Public-company narratives are no longer assembled by sell-side analysts. They are assembled by retrieval systems — and the systems are reading a different stack than the analysts ever did.
That is the structural shift. Everything downstream — IR strategy, disclosure drafting, executive visibility, crisis posture — runs through it.
The input stack the model reads.
Ordered by current retrieval weight across the major engines:
Wikipedia and Wikidata. Top-tier weight on every entity question. The 10-K does not beat Wikipedia in retrieval. Neither does the earnings transcript.
SEC EDGAR filings. Authoritative but under-cited. Models reach for filings second, not first.
Tier-one financial media. Reuters, Bloomberg, WSJ, FT, CNBC, Barron's.
Transcript vendors and aggregators. Motley Fool, Seeking Alpha, AlphaSense, FactSet, Capital IQ.
Trade and sector press. Vertical depth, sector authority.
Social training surfaces. Reddit (now licensed into multiple major engines), X/FinTwit, LinkedIn long-form.
Retail-investor content. YouTube transcripts, financial newsletters, Substack analysts.
The engines do not rank this stack the way a securities lawyer would. They rank it by Citation Dominance — repetition across surfaces, weighted by authority.
Entity Authority and Entity Drift.
The summary the model returns is anchored by Entity Authority — how consistently the company's name, segments, products, and executives appear in identical or near-identical form across the stack. Inconsistency creates Entity Drift: the model fragments the entity into pieces and reassembles it into a less coherent profile. An issuer that calls its core business by three different names across its filings, earnings scripts, and investor deck produces a fragmented retrieval profile and a confused summary.
The retrieval anchor.
Inside every public-company narrative there is a dominant frame the model repeats — the retrieval anchor. A tagline. A risk-factor phrase. A growth-story claim. Once anchored, it is hard to dislodge without sustained input-side intervention. The anchor is shapeable. Most companies are not shaping it. They are letting the most-repeated random phrase win.
Why the CEO drives the corporate retrieval profile.
Engines cluster company profiles around named individuals more reliably than around corporate entities. Companies rename, restructure, get acquired. People persist. A CEO with a dense content footprint pulls the company's profile in their direction. A CEO with no footprint leaves the profile to whatever the model can scrape — which is often less flattering and almost always less precise.
This is the structural reality: citation share is the new market share. Inside an AI-mediated buyer market — and the buyer market for public-company equity is now AI-mediated for a meaningful share of institutional discovery — the issuer that gets retrieved gets considered. The issuer summarized cleanly gets understood. The issuer summarized poorly gets discounted, often without anyone in the IR organization knowing why.
The next twenty-four months will sort the issuers who built Retrieval Governance functions from the ones who didn't. The sorting will be visible in two places: multiple expansion in the first group, and unexplained drag on the second.
Everything-PR. Reporting the AI-mediated capital markets layer. Publishing since 2009.
