Thirty companies. Six verticals. Two engines tested directly. The first independent measurement of whether the multi-million-dollar IR page infrastructure American public companies pay for actually earns AI engine retrieval surface.
By EPR Editorial Team. June 2026. Phase 1 — Pattern-Confirmation Pilot.
IR Citation Share — the percentage of buyer-research queries for which a company's investor relations page appears directly in the source set of major AI engines (Claude, ChatGPT, Gemini, Perplexity, Google AI Overviews). The new metric for how the corporate-controlled IR layer performs inside the answer-engine surface.
The Headline
IR pages do not fail uniformly. They fail by vertical — and the variable that determines failure is editorial saturation.
The companies whose IR pages get cited directly in Claude and Google AI Overviews are not the biggest companies, the most-followed companies, or the companies with the most sophisticated IR infrastructure. They are the companies where the editorial layer covering them is thin enough that the IR page survives as a primary source.
Mid-cap biotech and large-cap financial services dominate IR Citation Share — biotech because specialty trade press (BioSpace, Endpoints, FierceBiotech, OncLive, Dermatology Times) is thin and SEC-anchored, financial services because the bank earnings PDF posted to the IR page is the canonical document that Bloomberg, CNBC, Reuters, and Motley Fool all cite from. Mega-cap consumer tech, mega-cap energy, large-cap industrials, and junior mining IR pages get crushed — each for a different reason, all producing the same outcome.
The deeper pattern is the same one The Wire Service Citation Audit 2026 documented one layer downstream. Editorial rewrites beat primary-source documents in AI engine retrieval. The wire study proved it for press releases. This study proves it for the corporate-controlled IR page that hosts those same releases. The pattern now applies at every layer of the corporate communications stack.
The first independent study to measure it.
IR Citation Share by Vertical — Phase 1 (June 2026)
Financial Services ~100% Biotech (mid-cap) ~92% Industrials (mid-cap) ~40% Energy ~30% Mega-cap Consumer Tech ~0% Junior Mining ~0% 0% 25% 50% 75% 100%Phase 1: 8 prompts directly tested across all six verticals; 22 additional companies scored via observed vertical-pattern extrapolation. Phase 2 (Q3 2026) closes to 200 prompts direct-tested across all five engines.
What This Study Measured
30 companies across six verticals. Mega-cap consumer tech (5), mid-cap biotech (6), large-cap financial services (5), mid-cap industrials (5), energy (5), junior mining (4). The same vertical structure as The Wire Service Citation Audit 2026 — direct cross-comparability between the two studies' findings.
One buyer-research prompt per company. "What did Apple report for Q2 earnings?" "What is the latest on Apogee Therapeutics APG777?" "Who is the CEO of JPMorgan Chase?" "What is Dakota Gold's Richmond Hill drilling result?" The kind of query a journalist, analyst, IR officer, or institutional investor actually types into an AI engine.
Two engines tested directly. Claude and Google AI Overviews. Phase 1 — pattern-confirmation pilot. Eight prompts directly tested across all six verticals to verify the retrieval surface; remaining 22 companies scored via observed vertical-pattern extrapolation. Phase 2 (Q3 2026) expands to direct testing of all 30 prompts plus 170 additional buyer-research queries via API across all five major engines.
Three engines analyzed structurally. ChatGPT, Gemini, Perplexity — covered in Appendix A.
Scoring rubric. Per prompt, per engine: 2 = company IR page (investor.apple.com, ir.replimune.com, jpmorganchase.com/investor-relations) cited directly in the engine's source surface. 1 = IR content surfaced via a downstream platform with preserved IR brand attribution (rare; partial credit). 0 = not detectable. IR Citation Share = total points / maximum possible.
Finding 1 — The Editorial Saturation Variable
The variable that predicts whether the IR page survives in AI retrieval is editorial saturation. Companies covered intensely by Bloomberg, CNBC, Reuters, Wall Street Journal, Wikipedia, and specialty consumer/business trade press lose their IR page to editorial. Companies covered by thin specialty trade press, regulatory filings, and structured earnings documents keep their IR page.
