ACCURACY, ETHICS AND LEGAL RISK

Every document you sign your name to makes a quiet promise: that you checked. Most professional writing failures are not failures of style. They are failures of verification, of proportion, or of nerve — a number copied without a denominator, a patient detail left in a “de-identified” case, a performance note that says more than the writer could prove, an email that accidentally forms a contract. This chapter treats accuracy as a craft with steps you can practice, not a virtue you either have or lack. You will learn to sort claims by the proof they require, to check a source in five minutes, to report numbers without lying by arithmetic, to strip identifiers from a real case, to spot the sentences that create legal exposure, and to write the correction that repairs damage instead of compounding it. Legal material here is United States–centered, general, and not legal advice; rules vary by state, sector and employer.

Section 1What You Can Actually Prove

Four kinds of statements, four kinds of proof

Writers get into trouble when they treat every sentence as the same species. A staffing report might contain, in a single paragraph, a hard fact, a modeled estimate, a causal inference and a personal judgment — each demanding a different standard of support, each failing in a different way. Sorting them takes about thirty seconds and prevents most accuracy disasters.

A verifiable fact is something a competent stranger could confirm from a record: a date, a dose, a dollar figure, a policy number. An estimate is a number produced by a method; it is only as good as the method, so the method has to travel with it. An inference is a conclusion you drew from facts — the leap is yours, and the reader deserves to see the leap. An opinion is a judgment about value or priority; it cannot be true or false, but it can be well or poorly grounded.

The practical test is simple. Ask: If someone challenged this sentence tomorrow, what would I show them? If the answer is “the invoice,” you have a fact. If it is “my spreadsheet,” you have an estimate and should say so. If it is “well, it seemed obvious,” you have an inference wearing a fact’s clothing, and it needs to be relabeled before it goes out.

Checking a claim in five minutes

Most of us verify badly by default: we stare harder at the page in front of us, judging credibility by design quality, tone and confidence. Research on how professional fact-checkers work — notably studies from the Stanford History Education Group — found that skilled checkers do the opposite. They leave the page almost immediately and read laterally, opening other tabs to find out who produced the claim and what other sources say. Staying on the page is how you get fooled by a good template.

Mike Caulfield’s SIFT routine turns that into four moves you can do between meetings. Stop: notice that you are about to repeat something you have not checked. Investigate the source: spend sixty seconds finding out who they are and what they sell. Find better coverage: look for a stronger source saying the same thing. Trace claims to the original: follow the citation chain back to the study, statute or dataset rather than trusting the summary of the summary.

Tracing is where most errors surface. A vendor’s white paper cites a news article, which cites a press release, which describes a study that measured something narrower than everyone downstream assumed. When you finally open the study you discover the sample was 42 volunteers over six weeks. Nothing was fabricated; the claim simply grew as it traveled. Your job as the last writer in the chain is to stop the growth.

Numbers are claims too

Numbers feel objective, which is exactly why they are such efficient vehicles for error. Four failures account for most of it. Missing denominators: “80% of respondents preferred the new form” is meaningless if five people answered. Confusing percent with percentage points: a rise from 4% to 6% is a two-percentage-point increase and a 50% relative increase, and the second phrasing is technically true and rhetorically loaded. Mixing counts with rates: a unit with more falls may simply have more patients. And false precision: reporting 62.47% when your data cannot support more than two significant figures implies a measurement you never made.

The clearest fix is to give the reader both frames when they differ. If an intervention drops an adverse event rate from 2.0% to 1.0%, write both: a 1.0 percentage-point absolute reduction, a 50% relative reduction. Readers who see only the relative figure consistently overestimate the benefit — a pattern documented across health communication research and one reason regulators and journal editors increasingly ask for absolute numbers alongside relative ones.

Finally, carry the vintage of the number. “Current census is 28” is a claim about a moment. Write “census was 28 at 0700 on 14 March,” and the sentence stays true forever instead of decaying into a falsehood the next shift.

Section 2Honest Representation: Spin, Omission and Credit

The half-true sentence

Outright fabrication is rare in professional writing. What is common is the sentence that is literally accurate and functionally deceptive: true words arranged so the reader draws a false conclusion. “No safety concerns were reported” is true if nobody was asked. “Response times improved” is true if you improved from terrible to bad. “The committee reviewed the incident” is true if the review lasted four minutes and reached no conclusion.

The tell is usually a missing comparison, a missing timeframe, or a missing denominator — the same holes from Section 1, but this time left open on purpose or by convenience. A useful self-check before you send: What would a fair-minded reader assume from this sentence, and is that assumption true? If the answer is no, the sentence is dishonest regardless of how well it survives literal parsing.

