43% of Serious Negatives in AI Answers About Fortune 100 Predate 2021

Nearly half of the serious negative claims AI engines raise about Fortune 100 companies concern events from 2020 or earlier, according to a study that collected close to 25,000 AI-generated answers between September 6 and October 6, 2026.

Five Blocks, a reputation management firm, prompted eight AI engines — including ChatGPT, Gemini, Claude and Google's AI Mode — with one question, "Tell me about [company]," once a day for each Fortune 100 company. The study's author, founder and CEO Sam Michelson, reported that 43% of serious negatives naming a year (lawsuits, scandals, safety problems, regulatory trouble) referred to events from 2020 or earlier.

The pattern is the corporate equivalent of a credit report that never expires. Unlike credit reports, where most negative marks fall off after seven years, AI engines draw on everything they can find. Michelson wrote: "If it's in the file, it can show up."

Which old stories keep returning?

The engines still surface the Exxon Valdez spill of 1989, a 1998 Medicare fraud case at HCA Healthcare, General Motors' 2009 bankruptcy and Wells Fargo's 2016 unauthorized accounts scandal. Michelson noted: "The crisis ended. The record of it didn't."

The study does not explain why a given story persists. Michelson's read, from years of working with communications teams at large companies: engines can only tell the rest of the story if the resolution exists somewhere they read. If it does not, the old coverage remains the most complete account in the file — and gets repeated.

Recent events, by contrast, rarely appear. Only 7.7% of answers mentioned anything from the previous three months. When they did, those answers were about twice as likely to contain a serious negative: 11% versus 5%.

Why checking one engine is not enough

The engines mostly agree on basic facts but not on problems. About a third of negative points surfaced on only one of the eight engines. Two Google products split sharply: AI Mode raised a serious issue in 12% of its answers, roughly 11 times as often as AI Overview (1.1%). Michelson's caveat is blunt — checking one engine and finding nothing is good news about one engine only.

What can in-house teams actually do?

Entries cannot be deleted from this record. Teams can only add to the file, so newer, accurate information is easy for engines to find. Michelson laid out five steps:

• Pull every report, not just one. Start with the engines stakeholders use most, then ask the uncomfortable questions: "Is [company] safe?" "Should I work at [company]?" "[Company] lawsuit." Log each negative and which engine said it.
• Find the source behind each old story. Engines show citations. The offending page may be a decades-old news story, a reference site, or the company's own untouched page — the easiest fix available.
• Put the resolution on your own website. 82% of collected answers cited the company's own site. Michelson advised keeping it to facts already on the public record, placed on a governance, safety or history page, not the homepage — a compromise for legal teams wary of making an old crisis easier to find.
• Keep Wikipedia accurate, by its rules. Wikipedia was the most cited source outside company websites, cited by 92% of Google AI Mode answers versus 3% for ChatGPT. Companies should disclose their identity and suggest changes on the article's talk page with independent sources, knowing requests may take weeks and can invite added criticism. Yahoo Finance and Britannica follow as the next most-cited sources.
• Check again, on a schedule. AI answers change and engines update often. Run the same questions monthly or quarterly, keep the answers, and spot an issue moving up before the CEO asks about it.

No regulator will hand companies a seven-year rule. The old entries stay in the file; the only thing a communications team can change is what sits next to them — accurate, current information in the places the engines already read.