AI search engines can surface serious corporate problems years after they happened, according to a study of nearly 25,000 answers about Fortune 100 companies. The research also suggests companies cannot assume one AI tool will show the same problems as another, or that an old story will come with its resolution.
Watch Desk analysis
What happened
Sam Michelson, founder and CEO of Five Blocks, describes the study in PR Daily. His team asked eight AI engines, including ChatGPT, Gemini, Claude and Google AI Mode, “Tell me about [company]” once a day for each Fortune 100 company between 6 September and 6 October 2026.
Michelson says 43% of serious negative answers that named a year referred to events from 2020 or earlier. Only 7.7% mentioned anything from the previous three months. The study also found that about a third of negative points appeared in only one of the eight engines. These are findings from the company’s own study, as described by its CEO, not an independent audit of AI search results.
Key findings
- Old stories linger
43% of serious negatives that named a year concerned events from 2020 or earlier. - Recent issues were less common
Only 7.7% of answers mentioned events from the previous three months. - The engines disagree
About a third of negative points appeared in just one of the eight systems. - Company websites featured heavily
82% of the answers cited the company’s own site, according to the study.
Why it matters
A company’s past can remain part of what AI systems say about it, even after a crisis is over. Michelson argues that if reliable information about a resolution is hard to find, older coverage may remain the fullest account available to an engine. The study does not establish why individual stories recur, so that explanation should be treated as his interpretation, not a demonstrated cause.
There is a practical takeaway for communications teams: test more than one system, and check what sources each answer cites. The report recommends publishing clear, factual updates about resolutions on relevant company pages, and checking answers regularly. For Wikipedia, it advises companies to follow the site’s rules, disclose their connection and suggest changes on the article’s talk page rather than editing it themselves.
Our read
This is a useful reminder that an AI answer is not a tidy corporate archive with a seven-year expiry date. The study’s results are worth treating as a prompt to investigate, not a forecast of what every user will see. If you work in communications, start by asking several tools the same question and checking the citations before drafting a plan to “fix” anything.
What to watch
- Whether independent research finds similar patterns across a wider range of companies and prompts.
- How much results change between engines, and over time.
- Whether organisations publish clear, sourced updates that give engines relevant context.
Discussion spark: Should companies actively publish context about resolved crises for AI systems to find, or does that risk turning corporate websites into reputation-management material?
Sources and evidence
- Source update (11 November 2026, 11:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.