What the research supports, and what gets repeated anyway
One row per piece of common writing advice, labeled as a published standard, one researcher's finding, one practitioner's rule or folklore, with what the source actually says and what it does not.
Writing advice travels without its source. A figure from a 1997 study of web pages turns up as a rule for letters; a government guide's "writing experts recommend" turns up as a law. This page puts each piece of advice next to the source it came from, says what kind of source that is, and says what the source does not show. Every source was read on November 13, 2026, and each one is listed with its date in the resource library.
The four labels
| Label | What it means here | What it can tell you |
|---|---|---|
| Published guidance | Written by a government agency or another body that publishes advice for a kind of document, such as the federal plain-language guides on Digital.gov or one agency's guidance on comments. | What that body asks for, and in some cases what it will act on. Guidance is rarely a measurement; it says what to do, not how much difference it makes. |
| One researcher's finding | A study or a review of studies, with a method, a set of participants and a date. | What happened when those people were measured doing that task, at that time. It does not say what your reader will do. |
| One practitioner's rule | A working rule from one writer, one writing center or this manual, based on practice rather than measurement. | A rule that has worked for the people who wrote it down. Useful as a habit; unmeasured. |
| Folklore | Advice repeated widely for which we found no source in the form it is usually stated. | Nothing on its own. It may hold in some cases; nobody we read has shown it. |
A label describes the evidence behind a piece of advice, not whether the advice is good. "Folklore" means we found no source for the advice as it is usually stated, among the sources read for this manual on November 13, 2026. It does not mean the advice is wrong in every case, and a source we did not read may exist. No row predicts how a particular reader will decide on a particular letter. The research rows rest on abstracts and records, not full papers, as the research table says row by row.
The advice, row by row
Filter by label, or type a word such as "page" or "reason" into the box. Quotations are the source's own words; everything else in the third column is our paraphrase.
| The advice as people repeat it | Label | What the source actually says | What it does not show | Source |
|---|---|---|---|---|
| "Put the ask first." See The shape of a letter |
Published guidance | Digital.gov: "Put the most important information at the beginning." For a process, it says to give the steps in chronological order. | How much difference it makes to a decision. The guide gives no line count; "the ask in the first three lines" is this manual's working form of the rule. | Digital.gov (GSA), Organize the information |
| "Keep paragraphs under 150 words." See The shape of a letter and Cutting it down |
Published guidance, relaying an unnamed recommendation | Digital.gov: "Writing experts recommend" paragraphs "of no more than 150 words in three to eight sentences", with 250 words as the ceiling. It adds: "There is nothing wrong with an occasional one-sentence paragraph." | Which experts, or any study behind the figure. It is a recommendation the guide passes on, not a measured threshold. The draft check flags paragraphs over 150 and over 250 words for that reason, and calls neither an error. | Digital.gov (GSA), Clear and short |
| "Keep every sentence under a set number of words." See Cutting it down |
Folklore, as a number | The federal guide we read gives no sentence-length figure. Its advice is one idea per sentence, and breaking a complex sentence into parts. | Any particular word limit. A long sentence that holds one idea clearly is not an error by the guide's standard. The draft check reports the average and the longest sentence and claims no threshold. | Digital.gov (GSA), Clear and short |
| "People scan, they don't read." See The shape of a letter |
One researcher's finding: a usability firm's 1997 study | Nielsen reported that "79 percent of our test users always scanned any new page", and only 16 percent read word by word. | Anything about letters. It measured web pages, in 1997, and the article states no sample size. A claims reviewer working through a queue of appeals is reading a different document for a different reason. It is a reason to put the ask first, not a figure to repeat. | Nielsen Norman Group, How Users Read on the Web (September 30, 1997) |
| "Adults read about 238 words a minute." See Cutting it down |
One researcher's finding: a meta-analysis (a study that pools the results of many studies) | Brysbaert estimated average silent reading for adults in English at 238 words per minute for non-fiction and 260 for fiction, from 190 studies and 18,573 participants. Reading aloud: 183 words per minute, from 77 studies and 5,965 participants. | How fast your reader reads, or whether they read every word. It is an average over many readers. The draft check divides by 238 for a reading time, as a sanity check, not a promise. | Brysbaert (2019), record and abstract (OpenAlex) |
| "Give a reason and people will say yes." See Asking for something |
One researcher's finding: three field experiments, 1978 | Langer, Blank and Chanowitz tested whether people notice placebic information, a reason that adds no information. The abstract says "The predictions were confirmed for both oral and written communications." | That a reason-shaped sentence carries a large ask, or moves a reader who decides by a policy. The experiments concerned small requests. The record we read gives no participant count, and its abstract was reconstructed, so its wording may differ slightly from the published one. | Langer, Blank and Chanowitz (1978), record and abstract (OpenAlex) |
