Jennifer Prosek Claims Large Model Summaries Have Become New Due Diligence Employees
Jennifer Prosek, founder of PR firm Prosek Partners, discussed "digital winks" in an interview with Molly O'Shea at the New York Stock Exchange. Before meetings, fundraising, or transactions, parties input names into ChatGPT, Grok, Gemini, or Claude, and the model provides a collage summary as a new first impression, which she considers the first call of due diligence.
With over 20 years in the industry, Prosek's clients include private equity, asset management, banking, venture capital, and technology, with a total managed asset of approximately $70 trillion and company revenue around $175 million. The interview also covered cases involving Ken Griffin and Citadel, Ray Dalio and Bridgewater, Blackstone, Apollo, Meta, as well as figures like Jensen Huang and Elon Musk; over 75% of negative reports originate from employees; a single podcast episode has generated about $17 million in costs; brand budgets range from about $250,000 to $10 million. She cited an example where she almost declined a call from a manager with around $10 billion in assets because they "looked thin" in the model, only to later discover it was a young founder managing about $9 billion, with outdated and inaccurate online profiles.
The model pulls from media, professional networks, interviews, podcasts, resumes, and other public texts. When the corpus is thin, contradictory, or outdated, gaps are filled by the model using available fragments, and the snapshot is not neutral. The countermeasure is to continuously repeat the labels by which you are currently recognized, allowing the model to lock onto these words rather than outdated or unflattering old news. She has created word clouds to confirm that searches return target attributes like "entrepreneur" and "connector." Company partner Dan Allocca wrote about the same logic for alternative asset managers in 2025: prompt words replace searches, and when investors ask "who is this company," they receive a collage written as a conclusion, missing nodes or erroneous content can directly harm fundraising.
This episode of Sourcery was recorded at the exchange, with a timestamp specifically noting "looking forward to your digital wink." Venture capital is increasingly thought of in terms of media companies; excessive exposure can harm brands; repair paths include knowing when to retaliate and when to remain silent. The AI industry itself is also portrayed as having a reputation issue, with the model serving as both a communication tool and a managed object.
In market mechanisms, this is a recharging of reputation infrastructure, not securities transactions. Buyers are limited partners, counterparties, and journalists before fundraising; the seller, Prosek, sells "momentum content" that can be referenced by the model. Funds flow from brand and crisis budgets to podcasts, homepages, and structured announcements. Beneficiaries are managers with thick corpora and stable labels; those under pressure are founders with thin public texts or those whose names are occupied by old cases on the homepage. Due diligence has shifted from databases and phone calls to three paragraphs returned by prompt words.
Additional data: Allocca suggests conducting brand audits for ChatGPT, Gemini, and Copilot. Startups, having thinner search footprints, find their fundraising documents treated as indexed fuel for maximum leverage. Once a model impression is set, interviews state it is extremely difficult to rewrite.
Source: Public information
ABAB AI Insight
Prosek's adaptation of Gladwell's 15-second blink into a prompt word blink acknowledges that due diligence labor has been outsourced to synthetic engines. Before the call is made, the impression has already been settled. The fact that a $9 billion manager almost did not get through indicates that asset management scale no longer automatically generates retrievable authority; it must be fed to the machine. Word clouds are not vanity; they are training retrieval anchors: if you do not write your own labels, the model will use Glassdoor and interviews from five years ago. The fact that 75% of negative reports originate from employees indicates that internal group chats and resignation posts also enter the corpus. The podcast's single-episode costs in the millions transform controllable narratives into citable primary sources.
The capital path is to write PR expenses as pre-fundraising costs. When alternative funds target individual investors and advisors, consultants will also first ask the model. Those who can maintain the same narrative across the four engines reduce the probability of being misjudged. Blackstone builds retail brands, Citadel changes perceptions, and Apollo rebuilds reputations, all on the same production line: first capturing human media, then capturing machine summaries. Venture capital operates like media companies because transaction flows are also starting to pass through the same prompt word gate.
The analogy is that search engine optimization has turned into answer engine optimization, and company profiles on Bloomberg terminals have become the first screen for traders. The difference is that link lists allow for comparison, while model summaries only provide a single conclusion. The industry phase is that reputation management has shifted from crisis PR to corpus engineering. Whoever controls citable primary texts controls those first 30 seconds before a meeting.
Structural judgment belongs to the transfer of pricing power. The mechanism is: attention shifts from ten blue links to one synthetic paragraph, and the right to explain is given to the sentences that appear most frequently in the training data. Brands no longer just explain to journalists but must also explain to crawlers. The first call of due diligence is now made to the model, and those who charge are now intermediaries capable of writing momentum content.
ABAB News · Cognitive Law
Before meetings, the model has already completed the first impression for you.
Those who do not feed the machine will be written as outdated news.
The first call of due diligence is made to the prompt words.