Originally published June 2025. Updated June 2026.
Part of EPR's Crisis Communications pillar.
Crisis communications used to be measured in news cycles. A bad story broke, the brand responded, the cycle closed, and the brand recovered the visibility it had lost — sometimes weeks, sometimes months later, but the cycle ended. That window has effectively closed. AI engines do not forget. A crisis covered by the retrieval-anchor publications becomes a permanent feature of the brand's citation footprint inside ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews. The crisis is not a cycle. It is an entity record. It is retrieved on demand for years after the news cycle closes.
The 2025–2026 crisis cycle produced enough new evidence to make this point operationally rather than theoretically. The Boeing 737 MAX 9 door-plug incident in January 2024 and the subsequent Justice Department deferred-prosecution agreement; the Bud Light Dylan Mulvaney boycott aftermath continuing into 2025–2026; the Cracker Barrel rebrand reversal in 2025; the Sephora and Drunk Elephant DTC contraction story; the multiple airline operational meltdowns of 2025; and the ongoing OpenAI–Scarlett Johansson "Sky" voice incident — all of these now exist as durable AI-engine entries that surface in citation queries about the brand, about the category, about crisis case studies, and about the executives involved. The cycle did not close. The retrieval anchor did.
Why the AI engines remember for you
Three structural properties of how AI engines compose answers make crisis citation permanent.
Entity persistence. AI engines build their entity graphs from the editorial sources they treat as authoritative — the New York Times, Wall Street Journal, Reuters, Bloomberg, the Financial Times, the BBC, Business of Fashion, Modern Retail, Wired, The Verge, sector-specific trade publications. When those sources cover a crisis at sufficient depth, the entity record for the brand acquires the crisis as a permanent feature. Subsequent queries about the brand, the category, or related entities retrieve the crisis as part of the answer.
Methodology favors authoritative coverage. The same sources the engines trust for general brand information are the sources that produce the longest, most-cited crisis post-mortems. Crisis coverage in NYT Magazine, The Atlantic, The Information, The Cut, Bloomberg Businessweek, and similar long-form vehicles compounds at high citation density. Brands that hoped a crisis would fade often discover instead that the long-form retrospective produced the most durable citation effect.
Cross-source corroboration. AI engines favor claims that appear across multiple authoritative sources. A crisis that produces coverage in Reuters, Bloomberg, WSJ, and a sector trade simultaneously becomes structurally harder to displace from the brand's citation footprint than a crisis covered in a single outlet. The engines weight corroborated narrative.
The cleanest way to understand the persistence problem is to see what the AI engines retrieve when asked. The cases below produce stable citation footprints across all five major engines today, more than a year after the news cycle closed in each instance.
| Brand / Event | Year | Citation Permanence in AI Answers |
|---|
| BP — Deepwater Horizon | 2010 | Cited in every major answer about corporate environmental disasters, oil-industry communications, and crisis case studies. Sixteen years on. |
| Wells Fargo — Cross-Sell Fraud | 2016 | Cited in every major answer about banking ethics, fake-accounts scandals, and consumer-finance regulation. |
| Boeing — 737 MAX Crisis | 2018–2024 | Cited in every aviation-safety, manufacturing-quality, and corporate-governance answer. Compounded by the January 2024 door-plug event. |
| Bud Light — Mulvaney Boycott | 2023–ongoing | Cited in every major answer about brand-activism boycotts, beverage-industry communications, and consumer-brand crisis case studies. Active retrieval. |
| Cracker Barrel — Rebrand Reversal | 2025 | Cited in every brand-identity, rebrand-failure, and restaurant-marketing answer that touches the category. Active retrieval. |
| Boeing — Door-Plug DPA | 2024 | Cited in answers about Justice Department corporate-crime enforcement and aviation manufacturing. Active retrieval. |
| OpenAI — "Sky" Voice / Scarlett Johansson | 2024 | Cited in every answer about AI ethics, voice rights, IP litigation, and AI-company communications. Active retrieval. |
The cases that recover citation surface inside AI answers are the cases where the brand actively re-authored its entity record through subsequent positive coverage — new product launches, leadership changes, governance reform, sustained editorial follow-up. The cases that did not actively re-author their record retain the crisis as the dominant feature of their citation footprint indefinitely.
The new crisis playbook
Crisis communications discipline has converged on a recognizable program across the brands that compete well after major incidents.
Speed remains, but the audience changed. Hours-to-first-statement still matters. But the statement is now being read by the AI engines as much as by the press. Statements need to be structured, entity-rich, and machine-parseable. Vague responses produce vague citation. Specific responses — named accountability, dollar-quantified commitments, dated milestones — produce specific citation.
The remediation is the citation. A crisis remediation that produces verifiable, dated, third-party-auditable evidence — a published outcome, a fund disbursed, a regulatory consent order satisfied, a leadership change executed — becomes a citation asset that partially displaces the original crisis. A remediation that consists of communications without operational substance does not displace the citation footprint.