Omission is the harder ethical case, because no document can include everything. The working standard most professions converge on is materiality: you must include what a reasonable reader would want in order to make the decision the document is asking them to make. Leaving out the vendor’s two failed pilots from a procurement memo is not editing for length. Leaving out the color of their logo is.

Calibrated language: hedging that is honest, not cowardly

Students are often told to write with confidence, and separately told never to overstate. Those instructions collide until you understand that hedges are not weakness words — they are precision instruments. Calibrated language means the strength of your verb matches the strength of your evidence. “Causes” is a strong claim requiring strong design. “Is associated with” describes a relationship without asserting direction. “May” signals genuine uncertainty. Using the strong verb when you have the weak evidence is the error; using the weak verb when you have solid proof is a different error, and it costs you credibility too.

There is real disagreement in the research literature about how hedging reads. Some studies of scientific and workplace writing find that hedged claims are perceived as more credible and more trustworthy; others find that heavily hedged prose reads as evasive, particularly to non-expert audiences. The practical resolution is not to hedge less but to hedge specifically: replace vague softeners (“somewhat,” “it appears that,” “arguably”) with a named limitation. “This estimate excludes travel costs” is a hedge that adds information. “This estimate is somewhat preliminary” is a hedge that adds fog.

Compare: “The data arguably suggest a possible relationship between staffing and satisfaction.” Four hedges, no content. Now: “Units with higher staffing scored higher on satisfaction, but the comparison is observational and covers one quarter, so it cannot show cause.” One hedge, fully specified, and the reader knows exactly what to do with it.

Giving credit: plagiarism, patchwriting, copyright, fair use

Plagiarism is presenting someone else’s words or ideas as your own. It is an ethical and professional offense, and in a workplace it is usually adjudicated by your employer or licensing board, not a court. Copyright infringement is a legal violation of the owner’s exclusive rights. The two overlap but are not the same, and the difference matters practically: attributing a source cures plagiarism, but it does not by itself give you permission to reproduce protected material. Attribution is not a license.

Patchwriting — a term from writing researcher Rebecca Moore Howard — is the most common form students and professionals fall into unintentionally. You keep the source’s sentence structure and swap in synonyms. Even with a citation, this is usually treated as plagiarism, because the sentence architecture is borrowed. The reliable fix is mechanical: close the source, write the idea from memory in your own syntax, then reopen the source to check the facts.

Fair use in United States copyright law is a four-factor balancing test — purpose and character of the use, nature of the work, amount used relative to the whole, and effect on the market. It is a defense evaluated case by case, not a checklist with safe quantities. There is no magic word count. Materials released under Creative Commons licenses are easier: CC BY requires attribution; NC restricts commercial use; SA requires you to share adaptations under the same terms. Read the license, name it, and link it.

Section 3Confidentiality, Privacy and the Duty to Say Less

What counts as confidential

Four categories cover nearly everything a working professional handles. Protected health information (PHI) is health information tied to an identifiable person, governed in the United States by the HIPAA Privacy Rule and enforced against covered entities and their business associates. Education records are covered by FERPA at institutions receiving federal education funding. Personal data — addresses, government identifiers, financial accounts — is covered by a patchwork of state and sector laws, and by comprehensive regimes such as the EU’s GDPR if your organization touches those jurisdictions. Proprietary and confidential business information is protected by contract: your employment agreement, NDAs, and vendor terms.

Two misconceptions cause most trouble. The first is that de-identification is a formality. It is a technical standard: the HIPAA safe-harbor method requires removing eighteen categories of identifier, including all elements of dates other than year, all geographic subdivisions smaller than a state (with a narrow zip-code exception), and any other unique identifying number, characteristic or code. The second is that HIPAA is the only rule in play. It is not. A student posting about a clinical day may be outside HIPAA’s direct reach in some configurations and still face program dismissal, board discipline, and a defamation or privacy claim from the person described.

The professional habit that prevents most incidents is not legal knowledge. It is a reflex: before including a detail, ask whether the document’s purpose actually requires it. That reflex has a name in health privacy law — the minimum necessary standard — but it is good practice in any field. Most confidentiality breaches are not thefts. They are enthusiasms.

De-identifying a real example

Here is a case vignette a student submitted for a class discussion board. Read it as a privacy problem, not a clinical one: “On 3/14 I cared for a 78 y/o male from Millbrook admitted to 4-West after a fall at his daughter’s wedding. He is a retired fire chief, has a pacemaker, and his wife is a nurse on our own unit.”

Nothing here is malicious, and everything here is a problem. The date is a full date. Millbrook is a geographic subdivision smaller than a state. “Retired fire chief” is a near-unique occupational identifier in a small town. The wedding fixes an event that people attended. And the wife-on-the-unit detail identifies the patient to every coworker who reads it. The educational content of the vignette — a fall, an older adult, an implanted device — survives entirely without any of it.

