting it to Liberty Mutual.
  • Redirects, schema and knowledge-graph updates fix search; they don’t reliably change what an AI engine says, because models draw on training data and retrieved sources that update on their own schedules.
  • The fix is to check what each engine says about the old brand, follow the paths a confused customer would take, correct the sources you control, and re-test on a schedule.

When an insurer retires a brand, the work usually follows a familiar checklist: redirect the old domain, update the website, notify agents, issue a press release. Search engines pick most of that up. AI answer engines often don’t, at least not on the same timeline.

I described this in “Beware of ‘Digital Ghosts’” for Insurance Thought Leadership in October. This post is the practical companion: how to tell whether a retired brand is still appearing in AI answers, and what you can do about it.

What a digital ghost looks like

A digital ghost is not a hallucination in the usual sense. The brand existed and the facts were once true. The problem is synchronization: the AI answer still treats the old brand as a standalone option, sometimes recommends it, and may not connect it to the company that now owns it.

The example in the original article came from Brainpan’s Q2 2026 Insurance AIVI™ benchmark, which tracked 134 insurance brands across five AI systems. Liberty Mutual confirmed the retirement of the Safeco brand on April 25, 2026. In data collected after that date, Safeco still had real mentions in AI answers, a 43% recommendation rate, and zero citations connecting any of it back to Liberty Mutual.

Why it persists

  • Training data lags. A model’s underlying knowledge reflects what was written about the brand over years, not the last press release.
  • Retrieved sources lag too. Engines that browse before answering pull from comparison sites, directories, reviews and articles. Many of those pages still describe the old brand as it was.
  • Redirects don’t reach the answer. A 301 redirect changes where a visitor lands, not what an engine says. The model does not re-crawl your site each time someone asks a question.
  • Insurance has a lot of brand change. Carrier consolidation, MGA acquisitions and insurtech wind-downs all create retired or merged brands that customers may still ask about.

How to check whether you have one

  1. Ask the engines. Run the questions a customer might ask about the old brand (“Is [old brand] good for home insurance?”, “How do I get a quote from [old brand]?”) on the engines your customers use. Record whether the answer treats it as active, names the new owner, or recommends it.
  2. Check category questions too. Unbranded questions (“best home insurer for new buyers”) show whether the old brand still takes a place in recommendation lists.
  3. Follow the customer’s path. Where does the answer send people? Check the old domain, agent directories, the quote flow and the phone numbers an engine gives out.
  4. Look at the citations. Where an engine cites sources, note which pages still describe the old brand. Those are the pages to correct first.

What you can do

  • Make the transition explicit on pages you control. A clear, dated statement on the new brand’s site that the old brand is now part of it, and what that means for existing customers, gives retrieval-based engines a current source to find.
  • Correct third-party listings. Update directory and profile listings you manage, and request corrections from comparison sites and publishers that still describe the old brand as independent.
  • Work it like any other wrong claim. Capture each answer, verify it against the authoritative source, assign an owner and re-test. Our six-step correction workflow covers the process.
  • Re-test on a schedule. Engines change on their own timelines, so a single clean answer isn’t proof the problem is solved. Track the same questions over several cycles.

None of this guarantees an engine will update its answer by a given date. It does make the correct facts easier to find, and it gives you a record of what changed and when.

What we don’t claim

As in the original article, I’m not putting a dollar figure on digital ghosts. Any such number depends on assumptions about customer behavior that outside data can’t verify. What can be measured is whether the old brand still appears, how it is described, and whether that changes after you act.

Retired or merged a brand recently?

A pilot can include the old brand alongside the new one, so you see how each engine describes both and whether that changes after you act.

Request a pilot proposal AI visibility for insurers

Original article: Kevin Walsh, “Beware of ‘Digital Ghosts’,” Insurance Thought Leadership, October 13, 2026. Safeco figures are from Brainpan’s Q2 2026 Insurance AIVI™ benchmark as reported there.