| "Get a small yes first, and the big yes follows." See Asking for something |
One researcher's finding, and a later review | Freedman and Fraser ran two experiments on whether someone who agrees to a small request is more likely to agree to a larger one: the first with the same person making both requests, the second with different people. Burger's 1999 review names six processes proposed to explain the effect: self-perception, reactance, conformity, consistency, attributions and commitment. | A rule for letters. The foot-in-the-door literature (the name for this small-then-large sequence) is about two requests in a row; a single letter to a decider is one request. Neither record we read concerns a written appeal or complaint. | Freedman and Fraser (1966), record and abstract; Burger (1999), record (both OpenAlex) |
| "Always keep it to one page." See The shape of a letter and Proposing something to a group |
Folklore | No source we read measures what length gets read or acted on. | That a second page goes unread. This manual uses one page as a working limit that keeps you cutting, and says it is ours. When the facts will not fit, they go in an attachment or a numbered list, not smaller margins. | None found |
| "Emotion persuades; facts don't." See Tone and Proposing something to a group |
Folklore, as usually stated | Nothing in our sources supports it as a rule for a letter to someone who decides. The one agency guidance we read on public comments asks for "sound reasoning, scientific evidence, and/or how you will be impacted" (EPA). | That feeling has no place. EPA's list includes how the proposal affects you, next to reasoning and evidence. | None found for the claim; EPA, Commenting on EPA Dockets, for comments |
| "Never apologize." See Tone |
Folklore | No source we read addresses it. | Anything. This manual's own rule is narrower: an apology has a job in a letter that repairs something you got wrong, and usually no job in a request, an appeal or a complaint. That rule is ours. | None found |
| "Write at an 8th-grade reading level." See Cutting it down |
Published guidance, which says the opposite of how the scores are usually used | Digital.gov: "Only write for 8th graders if your audience is, in fact, an 8th-grade class." It also warns against writing for an 8th-grade class when your readers are PhD candidates, small business owners or working parents. | That a grade-level score (a formula that estimates reading difficulty from word and sentence length) measures clarity. The guide sets no grade target. Write for the reader you have. | Digital.gov (GSA), Principles of plain language |
| "Use the active voice." See Cutting it down |
Published guidance | Digital.gov: "Use the active voice", with "The company polluted the lake." set against "The lake was polluted by the company." | That every passive sentence is wrong. The guide gives no count or ratio. | Digital.gov (GSA), Writing for understanding |
| "Vary your words so it doesn't get repetitive." See Writing instructions and Cutting it down |
Published guidance, which says the opposite | Digital.gov: "You don't need to use synonyms to make your writing more interesting." Different words "can decrease clarity", it says. The same page names "absolutely", "really" and "very" as modifiers to cut. | A ban on any particular word. It is about clarity: one name for one thing, so the reader does not wonder whether "the form" and "the application" are two documents. | Digital.gov (GSA), Short and simple |
| "An accommodation request only counts if it's in writing and uses the legal words." See Asking for something |
Published guidance, which says the opposite | The EEOC: "Requests for reasonable accommodation do not need to be in writing." A person "may use 'plain English' and need not mention the ADA". An employer may ask for a form or a written request but "cannot ignore the initial request." | That writing is pointless. A written request dates what you asked and records it, which is why this manual recommends one. The guidance was issued October 17, 2002, before the ADA Amendments Act of 2008, and the page says so. | EEOC, Enforcement Guidance on Reasonable Accommodation and Undue Hardship under the ADA |
| "A public comment campaign wins on numbers." See Proposing something to a group |
Published guidance (one agency's) | EPA: "The comment process is not a vote". It says one well-supported comment is often more informative to the agency than a thousand form letters, and that mass-mail campaigns are posted as one sample plus a count. | How other agencies or local bodies weigh comments. It is EPA's guidance about EPA's dockets. | EPA, Commenting on EPA Dockets (last updated October 1, 2026) |
| "You need a big test group to know whether writing is clear." See Writing instructions and Cutting it down |
Published guidance | Digital.gov names three to five people for a usability test and 6 to 9 interviews for paraphrase testing. Its paraphrase-testing page says testing documents that go to a small number of people "probably isn't worth the time". | A method for one letter. This manual's read-back test, one reader and one question, is our scaled-down version for a single letter, and we say it is ours. | Digital.gov (GSA), Usability testing and Paraphrase testing |
| "A reference should say they're the best you've ever worked with." See Praise, references and reviews |