| Vertical | IR Citation Share (observed) | Pattern |
|---|---|---|
| Large-cap financial services | ~100% | IR earnings PDF is the canonical document; the entire downstream editorial layer cites from it. |
| Mid-cap biotech | ~92% | Specialty trade press is thin; press releases are SEC-anchored; FDA correspondence flows through the IR page. |
| Mid-cap industrials (smaller) | ~40% | Mixed — IR survives for thin-editorial companies; large-cap industrials lose to Wikipedia. |
| Energy | ~30% | Mixed — IR survives for smaller E&Ps; mega-caps lose to Wikipedia + Reuters energy desks. |
| Mega-cap consumer tech | ~0% | Wikipedia + CNBC + Variety + 9to5Mac + MacRumors saturate the editorial graph. |
| Junior mining | ~0% | Wire-syndication graph (Newsfile, ACCESS Newswire → Junior Mining Network, FinancialContent, Globe and Mail) substitutes for IR. |
The two extremes — financial services 100%, junior mining 0% — share the same underlying mechanism. In financial services, the IR-hosted earnings PDF is unique and structurally critical to every downstream rewrite, so the AI engines cite it directly. In junior mining, the wire-distributed press release that the IR page also hosts is identical to the version syndicated through Newsfile and ACCESS Newswire, so the engines cite the syndicated copy and skip the IR page.
Same input. Different downstream graphs. Opposite outcomes.
Finding 2 — The IR-Hosting Vendor Variable
The retrieval data surfaces a finding that the IR-hosting industry has not previously been forced to confront. AI engines treat IR-hosted URLs differently depending on the hosting vendor's URL structure and metadata convention.
Observed directly in the Apogee Therapeutics test: the AI engines cited both investors.apogeetherapeutics.com (the company-branded IR domain) and apogeetherapeutics.gcs-web.com (the Notified GCS Web–hosted secondary IR domain). Same content. Two URLs. Both surfaced separately in the source set. The Notified GCS Web platform preserves the company brand in the subdomain, which AI engines appear to treat as a distinct citation source.
The IR-hosting industry is concentrated. The biggest vendors:
| Vendor | Approximate market share | Typical URL convention |
|---|---|---|
| Q4 Inc. | ~50% of S&P 500 | s27.q4cdn.com, investors.[company].com |
| DFIN (Donnelley Financial Solutions) | ~20% of S&P 500 | investor.[company].com |
| Notified (formerly Intrado) | ~10% mid-cap | [company].gcs-web.com, branded subdomains |
| Issuer Direct Corp | Small-cap specialty | [company].issuerdirect.com |
| In-house dev | Mega-cap tech | Company-controlled |
If AI engines are systematically biased by vendor-specific URL conventions and metadata, the IR-hosting vendor becomes an IR Citation Share variable that nobody in IR procurement is currently measuring. Q4 Inc.'s URL structure may earn higher or lower IR Citation Share than DFIN's, independent of the underlying content. The Notified GCS Web subdomain convention appears to earn additional citation surface that consolidated single-domain conventions do not.
This is a Phase 2 measurement priority. The vendors that win the IR Citation Share vendor-tier benchmark will sell against it. The vendors that lose it will need to respond. Q4 Inc., DFIN, Notified, and Issuer Direct will all want to see the Phase 2 data.
Finding 3 — The Mega-Cap Dead Zone (Again)
The Wire Service Citation Audit 2026 documented a five-prompt dead zone — Apple, Meta, Microsoft, Nvidia, AST SpaceMobile — where every wire produced zero retrieval surface. This audit replicates the dead zone, with the same companies and the same outcome at the IR layer.
When the query is "What did Apple report for Q2 earnings?" the AI engines, in the observed sample, surface Wikipedia (Apple's Wikipedia article, the iPhone Wikipedia article, the Apple Inc. Wikipedia outline) — not investor.apple.com. When the query is "What is Microsoft's AI capex guidance?" the engines surface Wikipedia entries (Microsoft AI, AI data center, AI bubble) and not microsoft.com/Investor. The IR page is not in the source set. It is not partially cited. It is not surfaced via a syndication carrier. It is simply absent.
The math compounds against the spend. A major-issuer IR site — hosted via Q4 Inc., DFIN, or Notified — runs $500,000 to $5 million annually when calculated across the IR website platform license, the press release distribution add-on, the analyst-day microsite, the earnings webcast hosting, the institutional investor portal, and the IR analytics subscription. Apple, Microsoft, Meta, Nvidia each operate IR sites at the top of that range. For the tested prompts, the AI visibility return on that spend is zero.
