When a registered investment adviser (RIA) publishes a claim—that a strategy returned 8.2% over five years, that it outperformed a benchmark, or that it carried lower volatility than an index—the evidence behind that claim matters twice over. Once for compliance: the Securities and Exchange Commission's marketing rule (Advisers Act Rule 206(4)-1) requires the adviser to have a reasonable basis for believing it can substantiate the claim on demand. And once for discovery: a search engine or AI answer engine reading your page needs to recognize that evidence exists and is legitimate.
Most firms publish the evidence. They file performance footnotes, link to third-party verification, cite regulatory guidance. What many do not do is structure that evidence so algorithms can read it. A human can see a chart with a citation at the bottom and understand the connection. A search algorithm cannot. An answer engine cannot. It sees text, metadata, and structured data: tags that tell it what a claim is, what kind of evidence supports it, and whether that evidence comes from an authority it should trust.
Without that structure, your compliance file is complete and your pages still rank as unsubstantiated. You have written the proof but not told the machine where to find it.
What Evidence Architecture Means
Evidence architecture is the practice of linking claims to their sources in ways that machines can parse. It includes three elements: the claim itself, the reference to the supporting material, and the metadata that explains the relationship.
A human reader of an adviser's website might see a statement like this: "Our Core Equity strategy has outperformed the S&P 500 by an average of 1.8% annually over the past ten years." Below it, in smaller type, sits a note: "*Performance data as of December 31, 2024. See detailed performance report."
A search engine reading that same page sees text. It sees the claim. It sees the note. But without machine-readable markup, it has no way to know whether the note is a supporting citation, a disclaimer, a marketing flourish, or an advertisement. It cannot tell whether the "detailed performance report" is a verified source or a link to a competitor's site. It cannot assess authority.
Evidence architecture solves this by using structured data. These are invisible tags that tell algorithms exactly what they are reading. Schema markup, a standardized language search engines and answer engines understand, can label a claim as a factual statement, point to the evidence document, identify the source as verified, and note the publication date. When done properly, an algorithm recognizes not just that you made a statement, but that you backed it up with traceable, dated, verifiable material.
Why Algorithms Struggle Without It
Search engines and answer engines rank pages partly on authority. An answer engine, the AI system that powers Google's AI Overviews and similar products, needs to choose which source to cite when it answers a user's question. It looks for signals of credibility: citations, quotes from verified sources, links to regulatory documents, dates.
When those signals are structured, the engine reads them easily. When they are not, the engine must guess.
Consider a composite example. Derek runs a mid-sized RIA. His firm publishes a white paper on tax-loss harvesting strategies, and the paper cites several Internal Revenue Service (IRS) documents and a piece of peer-reviewed research on behavioral finance. The citations exist in the paper. They are there, in full. But Derek's web team embedded them as plain text, with no markup indicating that they are citations, no dates, no direct links to the original sources.
When a searcher asks "Is tax-loss harvesting effective?", an answer engine reads Derek's white paper. It sees citations. But because they are not tagged as such, the algorithm must infer their role. It may treat them as casual references rather than substantive evidence. It may not know whether the IRS documents are current or historical. It will certainly not be able to verify them without human intervention.
Now imagine Derek's team restructures the same citations using Schema markup. Each reference is tagged as a citation, with the source name, the original publication date, and a direct link to the authoritative document. The algorithm reads the same words. But this time it understands their function. It knows Derek is drawing on official regulatory guidance, not opinion. It can verify the IRS guidance is current. It can assess the citation's credibility.
The white paper's text has not changed. The evidence has not changed. The way an algorithm reads that evidence has.
The Three Components of Evidence Architecture
Claims and Assertions
The claim is the statement you are making. It should be specific and falsifiable. "Our strategy performed well" is not a claim in this sense. "Our Core Equity strategy returned 9.1% net of fees in 2023" is.
In HyperText Markup Language (HTML), you can mark a claim using the ClaimReview schema, which is designed exactly for this purpose. It tells algorithms that what follows is an assertion you can substantiate. The markup looks simple: a few lines of code. But it changes how an algorithm processes the sentence.
Derek's white paper might open with a claim tagged this way:
"Our tax-loss harvesting framework has reduced net tax liability by an average of 1.2% annually for clients in the highest federal tax bracket (as of December 31, 2024)."
The claim is tagged. Now the algorithm knows it is not marketing language. It is a factual statement that should be verifiable.
Citations and References
A citation is a pointer to the source material that backs the claim. It should include the source name, the author (if not the source itself), the publication date, and ideally a direct link to the original document.
