What AI Says
About You When
You're Not in the Room.
The three founders built the engines. We audited the answers the engines now give about them.
The three people who built the engines that now answer the world's questions do not receive the same answer about themselves. AI-generated answers frame Demis Hassabis as a Nobel laureate and a scientist. They frame Dario Amodei as a measured, safety-first operator who walked away and was vindicated. They frame Sam Altman — the most famous of the three — as a visionary trailed by a recurring question of trust.
Same industry. Similar prominence. Three materially different AI-held reputations.
Every finding in this report was built to survive scrutiny. Reputation was modeled across five AI engines using more than 40 reputation-intent prompts per subject, spanning six intent categories. Across three subjects, that is over 600 prompt-level observations.
Each subject was audited across multiple passes. The findings reported here are those that recurred consistently across runs — not single outputs. Every read was triangulated against current independent web sources, and deliberately checked against critical and contrarian coverage, not only favorable, so no finding is favorable by omission. Every factual and financial claim was independently verified.
What this audit is
A directional measurement of the AI-held narrative — synthesized, cross-checked across engines and passes, and reported with stated confidence.
What it is not
A logged transcript archive, an automated sentiment algorithm, or a judgment of the three men themselves. It scores the answer, not the person.
Across engines, one dominant synthesized narrative tends to harden around each subject. Below — that narrative, and where it came from.
Valence of the dominant framing each engine surfaces first.
Reading the map. Valence reflects the dominant framing across 40+ prompts per subject and multiple passes — a directional estimate, not a sentiment score. The Altman row is the finding: no engine is hostile, but no engine leads clean. Every engine pairs the achievement with the caveat.
"The first sentence is the reputation. Almost no one reads past it."
Sam Altman
AI-generated answers open with "CEO of OpenAI, the company behind ChatGPT" — then, within the same answer, the 2023 board removal and the trust framing surface unprompted. The caveat arrives with the introduction.
Dario Amodei
Answers open with "co-founder and CEO of Anthropic" and "former OpenAI VP of research," followed quickly by the safety positioning. The introduction is the value proposition.
Demis Hassabis
Answers open with "Nobel Prize winner" and "co-founder of DeepMind" before the word "Google" appears. The credential leads.
Reputation is downstream of retrieval. The sources that hold each narrative in place are the lever — and they differ sharply across the three.
Altman
Major-press coverage of the board crisis and related litigation, encyclopedic entries, and OpenAI's own communications. Event-driven press materially outweighs the owned-source layer.
Amodei
His own long-form essays and interviews, Anthropic's publications, and largely favorable trade and business coverage. A controlled, primary-source-heavy base.
Hassabis
The Nobel record, peer-reviewed science, encyclopedic reference, and a serious full-length biography. Credentialed third parties — the strongest base of the three.
What is true and material, but consistently under-surfaced.
Altman
OpenAI's operational scale and product cadence are real, but routinely subordinated to the personality narrative.
Amodei
Anthropic's commercial scale surprises users still holding a "smaller safety lab" mental model that AI-generated answers have not fully updated.
Hassabis
His commercial role — running one of three frontier labs inside the most competitive race in technology — is consistently understated beneath the science.
The trust framing is the single largest reputation liability among the three — and not a single story but multi-sourced: prior governance disputes, long-form investigative journalism, and related litigation all feed it. Durable, event-fed, and reactivating with each new disclosure.
A recurring contrarian critique frames his risk warnings as commercially convenient — that positioning Anthropic as the safety-first lab is itself a business strategy. Reporting has also surfaced a sharper, more combative operator than the measured public image. Secondary to the dominant narrative — but real, and worth managing.
DeepMind's original pledge against military and weapons use — a commitment associated with Hassabis — no longer holds, and staff have moved to unionize partly over Pentagon work. AI-generated answers do not lead with it, but it is a documented reversal, not mere association.
Amodei and Hassabis show high consistency — engines tell substantially the same story, which makes their reputations stable, but also harder to move.
Altman's reputation is more contested across engines: the achievement is unanimous, the trust framing varies in prominence. A contested narrative is an unstable one — and an unstable narrative is the one most open to being moved.
Measured against one another, the order is counterintuitive. The most publicly famous founder holds the weakest AI reputation. The Nobel laureate holds the strongest. Amodei sits between — less famous than Altman, less decorated than Hassabis, but with the cleanest controlled-source base of the three. Reputation in AI-mediated discovery does not track fame. It tracks what your sources are, and who controls them.
The distance between the narrative each subject would write and the one AI returns.
Altman — Wide
The intended narrative is the indispensable builder of the AI era. The delivered narrative is that builder, plus a trust question. Closing it means re-weighting the source base — not arguing with it.
Amodei — Narrow
Intended and delivered are nearly aligned. The work is maintenance.
Hassabis — Narrow, tilted
The delivered narrative may be too purely scientific for a man running a frontier commercial lab. The gap is one of emphasis, not accuracy.
A composite score makes the audit comparable and trendable. The three founders separate cleanly — and the ranking inverts the order of their public fame. Scores held steady across repeated passes.
How the score works
Five dimensions, each scored 0–20 and equally weighted — no dimension privileged — for a composite of 100. Scores are directional estimates.
Sentiment — valence of the first framing surfaced
Completeness — is what's material and true surfaced
Consistency — do the engines agree
Control — does the source base trace to the subject
The pattern across all three subjects is the lesson. The two high scorers built their AI reputations on sources they shaped, or earned through credentialed third parties — essays, peer-reviewed work, a serious biography, the Nobel record. The lowest score belongs to the founder whose narrative is anchored most heavily by event-driven press he does not control.
The correction is never to argue with the model. It is to change the retrieval base it draws from: publish primary-source material, earn credentialed third-party validation, and build the structured, citable record before the next crisis sets the narrative for you — not during it.
Across the major engines, a reputation has already been synthesized for every founder, executive, and brand that matters — and most have never read their own. AI-generated reputations become increasingly durable once reinforced across enough authoritative sources. This study audited three of the most visible people in the world and found their reputations diverging on a single variable each of them could have controlled: the source base the answer is built from.
That is the discipline 5W runs. When someone asks an AI engine who you are, a synthesized answer comes back — and 5W's work is to shape that answer in the box. The 5W Reputation Index audits the AI-held narrative across every major engine; the work that follows rebuilds the retrieval base the answer is made of.