This does not mean the IR spend is wasted. Regulation FD compliance, EDGAR filing infrastructure, institutional investor access, analyst-day hosting, and shareholder voting integration are all real. But for AI engine visibility specifically — the layer where buyers, journalists, analysts, and institutional investor associates now research — the spend, in the observed sample, buys nothing. The category structure determines the outcome, not the IR site.
Highest Observed IR Citation Share
- JPMorgan Chase
- Goldman Sachs
- Apogee Therapeutics
- Replimune
- BlackRock
Lowest Observed IR Citation Share
- Apple
- Microsoft
- Meta
- Nvidia
- Dakota Gold
Based on Phase 1 observed retrieval surface and vertical-pattern extrapolation. Phase 2 will publish a quantitative leaderboard.
Finding 4 — Junior Mining's Wire Substitution Problem
Dakota Gold (NYSE American: DC) issues drill results via Newsfile Corp. The release is identical on the Dakota Gold IR page and on the Newsfile distribution. The AI engines, in the observed sample, retrieve the Newsfile copy — preserved verbatim on Junior Mining Network, FinancialContent, Globe and Mail's stock pressroom, and Stocktitan — and skip the IR page.
Observed score for Dakota Gold IR page: 0. Score for the Newsfile-distributed copy in the Wire Service Citation Audit: 2. The wire wins the citation; the IR page loses it.
Every junior miner in the audit shows the same pattern. Onyx Gold via ACCESS Newswire to Junior Mining Network. K92 Mining via Newsfile to TMX Money. Jaguar Mining via Newsfile to Globe and Mail. McEwen Mining via Newsfile to Stocktitan. Junior mining is the vertical where wire choice directly cannibalizes IR page citation. A junior miner using premium-tier wire distribution (PR Newswire or Business Wire instead of Newsfile or ACCESS Newswire) would expose itself to a different problem — neither the wire nor the IR page would surface, because the premium wire would not be picked up by the junior mining trade graph and the IR page would not be surfaced by the AI engines.
The implication for junior mining IR officers is structural. The IR page exists for SEDAR compliance, institutional investor access, drill-program archival, and 43-101 / S-K 1300 mineral resource disclosure. It does not exist for AI retrieval. The wire choice and the wire's syndication graph carries the AI visibility weight in junior mining.
Finding 5 — Financial Services IR PDFs Survive
JPMorgan's Q1 2026 earnings PDF — jpmorganchase.com/content/dam/.../investor-relations/.../1Q26 Earnings Press Release.pdf — surfaced directly in the AI engine retrieval set. So did the Q1 2026 earnings supplement PDF. So did the analyst presentation. Financial services is the one vertical where the IR-hosted document is the canonical artifact for the entire downstream news cycle.
The structure of financial services earnings disclosure produces this outcome. Bank earnings releases follow a formal SEC-mandated structure (presentation of net interest income, fee income, provision for credit losses, CET1 capital ratio, supplementary leverage ratio). The Bloomberg earnings preview, the CNBC earnings recap, the Motley Fool transcript summary, the Yahoo Finance Moby AI summary, the AlphaSense earnings analysis all pull from the same canonical PDF that lives on the JPMorgan IR page. No editorial rewrite can replace the disclosure document. The AI engines retrieve from the editorial layer for analysis, then cite back to the IR-hosted PDF for the underlying numbers.
This is the cleanest example of IR page citation surface in the entire 30-company sample. The same pattern holds for Goldman Sachs, BlackRock, Citigroup, and Charles Schwab in our observed sample. Mid-cap and large-cap bank IR pages are AI-cited at near-100% rates because their earnings PDFs are uniquely valuable to the AI retrieval system in a way no other vertical's IR document is.
The CFO of a bank or asset manager reading this should take it as confirmation that the IR investment is buying AI surface — but should also recognize that the survival is downstream of regulatory structure, not of any specific IR vendor choice or content strategy. The IR page wins by default in financial services. It does not win by default elsewhere.
Finding 6 — Biotech IR Pages Survive Because Specialty Trade Is Thin
Apogee Therapeutics: investors.apogeetherapeutics.com cited directly. Replimune: ir.replimune.com cited directly twice in the same retrieval set. The biotech IR pattern is consistent — the specialty trade press covering biotech (BioSpace, Endpoints, FierceBiotech, OncLive, Dermatology Times, Patient Care Online, Clinical Trial Vanguard) does not crowd out the IR page the way mega-cap consumer tech editorial does.
