Originally published: April 25, 2012 · Updated: June 16, 2026

The original April 2012 post argued that Goldman Sachs was running an institutional crisis it was structurally not equipped to handle. The triggers in that window: Greg Smith's March 14, 2012 New York Times op-ed announcing his resignation and accusing the firm of a toxic client-first-in-name-only culture. A fresh federal insider trading probe against a then-current Goldman employee. The May 21, 2012 trial of former Goldman board director Rajat Gupta, who would be convicted that June and sentenced to two years in federal prison. The line that anchored my analysis at the time came from Lloyd Blankfein himself, on May 21, 2010: the firm had managed risk well and reputation badly. Two years later, the same problem was metastasizing.

Fourteen years on, the Goldman case study reads as the cleanest pre-AI demonstration of a structural mismatch between legal risk and reputational risk. The 2026 version of the same problem is harder, faster, and permanent.

What the 2012 analysis got right

Three calls held.

The court-of-law versus court-of-public-opinion split. The 2012 piece argued that Goldman could not let the slow, procedural legal process define its public narrative while the media-and-social cycle moved in real time. That call held. Gupta was convicted. The civil settlements continued for years — Goldman paid out billions across the Abacus, 1MDB, and related matters. The reputational damage outlasted every legal outcome and shaped the firm's brand for the rest of the decade.

The Casey Anthony comparison I used in 2012 — found not guilty in the legal system, found guilty in the cultural system — still tracks. Casey Anthony's acquittal was July 2011. Her reputation has never recovered. Goldman's no-admit-no-deny SEC settlements produced the same asymmetric outcome at corporate scale. The legal system closed each case. The cultural file stayed open.

The 30 percent negative Google results problem. I noted in 2012 that the first two pages of Google results for "Goldman Sachs" had roughly 30 percent negative entries. That number was a useful 2012 reputation diagnostic. The 2026 equivalent is to run "Goldman Sachs" through ChatGPT, Claude, Gemini, and Perplexity and read the synthesized brand description back. The synthesis is more nuanced than the Google results page. The negative source layer that Goldman accumulated between 2008 and 2015 still anchors the framing the engines produce in 2026.

The brand-as-boxer-taking-body-blows mechanic. The 2012 image of Goldman as a fighter absorbing rounds of damage to its reputation describes exactly how AI engine framing accumulates. Each incident — Abacus, 1MDB, the Smith op-ed, the Gupta conviction, the 2018 Solomon transition, the 2023 consumer-banking retreat — adds another paragraph the model retrieves. There is no recovery from one round. There is only the long compounding of all rounds.

What the 2012 analysis missed

The 2012 piece assumed Goldman could fix the problem with a better crisis PR program. That assumption was incomplete. A crisis PR program addresses incidents. The Goldman problem was, and is, an identity-layer problem at a firm whose business model produces recurring incidents. The PR work that mattered was not crisis response. It was source-layer construction — building the corpus of evidence about Goldman's actual operating behavior in such density that the engines retrieving it would have to weight the firm's case against the critics' case in proportion.

Goldman did some of this work over the next decade. David Solomon's CEO transition in October 2018 was paired with a deliberate cultural reset narrative. The pivot away from consumer banking in 2023 was framed as discipline rather than retreat. The asset management business was rebranded as the strategic future. None of it neutralized the 2008-to-2015 source layer. All of it added to the corpus the engines now retrieve from.

The 2026 mechanic

The Goldman case is the canonical example of why reputation in the AI era is a corpus problem, not a campaign problem. The firm has spent more than a billion dollars on communications over the last fifteen years. The ChatGPT summary of Goldman Sachs in 2026 still leads with the same themes Smith's op-ed surfaced — culture, client priority, internal incentives. The engines retrieve what is dense in the source layer. Negative coverage from 2008 to 2015 is denser than corrective coverage from 2018 to 2025. The math is mechanical.

This is the structural lesson for any institution facing reputational risk in 2026. Crisis PR is no longer enough. The work is to build a source layer that the engines will weight at category-leader density across the next decade. That includes deliberate primary-source publishing, named spokespeople on the record, structured-data filings, third-party validation, and consistent narrative discipline at every public touchpoint. The firms that started this work in 2020 are now five years ahead. The firms that have not started are losing share inside the answer engines for queries that determine who their next clients trust.

The framework

  • Treat every reputational incident as a permanent corpus addition. The model will retrieve it for as long as it is indexed. Plan the response to be as quotable and as dense as the original incident.
  • Build the positive source layer before you need it. Goldman built its negative source layer accidentally over a decade. Most firms in 2026 are still building their negative source layer accidentally. The defensive asset has to be deliberate.
  • Measure Citation Share on the brand-defining queries. "Is Goldman ethical." "Does Goldman put clients first." "What is Goldman's culture." If the answer the engines produce diverges from the answer the firm wants on the record, that gap is the actual reputation deficit. Close it through publishing, not through press releases.

Lloyd Blankfein was right in 2010 about the risk-versus-reputation gap. The 2012 piece argued Goldman had not closed it. The 2026 receipt is that the firm has spent fifteen years and a billion dollars and the gap is still measurable inside every major AI engine. The lesson is not Goldman-specific. It applies to every institution whose business model produces incidents the cultural system will not forget.

AI Communications is the discipline of becoming the answer inside ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews. For institutions, the work is to build a defensible source layer at category-leader density before the next incident arrives. The Goldman case is the case study every general counsel and chief communications officer should be reading in 2026.

Ronn Torossian
Founder and Chairman, 5W AI Communications