The revision: “I cared for an adult in their seventies admitted after a mechanical fall at a family event. The patient has an implanted cardiac device, which shaped our approach to imaging and telemetry.” Same teaching value. No identifiable person. Notice what the rewrite did: it generalized the age, removed the date and place, deleted the occupation, deleted the relationship, and kept only the clinically relevant device.

Channels, forwarding and the accidental disclosure

The document is only half the risk; the pipe matters as much. Three habits prevent the majority of accidental disclosures. First, check the recipient list before every send, and be suspicious of autocomplete — misdirected email is one of the most frequently reported causes of small-scale data breaches in organizational reporting. Second, never forward a thread without reading down to the bottom; confidential material three replies deep travels with the message. Third, do not move regulated information into unmanaged channels. A personal phone, a consumer messaging app, an unapproved AI assistant, or a home printer takes the data outside your organization’s controls, and in many cases outside its legal ability to account for it.

Attachments deserve their own paragraph. Spreadsheets carry hidden columns, filtered rows, second worksheets and revision history. Word files carry tracked changes, comments and author metadata. PDFs generated by drawing a black box over text often leave the text underneath fully selectable — a redaction failure that has embarrassed law firms and government agencies repeatedly. If a document must be redacted, use a tool that removes the content, then reopen the output and search it for the terms you thought you removed.

Finally, write the subject line as though it will be read on a lock screen in a crowded elevator, because it will be. “Re: Jane Doe pathology results” discloses before anyone opens anything. “Re: patient 4 — results available in the chart” does the same job with none of the exposure.

Section 4Where Everyday Writing Creates Legal Exposure

Defamation, references and performance notes

In broad United States terms, defamation requires a false statement of fact, communicated to someone other than the subject, made with the required degree of fault, causing harm. Truth — or substantial truth — is a defense. Public officials and public figures must additionally prove actual malice, meaning knowledge of falsity or reckless disregard for the truth — the standard set for public officials in New York Times Co. v. Sullivan (1964) and extended to public figures three years later in Curtis Publishing Co. v. Butts. Private individuals generally face a lower bar, typically negligence, and the specifics vary considerably by state.

The dangerous belief in workplaces is that labeling something an opinion immunizes it. The Supreme Court addressed this in Milkovich v. Lorain Journal Co. (1990), declining to recognize a blanket constitutional privilege for statements framed as opinion. What matters is whether the statement implies undisclosed, provably false facts. “In my opinion, Terrell is a thief” implies you know of a theft. That is why performance documentation should describe observed behavior and its date rather than assign character.

Employment references sit in a particular zone. Many states recognize a qualified privilege for good-faith references given to those with a legitimate interest, and a number have enacted reference-immunity statutes; the protection typically evaporates if the statement is knowingly false or made with malice. Practically, this argues for the same discipline you would use in a performance note: state what is documented, stay within what you personally observed or what the record shows, and route the request through whatever channel your employer has designated.

Words that make promises

An email can form a contract. Under the federal E-SIGN Act and the Uniform Electronic Transactions Act adopted in most states, electronic records and signatures are not denied legal effect merely because they are electronic, and courts have in various cases treated an email exchange — sometimes even a signature block — as satisfying a signed-writing requirement. Whether any particular exchange forms a binding agreement depends on the facts and the jurisdiction, but the practical lesson is settled: casual language about price, scope and timing in negotiation email carries real weight.

Product and service claims carry a parallel risk. Under Article 2 of the Uniform Commercial Code, an affirmation of fact or a description of goods that becomes part of the basis of the bargain can create an express warranty. Vague sales enthusiasm — puffery — generally does not. “Best-in-class support” is puffery; “guaranteed four-hour response, twenty-four hours a day” is a promise your operations team now has to keep. And the FTC expects advertisers to have a reasonable basis, or substantiation, for objective claims before making them, with health-related claims typically requiring competent and reliable scientific evidence. Its Endorsement Guides also require clear disclosure of material connections between an endorser and a brand.

Internal documents make promises too. Courts in some states have found that employee handbooks and written policies can create implied contractual obligations, which is why well-drafted handbooks carry explicit at-will and no-contract disclaimers. When you write a policy or procedure, you are writing a standard someone can later be measured against. Write “staff should verify identity using two identifiers” and you have created a benchmark; write “staff will always verify” and you have created an absolute your organization will be asked to prove it met, every time.

Everything you write may be read later

In litigation, parties exchange relevant documents through discovery, and electronically stored information — email, chat, drafts, metadata, texts — is squarely included. The mental model to adopt is simple: assume a stranger with an opposing interest will read every message, out of order, without context, years from now. That is not paranoia; it is a description of how a deposition exhibit binder is assembled.