For Derek's tax-loss harvesting statement, a proper citation might point to:
- A peer-reviewed academic paper on tax-loss harvesting outcomes, with a stable link
- Internal performance data, audited by a third party, with a publication date
- Regulatory guidance from the IRS or SEC, with a direct link to the current version
Each citation should be tagged in a way that tells an algorithm what kind of source it is. A regulatory document carries different weight than a blog post. An audited performance report carries different weight than an internal calculation. Schema markup lets you specify the source type: GovernmentOrganization, AcademicArticle, FinancialService, and others.
Metadata and Dates
Metadata is information about information. It includes the publication date of your claim, the date the supporting evidence was current as of, and the date you last verified the citation.
This matters because algorithms penalize stale evidence. A tax guide from 2016 might contain useful information. But if it is not marked as historical, an algorithm may treat it as current guidance. If Derek publishes a white paper on tax law in 2024 and cites a 2019 analysis without noting the date gap, an algorithm may assume the analysis still applies.
Proper metadata includes:
Publication date of your claim or article. When did Derek's white paper go live?
As-of date of the evidence. Derek cites performance data "as of December 31, 2024." That date is material.
Last verified date. Derek's team confirms annually that the IRS guidance cited is still current. That verification date matters to algorithms assessing freshness.
When these dates are tagged and visible, algorithms can assess whether your evidence is still relevant. When they are buried or missing, algorithms treat all citations as equally current or equally suspect.
How to Implement Evidence Architecture in Practice
Start with your compliance infrastructure. Most RIAs maintain a document repository: a shared drive, a governance platform, or a compliance management system (CMS) where they store performance reports, audit letters, regulatory filings, and source materials. This is where evidence lives.
Next, identify the claims on your website or in your content that require substantiation. For an RIA, these typically include:
- Performance figures (returns, benchmarks, volatility, fees)
- Regulatory approvals or accreditations
- Third-party ratings or recognitions
- Research findings or investment theses
For each claim, establish the source of truth. Not the marketing version. The compliance version. The audited performance report. The SEC filing. The original academic paper. The regulatory guidance. Document the publication date and the effective date.
Then, work with your web team or your content platform to add structured data markup. This does not require rebuilding your site. It means adding invisible tags to the HTML code behind each claim. Tools like Google's Structured Data Testing Tool let you verify the markup is correct before you publish.
For Derek's white paper, the code might look like this:
<div itemscope itemtype="http://schema.org/ClaimReview">
<span itemprop="claimReviewed">Our tax-loss harvesting framework has reduced net tax liability by an average of 1.2% annually for clients in the highest federal tax bracket.</span>
<span itemprop="datePublished">2024-10-15</span>
<span itemprop="author">Composite RIA</span>
<div itemprop="reviewRating" itemscope itemtype="http://schema.org/Rating">
<span itemprop="ratingValue">substantiated</span>
</div>
<div itemprop="citation">
<a href="[link to performance report]">Audited Performance Report, December 31, 2024</a>
</div>
</div>The words on the page are the same. But the code now tells an algorithm: this is a claim. It was published on this date. It has been substantiated. Here is where to find the evidence.
Why This Matters for Findability
Search engines and answer engines prioritize sources that show their work. A page with unsubstantiated claims ranks lower than a page with the same claims backed by traceable evidence. But evidence that is not machine-readable looks like no evidence at all.
When Derek's white paper cites the IRS guidance in plain text, a search engine sees the citation but cannot verify it without human review. When the same citation is tagged and linked directly to the IRS.gov page, the algorithm can verify it instantly. It can confirm the source is authoritative. It can confirm the guidance is current. It treats Derek's white paper as more credible.
The effect compounds across multiple claims. A ten-page white paper with five substantiated claims, each with proper citations and metadata, signals authority far more clearly than the same paper with citations buried in footnotes or embedded as plain text.
For answer engines, the difference is even sharper. When an AI Overviews system answers a user's question about tax-loss harvesting, it pulls from multiple sources. It prioritizes sources where the evidence is visible and verifiable. An answer engine will more readily cite Derek's white paper if the citations are structured and transparent than if they are hidden in the formatting.
Compliance and discovery become the same act. The evidence you collect for regulators becomes the evidence that helps algorithms recognize your authority. You do not need to maintain two separate documents or two separate processes. You structure once, and both audiences benefit.
Getting Started Without Overhaul
You do not need to redesign your entire site to implement evidence architecture. Start with your highest-value content. These are the pages that receive the most traffic, or the claims that are most central to your firm's positioning.
For an RIA, that often means your performance pages, your research or insights section, and your regulatory disclosures