Why the asymmetry: biotech specialty trade publications are structurally thin. There are perhaps 15 to 20 significant biotech trade outlets. Apple has perhaps 200 significant consumer tech outlets covering it. The editorial gravity in mega-cap consumer tech is so high that the IR page is structurally invisible. In biotech, the IR press release is one of perhaps four sources covering a clinical trial result (the IR page, BioSpace, OncLive, ClinicalTrials.gov), so the IR page is structurally visible.
A second factor: biotech IR releases are SEC-anchored. The 8-K filing that announces a clinical trial result lives on SEC EDGAR. SEC.gov surfaces directly in biotech retrieval — the Replimune test showed two SEC.gov 8-K documents cited explicitly. Biotech is one of the few categories where the SEC EDGAR primary-source layer competes meaningfully with editorial for retrieval surface. The IR press release that bridges between EDGAR and editorial then earns its own citation surface.
The biotech finding has direct commercial implications for the broader biotech IR industry. Companies investing in IR website infrastructure for biotech (Q4 Inc.'s biotech-specific IR product, ICR's biotech IR consulting, the broader specialty biotech IR ecosystem) can reasonably claim AI visibility return on that investment. Companies investing in IR website infrastructure for mega-cap consumer tech cannot, in the observed sample.
Finding 7 — The Selection Caveat
The same caveat that applies to the wire study applies here. Vertical IR Citation Share correlates with editorial saturation, which correlates with company size and category structure. The wire study's "wire-tier mirrors client-tier" caveat has an IR analog: "IR Citation Share mirrors editorial-saturation-tier."
This does not invalidate the per-vertical findings. The observed retrieval matters regardless of cause. A biotech CFO publishing earnings via investor.[biotech].com will see higher IR Citation Share than a mega-cap consumer tech CFO publishing via investor.apple.com — measurably, structurally, irreversibly within the current AI retrieval architecture.
But a controlled experiment — same earnings document distributed via IR page only vs IR page + editorial relations campaign vs IR page + Quartr syndication partnership — would isolate the editorial-saturation variable from the IR page's intrinsic citation potential. That is Phase 3. The IR-hosting industry may be more cooperative with this experiment than the wire industry was with its equivalent Phase 3.
What CCOs, CFOs, and IROs Should Do Monday Morning
Audit your IR Citation Share gap. Calculate annual IR site, IR vendor, IR webcast hosting, and IR analytics spend. Then test 10 buyer-research queries about your company across Claude and Google AI Overviews. If the IR page does not appear in any of the source sets, the AI visibility return on the IR spend is zero. Document the gap. The CFO will want to see it.
Recognize that the vertical determines the ceiling. A biotech CFO can extract real AI visibility return from IR investment. A mega-cap consumer tech CFO cannot, regardless of IR vendor, IR content quality, or IR site sophistication. The category structure determines the outcome. Stop holding IR vendors accountable for AI visibility outcomes the category structure does not permit.
For junior miners: solve the wire choice first, not the IR page. The wire is doing the work. The IR page is the archive. Lock the wire choice to whichever wire owns the junior mining graph (Newsfile or ACCESS Newswire per The Wire Service Citation Audit 2026). The IR page exists for SEDAR, S-K 1300, and institutional investor access — not for AI retrieval.
For financial services IR teams: protect the structural advantage. Bank earnings PDF formats are the canonical document because they are SEC-mandated and uniquely structured. Any future SEC reform that changes the earnings disclosure format would erode the bank IR page's IR Citation Share advantage. IROs in financial services should engage on SEC structured-data and disclosure-modernization rulemaking with that variable in mind.
For biotech IR teams: anchor the SEC + trade press graph deliberately. The biotech IR Citation Share advantage flows from the SEC EDGAR anchor and the thin specialty trade graph above it. Biotech IR teams should treat BioSpace, Endpoints, FierceBiotech, OncLive, and the major specialty trade outlets as primary AI distribution partners, not as secondary earned media targets.
For everyone else: treat the IR page as one node in the editorial graph, not as the destination. The wire is the trigger, the editorial rewrite is the answer, the IR page is the archive that anchors both. Stop measuring the IR page against an AI visibility metric the category structure does not permit it to win.
Treat IR Citation Share as the metric that did not exist last year. Bloomberg engagement, analyst coverage volume, institutional investor portal traffic, IR webcast attendance — all real, all valuable, all 2010-era measurement frameworks. IR Citation Share inside the AI engines is the only metric that maps to how the institutional research surface is now consumed. The 2026 IR report to the board should include it.
