An issue is contained: one article, one platform comment, one customer complaint that has not picked up amplification. Most issues never become crises and the right response to most of them is monitoring rather than action. A crisis has reached escape velocity in at least one dimension: it is being repeated across credentialed outlets, it is moving on social, the AI engines have absorbed it into their narrative, or stakeholders (investors, regulators, employees, customers) are reaching out. The threshold matters because the operational tempo and the tools are different. An issue is handled by the comms team; a crisis pulls in legal, IR, the board, and reputation specialists working in coordination.

The framing matters. Suppression suggests pushing something down, which is reactive and rarely durable. Elevation is what actually works: building authoritative competing content that the engines have independent reasons to rank, strengthening entity signals so the brand or person is recognized clearly across the Knowledge Graph and Wikidata, securing fresh third-party coverage in the outlets the AI engines weight most, and using legitimate platform channels on the narrow set of cases where takedown is actually available. Done at sufficient sustained volume, elevation moves the contested content off page one and changes the AI narrative durably. The work compounds because each authoritative asset stays authoritative once it ranks. Suppression-led tactics produce the opposite: short-term wins that decay the moment the program stops running.

PR firms run crisis communications: the statement, the press strategy, the executive interviews, the stakeholder calls. We run crisis reputation management: the parallel work to ensure that what stakeholders find when they go look – on Google, in ChatGPT, in the Wikipedia article – actually reflects the message the comms team is putting out, and not the version the contested article wants to tell. The two functions are complementary and work best when they are coordinated from the first call. A statement that does not also exist in citable form on the corporate site rarely makes it into the AI engines. A press placement that the SERP does not show above the contested article does not get seen by the people the comms team was trying to reach.

A useful severity read is structured rather than vibes-based. Media reach scores who has run the story and what tier of outlet they are: a tier-one piece in WSJ or Bloomberg is a different severity than a piece in a trade publication, which is different from a niche newsletter. Social amplification rate measures whether the story is gaining or losing reach hour by hour. AI narrative formation, tracked daily in AIQ, measures how many of the eight engines have absorbed the story and how confidently they are stating it. Stakeholder reach measures the calls and emails coming in from investors, regulators, customers, and employees. The four signals together produce a severity picture that is much more decision-useful than any one of them alone. The same WSJ article can be a manageable issue or a category crisis depending on how the four signals are moving.

Two mechanics make AI crisis amplification distinctive. First, synthesis: an AI engine reading ten articles about an event does not produce ten different summaries; it produces one consolidated narrative that reads as confident fact, and that narrative is what users see. Second, source-set persistence: once the engines have absorbed a story into their training corpus or their retrieval index, the narrative tends to persist even after fresh contradicting information emerges, because the engines weight breadth of citation over recency. The practical effect on a crisis is that the AI version of events can be both worse and more durable than the press version. Daily AIQ monitoring during a crisis catches the formation, identifies the sources the engines are weighting, and points to where source-level intervention will actually move the picture.

Autocomplete reflects searcher behavior in aggregate. When many users are typing a contested phrase, autocomplete shows it; when the volume drops, autocomplete drops it. The implication is that fighting autocomplete directly with content production rarely works – the lever is the underlying news cycle generating the searches, not the search box itself. Google does offer takedowns for specific categories (illegal content, harassment, certain personal information), and we pursue those where they apply. For the more common case of an autocomplete result reflecting a real news event, the practical path is patience plus the broader crisis work: as the news cycle ages and stakeholder attention shifts elsewhere, the search volume drops and the autocomplete softens within weeks to months.

Not every crisis fire needs to be fought, and the wrong ordering wastes the first week. Our triage runs by leverage and durability rather than by emotional intensity. The first priority is page-one Google results for the priority queries, because they are what stakeholders will actually see and they tend to be the most durable layer. The second is AI narrative threads that have appeared across three or more engines, because that is the threshold at which a story consolidates and becomes hard to dislodge later. The third is the Wikipedia article, because Wikipedia framing flows directly into AI engines and Knowledge Panels. The fourth is social platforms where the story is still picking up amplification. The fifth, often left for last because it is what the comms team is naturally drawn to first, is responding to individual outlets, which usually has the least durable effect for the time invested.

Most viral social posts do not become durable reputation problems. The instinct to respond publicly and forcefully often fuels the reach and converts a 72-hour social moment into a multi-week press story. The discipline is to assess the actual reach and trajectory in the first 24 hours, prepare authoritative content that addresses the specific factual claims on owned properties (where stakeholders looking for the brand’s version of events can find it), engage platforms only on clear policy violations rather than on every offensive post, and let the engagement curve do most of the work. If the post does break through to mainstream coverage or starts influencing AI engine responses (tracked daily through AIQ during a live event), the response escalates. If it does not, restraint is the correct strategy. We use a structured monitoring layer across the relevant platforms during active situations.

A crisis engagement starts with the discovery call and a same-week diagnostic: SERP and AI baseline, source map of what is driving the narrative, identification of the highest-leverage interventions, and a written program scope. The first 90 days run at intense cadence – daily monitoring through IMPACT and AIQ, weekly strategy calls, fast content production, source-level work on the articles and Wikipedia framing that are doing the damage. After 90 days, if the trajectory has shifted, the engagement transitions to a sustained 6-12 month durability program: continued monitoring, ongoing authoritative content, peer benchmarking, and the kind of source-layer maintenance that prevents the crisis from resurfacing. The full engagement is documented in a Letter of Engagement at the start; scope changes mid-stream are documented in addenda.

Active litigation collapses the space for public messaging and expands the importance of durable infrastructure. The first rule is that nothing goes public without counsel’s sign-off, including content that seems unrelated, because plaintiffs’ lawyers cite reputation work back in court routinely. The work that remains effective under that constraint is the work that does not require public statements: entity-layer strengthening (Knowledge Graph, Wikidata, schema), authoritative owned content on the company’s broader record, source-level correction requests through editorial channels where the reporting contains factual errors, and Wikipedia work through Talk-page edit requests with disclosed COI. We have run many programs under this constraint, and the substantive volume of work available is much larger than most clients expect going in.