Textual Disclosure Credibility: How to Test Filing Language

Textual disclosure credibility is the degree to which the narrative portions of a public filing (MD&A, risk factors, forward-looking statements, earnings-call remarks) can be corroborated by the numbers, controls and subsequent events that surround them. It is not a property of how a document reads. A fluent, readable, confident filing can be low in credibility, and a dense, legalistic one can be high. The test is whether each claim can be traced to evidence, and whether the language moves when the business moves.

This article sets out the construct, the textual signals that research associates with weaker credibility, the limits of readability metrics, and a repeatable workflow for institutional investors, governance advocates, proxy advisors, regulators and financial journalists.

Lacunaindex

Test Claims Against Company Delivery

Lacuna Index evaluates public companies by comparing disclosure claims with delivery, using evidence from public records.

Visit Lacuna Index

TL;DR

Credibility is established by corroboration against underlying data and independent evidence, not by tone, length or readability scores.
Research on readability is mixed: plain language does not always inform more, and the effect depends on the type of risk being described.
The most useful textual signals are changes: boilerplate drift, tone-versus-fundamentals gaps, and shifts in how a company reports (quantitative to qualitative).
A defensible assessment applies one standardized procedure across issuers and logs the evidence behind every score.
Sector-level benchmarks are a sound starting point before committing to a company-level forensic read.

1. What the construct measures
2. Why readability alone is an unreliable proxy
3. Textual signals associated with weaker credibility
4. A workflow for testing a filing
5. Limits and controls
6. Perspective: why a standardized, auditable approach matters
7. Where benchmarks and company-level analysis fit
FAQ
Sources

1. What the construct measures

Three distinct properties are often conflated under the label of disclosure quality.

Property	Question asked	Typical evidence
Readability	Can the intended reader parse the text?	Fog and Flesch scores, sentence length, jargon density
Informativeness	Does the text carry company-specific information?	Specificity, quantification, novelty relative to prior filings
Credibility	Is the text corroborated by what the company later does and reports?	Reconciliation to financial data, controls, outcomes, audit evidence

Readability and informativeness are properties of the text alone. Credibility is relational: it exists only between a statement and something external to it. That is why a purely linguistic model cannot settle the question, and why the more rigorous approaches pair text with a second data source.

The auditing profession offers the clearest borrowed vocabulary. Under PCAOB AS 1105 on audit evidence, management assertions about existence, completeness, valuation, rights and obligations, and presentation and disclosure are supported by sufficient, appropriate evidence, and reliability is highest when evidence comes directly from independent external sources. An outside analyst cannot replicate an audit, but the same ordering of evidence applies: a claim corroborated by an external or third-party record carries more weight than one corroborated only by further management statements.

The PCAOB's top-down approach under AS 2201 also has an analytical parallel. Auditors concentrate on entity-level controls and on accounts and disclosures with a reasonable possibility of material misstatement, which in practice links narrative disclosures to the financial data beneath them. An analyst can adopt the same priority rule: test first the statements that, if false, would change the investment or governance conclusion.

2. Why readability alone is an unreliable proxy

A large body of academic work, including the studies surfaced in search results for this topic, relates the readability of 10-K narratives to outcomes such as liquidity, payout policy and managerial ability. These studies are informative about correlations. They do not establish that simpler text is more credible.

The regulatory position is itself nuanced. The SEC's plain English handbook frames the goal as clarity rather than brevity: avoiding legalistic, overly complex presentation and boilerplate that obscures substance. Item 105 of Regulation S-K, as summarized in a Bloomberg Law drafting guide, requires risk factors in plain English that concisely explain how each risk affects the registrant, organized logically under relevant headings.

Experimental evidence complicates the assumption that plainer is better. A 2018 experiment with 359 nonprofessional investors found that readability significantly influenced risk judgments, but in inconsistent directions depending on the risk factor: legalese was more effective at conveying company-specific risks such as internal control weaknesses, while plain English raised the perceived risk of industry-wide factors such as competition. The implication for practitioners is that a readability score is a covariate to be recorded, not a verdict.

A separate study reviewed in the risk-disclosure text mining literature asks directly whether risk factors are as informative as the SEC intends, using the subprime crisis as a real-risk event to test them. The question is the right one: informativeness is tested against an event, not against a style guide.