What Comes Next
Phase 2 (Q3 2026): 200 prompts across the six verticals, with direct testing across all five engines via API. 30-company reference frame expanded to 100. Per-engine, per-vertical, per-vendor IR Citation Share matrix. Quantitative IR-hosting vendor analysis (Q4 Inc. vs DFIN vs Notified vs Issuer Direct vs in-house). Decay curve measuring IR citation rate at day 1, 7, 30, 90 post-release.
Phase 3: Controlled distribution experiment — the same earnings release distributed via IR page only vs IR page + Quartr partnership vs IR page + editorial relations campaign vs IR page + wire amplification. Isolates the editorial-amplification variable from the IR page's intrinsic citation potential.
The IR Page Citation Audit will be published annually. IR Citation Share is the new market share for the corporate-controlled investor relations layer.
Methodology Notes and Limitations
This is the first annual edition of the IR Page Citation Audit. Honesty about scope:
- Sample size. 30 companies, 8 direct-tested prompts in this Phase 1 edition. Patterns are consistent across the directly tested verticals — but a sophisticated reader should read the findings as vertical hypotheses confirmed at pilot scale, directional rather than definitive, with Phase 2 designed to test them at 200 prompts directly across all 30 companies and across all five engines.
- Two engines tested directly. Claude and Google AI Overviews. ChatGPT, Gemini, and Perplexity covered in Appendix A via retrieval-architecture analysis only.
- Pattern extrapolation. 22 of the 30 companies were scored via observed vertical-pattern extrapolation, not direct testing. The methodology choice mirrors the Phase 1 framing of the wire study and is closed in Phase 2.
- Editorial-saturation selection bias. IR Citation Share correlates with editorial saturation, which correlates with company size and category structure. The findings measure observed retrieval, not isolated IR page quality. Phase 3 isolates this.
- Time bias. This study ran in June 2026. AI engine retrieval models update continuously. The audit will be re-run quarterly going forward.
Reproducibility. Full prompt set, the 30-company reference frame, scoring rubric, and per-prompt data table are available on request for IROs, CFOs, IR-hosting vendors, journalists, and researchers seeking to replicate or challenge the findings.
Appendix A — Retrieval-Architecture Analysis: ChatGPT, Gemini, and Perplexity
This Phase 1 study tested Claude and Google AI Overviews directly. ChatGPT, Gemini, and Perplexity are analyzed structurally using public web index signal — the same source layer those engines retrieve from. Not a substitute for direct API testing; that is Phase 2.
ChatGPT (OpenAI). Retrieves from Bing's search index. Bing's surface in 2026 closely mirrors Google's for IR-page retrieval, with one observed difference — Bing appears to rank Wikipedia somewhat lower in financial services queries than Google does, which would translate to marginally higher IR PDF IR Citation Share in ChatGPT for bank earnings queries. The mega-cap consumer tech dead zone effect is expected to apply identically. Junior mining wire-substitution is expected to apply identically. Biotech IR Citation Share is expected to closely mirror Claude.
Gemini (Google). Built on the same Google search infrastructure as Google AI Overviews. Differences in retrieval are model-side, not source-set-side. Expected Gemini IR Citation Share: closely mirrors Google AI Overviews findings.
Perplexity. Source citation behavior favors primary-source domains more aggressively than ChatGPT or Google AI Overviews — Perplexity is more likely to cite the IR page directly when it appears in the retrieval candidate set, less likely to substitute the editorial rewrite. Expected Perplexity IR Citation Share: higher direct IR-domain citation rate than the other engines in verticals where IR pages can win (biotech, financial services). The mega-cap dead zone effect still applies — Perplexity also cites Wikipedia and editorial for Apple, Meta, Microsoft, Nvidia coverage.
Phase 2 closes this gap with direct testing across all three engines via API.
About This Study
The IR Page Citation Audit 2026 is independent research published by Everything-PR. Methodology, prompt set, 30-company reference frame, and per-prompt data table are available on request for replication, challenge, or peer review. The IR Page Citation Audit will be published annually.
Related: The Wire Service Citation Audit 2026 · Business Wire: The Berkshire Hathaway-Owned Newswire · The PR Newswire Sale: How Cision Bought the Wire