Once litigation is reasonably anticipated, organizations have a duty to preserve relevant material — a litigation hold. Destroying or altering material after that duty attaches is spoliation, and under the federal rules governing electronically stored information, courts can impose serious sanctions — including, where a party acted with intent to deprive its opponent of the material, instructing a jury to assume the lost material was unfavorable. Separately, federal law criminalizes destroying records with intent to obstruct an investigation. The practical rule is unglamorous and absolute: when a hold notice arrives, stop all routine deletion in its scope immediately, including auto-purge rules you set up and forgot.

The mirror-image error is hoarding. A records retention schedule exists because keeping everything forever is also a liability — more to breach, more to produce, more to contradict you. Follow the schedule your organization publishes, in both directions, and do not improvise. If you find yourself reasoning about what to delete based on how a document might look, stop and call someone. That instinct is the beginning of a much worse problem than the document.

Section 5When Something Goes Wrong: Correction, Escalation, Repair

The correction ladder

Not every error deserves the same response, and treating a typo like a scandal is its own failure of judgment. Match the response to two things: how far the document traveled, and whether the error could change a decision. A misspelled name in a draft nobody read gets fixed silently. A wrong dose in a distributed protocol gets a notification to every recipient, today, with the corrected document attached.

A good correction has five parts and takes ninety seconds to write: what was wrong, where it appeared, what is correct, what the reader should do now, and what you changed to prevent recurrence. What it does not have is a defense. The instinct to bury the fix inside an explanation of how the error was understandable is the single most damaging habit in this genre, because it moves the reader’s attention from the fact to your feelings about the fact.

Compare a weak and a strong version. Weak: “Hi all — there may have been some confusion around the figures in Tuesday’s memo, which was put together quickly given the deadline. Please disregard anything that doesn’t look right.” Strong: “Correction to Tuesday’s staffing memo. The overtime figure on page 2 read 498 hours; the correct figure is 398. The 61% increase should read 28%. A corrected memo is attached; please delete the earlier version. The error came from a transposed cell, and I have added a second check before distribution.” The second version is shorter, harder to write, and immeasurably better.

Raising a concern in writing

Ethical problems in organizations are rarely resolved by a single dramatic memo. They are resolved — when they are — by a documented, escalating, unemotional record. The structure that works is the same one that works for any persuasive short document: observation, standard, gap, ask.

A model: “On 14 and 21 June, the vendor invoices I processed (numbers 4471 and 4488) each included a $2,400 line item labeled ‘consulting’ with no supporting statement of work. Our purchasing policy, section 4.2, requires a signed SOW before payment of consulting charges. I have held both invoices. Could you confirm whether an SOW exists, or advise how you would like me to proceed? I want to make sure I am following the policy correctly.” Notice the absence of accusation, adjectives and inference. It states dates, documents, the written standard, the gap, and a request. It is nearly impossible to characterize as insubordinate, and nearly impossible to ignore.

Keep the escalation sequence and your own copies aligned with policy. Go to your direct supervisor first unless the concern involves them. Use the compliance hotline, ethics office, or incident system your organization provides — these channels frequently carry procedural protections that a personal email does not. Whistleblower protections exist under a range of federal and state statutes and vary enormously by sector, employer and the kind of concern raised; do not assume you are covered, and if the stakes are high, get advice before you act. Retaining documents you were not authorized to take can create problems even when your underlying concern is entirely valid.

Apology, accountability and the paper trail

Three sentences separate a repair from a further injury. Name what happened in plain language. Say what you are doing about it. Do not explain why it was reasonable. “I sent the report without the safety appendix; the complete version is attached, and I have added the appendix to the pre-send checklist” ends the incident. “Due to an unusually compressed timeline and multiple competing priorities, the appendix was inadvertently omitted” extends it, because the reader now has to decide whether to argue with you.

Institutional apologies carry additional considerations, and honest instructors should tell you the evidence is contested. Many United States states have enacted apology or benevolent-gesture statutes making certain expressions of sympathy inadmissible in civil proceedings, though their scope varies sharply — some cover only sympathy and not admissions of fault. Whether these laws reduce claims is genuinely unsettled: some studies of disclosure-and-offer programs report fewer claims and lower costs, while other analyses have found apology laws associated with no reduction or even increases in liability for some specialties. Where an event involves potential liability, your organization almost certainly has a disclosure process. Use it rather than improvising.

Related, and equally counterintuitive: disclosure is not a cure-all. Behavioral research by Cain, Loewenstein and Moore found that disclosing a conflict of interest can sometimes make advice worse, because disclosure appears to morally license the adviser to be more biased while listeners fail to discount enough. That does not argue against disclosing — disclose, always, and early. It argues that disclosure is a floor, not a resolution. When a conflict is serious, the right move is often to step back from the decision entirely, and to write down that you did.

Words worth knowing