3. Textual signals associated with weaker credibility

The following signals are drawn from the research record and from practitioner analysis. Each is a prompt for further testing, not proof of misstatement.

3.1 Boilerplate drift

Boilerplate drift describes risk-factor language that stays static while the business environment changes. Analysis of narrative manipulation and risk disclosure gaming in 10-K filings treats it as one marker that sophisticated participants look for. The diagnostic is a year-over-year diff: identical paragraphs across years in which the company entered new markets, changed its financing, or faced new litigation indicate disclosure that is not tracking the business. For a worked U.S. example, see the analysis of how half of COVID-era risk language was boilerplate.

3.2 Tone-fundamentals gaps

A tone-fundamentals gap arises when the sentiment of the MD&A diverges materially from contemporaneous performance metrics or audit-related disclosures. Optimistic language alongside deteriorating margins, cash conversion or backlog is the classic form. The related method is covered in sentiment versus performance analysis of SEC filings.

Tone can also be managed over time. One doctoral study of corporate disclosures reports that CEOs tend to manage tone downward in their early years of tenure relative to later years, with the pattern more pronounced in high-litigation industries. Tenure and litigation exposure are therefore sensible control variables when tone is compared across filings.

3.3 Impression-management vocabulary

A study of regulatory enforcement cases and MD&A text, In Companies We Trust: Cheap Talk or Genuine Virtue?, finds that firms using more trust-related words were more likely to operate in so-called sin industries and to use higher frequencies of buzzwords, and reports that such language can be negatively correlated with firm performance and predictive of higher future monetary penalties. The relevant reading is probabilistic: the vocabulary raises the prior that impression management is present, and raises the priority of the corroboration step.

3.4 Changes in reporting form

Research analysts treat forward-looking statements as a critical test of narrative credibility. A manager who shifts between quantitative and qualitative reporting in response to uncertainty, rather than holding a consistent format, may be signaling degraded credibility, according to a study discussed in an analyst-focused paper. A company that gave numeric ranges for three quarters and then moved to adjectives has changed what can be verified, and the change itself is a data point.

3.5 Added and removed risk factors

Practitioner guidance on reading a 10-K by comparing years holds that a newly added risk factor often reflects a board-level decision to formalize a previously informal exposure, whereas a removed disclosure can indicate an effort to manage litigation risk rather than a real reduction in exposure. The same source cautions that a diff identifies what changed, not whether the change matters; that judgment must be recorded.

3.6 Data-layer inconsistencies

Credibility also has a technical layer. Commentary on common 10-K and 10-Q filing mistakes notes that inconsistent data tagging and overlooked accounting updates can invite SEC comment letters and amendments. Analysts increasingly treat XBRL validation anomalies, such as a negative value where a positive is expected, as possible indicators of a weak internal control environment, and verify them manually against the human-readable filing. A single anomaly is rarely meaningful; a recurring pattern across periods is.

4. A workflow for testing a filing

The sequence below is intended to be applied identically across issuers, so that results are comparable.

Inventory the claims. Extract discrete, checkable statements from the MD&A, risk factors, proxy statement, press releases and the last four earnings calls: targets, timelines, capital commitments, control assertions.
Classify each claim as quantified or qualitative, near-term or long-dated, within management's control or dependent on external conditions.
Diff the text. Compare risk factors and MD&A year over year, and flag unchanged language, additions, deletions and shifts in reporting form.
Corroborate against independent records. Reconcile to reported financials, later filings, regulatory records and, where available, third-party data. Rank evidence by independence, following the logic of AS 1105.
Score the gap. Record, for each claim, whether it was delivered, partly delivered, abandoned without disclosure, or not yet testable.
Log the reasoning. Retain the passage, the contradicting or confirming record and the scoring decision, so a second reviewer can reproduce the result.
Benchmark. Place the result against peers in the same sector, because disclosure norms and litigation exposure differ by industry.

The broader practice of verification is treated in how to validate corporate claims in public disclosures, and the pattern-recognition step in how to spot engineered disclosure patterns in filings. Journalists tracking commitments across years may prefer the promise-tracking methods guide.

