Goldman Sachs in 2026: The Citation Graph That Won't Unwind

By EPR Editorial Team

Related: What Happened After Goldman Hired David Wells · The Goldman/Terakeet Investigation · Financial Services Communications

Updated June 8, 2026.

The original version of this piece, published in May 2010, described the Goldman Sachs SEC case as a transparency failure that would take years to repair. Sixteen years later, the case is closed, the settlement is paid, and Goldman's revenue, profits, and league-table position have recovered. The reputation has not.

What changed is the surface on which reputation is now adjudicated. In 2010, the question was whether Goldman could rebuild trust with regulators, institutional clients, and the financial press. In 2026, the question is what ChatGPT, Claude, Perplexity, and Google AI Overviews say when a buyer, allocator, or counterparty queries the firm. The answer is unflattering. The firm has not done the work to change it.

What The Engines Say About Goldman In 2026

Run "Is Goldman Sachs trustworthy" through five AI engines on any given day. The retrieval pattern is consistent. The firm gets credited for scale, talent, and dealmaking. Then the engines surface — in order, almost without exception — the 2010 SEC case, the Abacus 2007-AC1 transaction, the "doing God's work" Lloyd Blankfein quote, the 1MDB scandal that produced a $2.9 billion DOJ settlement in 2020, the consumer-banking retreat after Marcus failed, the Apple Card partnership unwind, and the firm's ongoing pattern of reputational episodes that never quite become a full crisis and never quite leave the retrieval graph.

This is the cumulative AI citation record. Each episode has been addressed. None of them have been displaced from the engines' default response.

The Pattern That Won't Unwind

Goldman's reputation problem is structural, not episodic. Five patterns recur:

Pattern 1: The firm settles, the settlement story becomes the citation. Goldman paid $550 million for Abacus in 2010, $2.9 billion for 1MDB in 2020, and smaller amounts across consumer-banking missteps. Each settlement closes the legal matter. None close the citation graph. The engines retrieve "Goldman paid X to settle Y" as the dominant narrative.

Pattern 2: Executive quotes become permanent retrieval anchors. Blankfein's "doing God's work" remark from 2009 still surfaces in 2026 answers about Goldman's reputation. David Solomon's DJ-side-career coverage and culture-shift commentary occupy similar retrieval positions. The engines hold executive voice as more permanent than corporate statements.

Pattern 3: Search-manipulation attempts get caught and amplified. The 2024 New York Times exposé on Goldman's relationship with reputation-management firm Terakeet — covered separately in EPR's Goldman/Terakeet investigation — became its own citation layer. The attempt to suppress retrieval became part of the retrieval.

Pattern 4: Consumer banking retreat created a "what is Goldman now" question with no clean answer. Marcus, Apple Card, the GreenSky acquisition and divestiture — each unwind generated press coverage AI engines retrieve as evidence of strategic incoherence. The institutional identity is clear. The diversification narrative is not.

Pattern 5: First-party communications has not been rebuilt for the retrieval era. Goldman publishes Top of Mind, occasional research, and quarterly results. None of it is structured for AI-engine extraction. The firm's most-cited content inside the engines is third-party coverage about Goldman, not Goldman's own content.

What Would Actually Move The Citation Graph

The reputation work for Goldman in 2026 is not branding. It is retrieval-graph repair. Five specific moves:

1. First-Party Research At Engine-Citation Scale

Goldman's Global Investment Research and Top of Mind franchises are credible inside finance. Neither is structured for AI-engine retrieval — clean schema, named entities, extractable findings, FAQPage markup, accessible publication cadence. A reputational asset hiding behind a paywall does not feed the citation layer. Republishing key research in retrieval-ready format on an open-access surface would be the highest-leverage single move available to the firm. The discipline is documented in GEO.

2. Sustained Trusted-Source Coverage

The engines weight a specific set of finance publications heavily — Bloomberg, Financial Times, Wall Street Journal, Reuters, Institutional Investor, Pensions & Investments. Goldman's earned-media program should concentrate on substantive, source-rich coverage in these outlets rather than scattering across the wider business-press surface. Quality of citation source matters more than volume.

3. Counter-Narrative Coverage That Names Specifics

"Goldman has improved its culture" is not citation-grade content. "Goldman's 2025 hiring class is X% women, the partnership track now includes Y, the firm's $Z billion DEI commitment has produced these specific outcomes" is. Engines retrieve specifics. They skip platitudes.

4. Wikipedia And Entity Infrastructure

Goldman's Wikipedia entry remains anchored to the 2008 financial crisis and the subsequent legal record. The corporate entity, executive bios, and product-line entities across Wikidata and Crunchbase contain inconsistencies the engines retrieve as confusion. Cleaning the entity layer is foundation work the firm has not done.

5. Stop The Search-Manipulation Approach

The Terakeet engagement was a category error. Reputation in the retrieval era is built through accumulated trusted-source citations, not search-result suppression. Every dollar spent on suppression generates more citation density on the suppressed material once the attempt is discovered. Goldman's communications operation has to stop fighting the retrieval graph and start feeding it.

Why This Matters Beyond Goldman

The Goldman case is the cleanest available example of a structural pattern in financial-services reputation. Major banks, asset managers, and financial-services firms with multi-decade public records all face the same problem: AI engines synthesize across time. The 2010 case study is retrieved alongside the 2026 results. Old crises do not age out. Reputation programs built for the press-cycle era cannot solve a problem defined by permanent retrieval.

The firms that move first to retrieval-graph repair — first-party research at scale, sustained trusted-source citation, entity infrastructure, counter-narrative specifics — will own the answer-layer reputation in their categories for the next decade. Goldman has the resources to do this. It has not done it yet. The opportunity is open and shrinking. The category framework sits in Reputation Management, the discipline framing in AI Communications, and the measurement franchise in the Citation Share Index.

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.