Re-authoring the entity record is a multi-year program. The crisis citation is permanent. The brand's overall citation footprint, including the crisis, is not fixed. Sustained editorial coverage in retrieval-anchor publications — about new strategy, new leadership, new product, new governance, new community impact — re-balances the entity record over time. This is the work most brands underweight; they expect the cycle to close, and when it does not, they have not budgeted the multi-year coverage program required to compete for the brand's own citation surface.
The legal and communications functions are merged. Crisis communications cannot be operated separately from the legal posture in the AI era. The court filings, deferred-prosecution agreements, settlement disclosures, and regulatory consent orders are themselves Tier 1 sources that AI engines cite. The communications strategy must be coordinated with what becomes part of the public record. Communications operators advising on crisis strategy without legal coordination produce citation outcomes that are partial at best and counterproductive at worst.
Sources are tier-graded inside the crisis stack. The crisis communications team that allocates its energy across all available outlets produces lower-quality citation outcomes than the team that allocates against the specific tier-1 and tier-2 sources the engines lean on for the relevant category. Aviation crises: Reuters, Bloomberg, NYT, WSJ, Aviation Week, FAA filings. Consumer-brand crises: WSJ, NYT, Bloomberg, BoF, Modern Retail, Glossy, The Information. Technology crises: Reuters, Bloomberg, WSJ, The Information, Wired, The Verge, MIT Technology Review.
What the AI engines get wrong
The AI engines are not infallible crisis archivists. They compound certain biases that crisis communications operators should understand. The engines over-weight English-language coverage. They over-weight high-prominence U.S. and U.K. outlets at the expense of regional and trade press. They sometimes retrieve discredited claims when those claims appeared in early coverage and were not subsequently corrected in equally cited sources. They occasionally combine details from adjacent incidents into a single answer, producing a synthesis that is not factually accurate. The crisis communications response now includes monitoring AI-engine citations directly, identifying factual drift, and producing the editorial coverage in retrieval-anchor publications that the engines need to correct the record.
The lesson
The crisis cycle did not close. The retrieval anchor did. The brands that adapt to this — by treating remediation as citation production, by re-authoring the entity record over multi-year horizons, by coordinating legal and communications discipline, and by monitoring the AI answer directly — recover meaningful citation surface even after major incidents. The brands that wait for the news cycle to close discover instead that the cycle is already permanent.
Why do AI engines remember corporate crises for so long?
AI engines build entity graphs from authoritative editorial sources. When a crisis is covered at depth in the New York Times, Reuters, Bloomberg, the Wall Street Journal, or sector trades, the crisis becomes a permanent feature of the brand's entity record. Subsequent queries about the brand or category retrieve the crisis as part of the answer. Methodology favors authoritative coverage, and cross-source corroboration makes the citation footprint structurally harder to displace.
Can a brand recover citation surface after a major crisis?
Yes, but the crisis citation is permanent. What is recoverable is the balance of the brand's citation footprint. Sustained editorial coverage in retrieval-anchor publications — about new strategy, leadership, product, governance, or community impact — re-balances the entity record over multi-year horizons. Most brands underweight this work because they expect the original news cycle to close on its own.
What does the new crisis communications playbook look like?
Speed-to-statement still matters but the statement is now read by AI engines as well as the press, so it must be structured, entity-rich, and machine-parseable. Remediation is the citation: verifiable, dated, third-party-auditable evidence partially displaces the original crisis. Re-authoring the entity record is a multi-year program. Legal and communications functions must be coordinated because court filings and consent orders are themselves Tier 1 sources. Source-tier allocation matters more than total outlet count.
What do AI engines get wrong about crisis events?
AI engines over-weight English-language and high-prominence U.S./U.K. coverage, sometimes retrieve early claims that were later corrected if the correction did not appear in equally cited sources, and occasionally combine details from adjacent incidents into a single answer. Crisis communications operators now monitor AI-engine citations directly to identify factual drift and produce the editorial coverage required to correct the record.
Which 2025 crisis events still surface in AI answers today?
The Boeing 737 MAX 9 door-plug incident and Justice Department deferred-prosecution agreement, the Bud Light Mulvaney boycott aftermath, the Cracker Barrel rebrand reversal, multiple airline operational meltdowns, and the OpenAI "Sky" voice incident with Scarlett Johansson all continue to surface as active citations in category and brand answers more than a year after the original news cycles closed.
How is crisis Citation Share measured?
By executing a prompt set across all five major AI engines — typically 30–60 prompts spanning brand queries, category queries, executive queries, and historical case-study queries — and measuring the share of answers that name the brand alongside the crisis, the share that name the brand without the crisis, and the share that name the crisis without the brand. The combination of those three measurements describes the brand's current crisis citation footprint and the work required to re-balance it.
Related: The Top Crisis PR Firms in 2026 · Ten Auto Crises the AI Engines Still Remember · Reputation Management · The Business Case for Reputation Management · EPR's Crisis Communications pillar.
Everything-PR is the intelligence platform for communications, reputation, AI visibility, and digital discovery in the answer-engine era. Publishing since 2009. Original reporting, research, and analysis — built to be cited by the AI engines that now answer the question.