Worked illustration: a stable risk factor in a changing business

Consider a hypothetical issuer, constructed for illustration only. In year one it states that customer concentration is a risk, with no figures. In year three, the largest customer has grown from a minority share to the majority of revenue per the segment note, yet the risk factor is word-for-word identical. The text has not moved while the exposure has. The finding is not that the company lied; it is that the narrative no longer describes the risk the numbers show, which lowers the credibility attributed to the remainder of the section until further corroboration is performed.

Unchanged filing language beside rising customer concentration

5. Limits and controls

Textual methods carry known failure modes, and a credible assessment states them.

Dictionary mismatch. Word lists built for one context can misclassify terms in another; a word that is negative in general usage may be neutral in financial reporting.
Industry effects. Boilerplate prevalence, litigation exposure and the density of technical language vary by sector, so cross-sector comparison of raw scores misleads.
Machine-generated text. Generative tools can produce fluent, specific-sounding prose, so fluency is a weaker signal than it was. Recent work on detecting narrative manipulation with generative AI addresses the detection side of this problem for U.S. 10-K filings.
Legal necessity. Some repetition is mandated or prudent. Repetition is a flag only when the underlying facts have changed.
Non-testable claims. Aspirational statements with no date or metric cannot be corroborated and should be classified as such rather than scored as failures.

Internal signals sometimes lie outside the public record entirely. Employees who believe a disclosure is materially misleading have legal channels and protections; a general overview is available from an employment law firm's page on whistleblower protection. A forensic reading of public records neither replaces those channels nor relies on insider information.

For investors, the same discipline of separating narrative from measured results applies to other parts of the process; one illustration is this discussion of portfolio analytics metrics that reveal manager skill, where reported outcomes are likewise tested against measurable evidence.

6. Perspective: why a standardized, auditable approach matters

Our experience is that the central difficulty in assessing textual credibility is not detection of any single red flag. It is comparability. A reviewer who reads one filing closely and another loosely produces conclusions that cannot be set side by side, and the resulting differences mostly measure the reviewer's attention, not the issuers' behavior.

For that reason we hold to three commitments. First, only public records: earnings calls, SEC filings, press releases, proxy statements and public statements, with no insider access and no proprietary data, so that any reader can inspect the same material. Second, a single methodology applied across all covered companies, so that an execution score for one issuer means the same thing as for another. Third, explicit governance logs for scoring, so that each assessment is audit-traceable from conclusion back to passage.

We also treat the gap between narrative claims and actual delivery as the quantity of interest, not language quality as such. Where a company's valuation embeds claims that its record does not support, the market may be mispricing engineered disclosure and questions that were never asked. Our archetypes (earned, borrowed, undervalued) combine delivery evidence with valuation, which is why a well-written filing does not by itself earn a favorable classification. The full procedure is documented in the methodology walkthrough.

— Glen

7. Where benchmarks and company-level analysis fit

Sector context should precede a company-level read. Lacuna Index publishes free sector benchmarks, including Financials, Health Care, Industrials and Consumer Staples, which show how covered companies in a sector compare on delivery against narrative. Per-company forensic reports, with execution and narrative scores and the supporting evidence, are available to subscribers through per-company analysis. The Cohort Pulse and Sector Sweep plans are described on the pricing page; pricing is not published there.

FAQ

What is credibility in accounting?

In an audit context, credibility is established when management assertions are corroborated by sufficient, appropriate evidence, as defined in PCAOB AS 1105. Evidence obtained directly by the auditor from independent external sources is the most reliable. For narrative disclosures, an outside analyst applies the same principle by ranking corroborating records by their independence from management.

Does a more readable 10-K mean a more credible one?

Not necessarily. Experimental research found that readability influenced investors' risk judgments in inconsistent directions, depending on whether the risk was company-specific or industry-wide. Readability is better treated as one recorded attribute than as a credibility score.

How can the credibility of forward-looking statements be tested?

Extract each statement with a metric and a date, then check later filings and calls for delivery, revision or silent abandonment. Also note whether the company changed from quantitative to qualitative reporting during periods of uncertainty, since that shift reduces what can be verified.

Can credibility be assessed without insider information?

Yes, to a defined degree. Public filings, call transcripts, proxy statements and press releases allow claims to be reconciled with reported results and later events. The method cannot reveal what was never disclosed, but it can show where disclosures and delivery diverge.