One practitioner's rule: one writing center | UNC's Writing Center says unsupported superlatives "are likely to meet with skepticism, unless the writer includes specific evidence". It asks the writer to say how long they have known the person, and in what capacity. | A measured effect. It is one university writing center's guidance, written for faculty recommending students, applied here to everyday references. | The Writing Center, UNC Chapel Hill, Letters of Recommendation |
| "If someone asks you for a reference, you have to say yes." See Praise, references and reviews |
One practitioner's rule: one writing center, saying the opposite | UNC's Writing Center says it is fine to decline when you could not write a positive or detailed letter, and lists reasons: you know too little about the person, the field is not yours, you lack the credentials, or there is not enough time. | How the person will take it. It is guidance on whether to write, not on how to say no. | The Writing Center, UNC Chapel Hill, Letters of Recommendation |
| "Put it in writing and you're protected." See Sending it, and after |
Folklore | Nothing we read says a letter preserves a legal right by itself. USPS describes Certified Mail as "Prove you sent it", with delivery information and, with Return Receipt, a signature. | What was in the envelope, or that sending it settled anything. A written record proves what you sent and when, if you kept a copy. Whether a right is preserved is a lawyer's question. | None found for the claim; USPS, Insurance & Extra Services, for what proof of mailing covers |
| "Always appeal; most appeals succeed." See Appealing a decision |
Folklore | No figure we read covers appeals in general. The agency pages give levels and deadlines, not odds. | Your chances. This manual states deadlines and routes, in Where it goes next, and no success rate. | None found |
How to read a claim
When a piece of writing advice comes with a number or "studies show", five questions sort it in a minute. Ask them in order, and stop at the first one you cannot answer.
- Who measured it, or who wrote it down? A government guide, a researcher, a firm that sells usability work, a writing center or nobody. The answer is the label.
- What did they measure? How people read a web page, words read per minute, whether someone agreed to a small favor. A figure about one task says little about a different task.
- On whom? Test users, people asked a favor, 18,573 readers pooled from 190 studies. A decision-maker reading a letter about their own rules is rarely in the sample.
- When? 1966, 1978, 1997 and 2019 are the research dates on this page. A date tells you what a study could have measured: a 1997 study of web pages measured the web of 1997. The 1966 experiment was followed by a body of studies, which Burger reviewed in 1999; that is why the review is cited beside it.
- For which kind of document? A web page, a spoken request, a comment to an agency, an academic reference. If it is not a letter asking someone to decide, the finding is a reason to try something, not a rule.
Worked through once. "People don't read, they scan." Who: one usability firm. What: how test users read new web pages. On whom: test users, number not stated. When: 1997. Which document: web pages. Every answer is clear, and none of them is "a letter to someone who has to decide". So the claim supports putting the ask at the top, which the federal guide says anyway, and supports nothing about how long your letter may be.
A sixth question applies to this page in particular: how much of the source did we read? For the research rows, the answer is the abstract or the database record, not the full paper. The next table says which.
The research sources, and how much of each we read
| Source | Year | What was measured | On whom | What we read |
|---|---|---|---|---|
| Brysbaert, Journal of Memory and Language | 2019 | Silent and oral reading rate, pooled across studies | 18,573 participants in 190 studies (silent); 5,965 in 77 studies (oral) | The abstract, through the OpenAlex record; not the full text |
| Nielsen, Nielsen Norman Group | 1997 | How test users read web pages | Test users; the article states no sample size | The article |
| Langer, Blank and Chanowitz, Journal of Personality and Social Psychology | 1978 | Whether people notice a reason with no information in it | Three field experiments; no participant count in the record | An abstract reconstructed by OpenAlex; wording may differ slightly from the published abstract |
| Freedman and Fraser, Journal of Personality and Social Psychology | 1966 | Agreement to a larger request after a smaller one | Two experiments; no participant count in the record | An abstract reconstructed by OpenAlex |
| Burger, Personality and Social Psychology Review | 1999 | A review of the foot-in-the-door studies and the explanations offered for them | A narrative review, with no sample of its own | The OpenAlex record |
As far as the records we read show, none of the five studied a letter to someone who decides by rules: a claims reviewer, a landlord's office, a board. That is the reader this manual is written for. Where the research is silent, the manual relies on the published guidance above and on its own rules, labeled below.
Where this manual's own rules sit
Some advice in this manual has no outside source. We label it where it appears, and list it here so nobody mistakes it for a finding.
- Preference One page as a working limit. This manual's rule, to keep you cutting. No source we read measures what length gets read. See The shape of a letter.
- Preference The ask in the first three lines. This manual's working form of the published rule to put the most important information first.
- Preference The absolute-word list in the draft check (always, never, cannot, must not, everyone, nobody). This site's own editorial rule. The draft check says the list is ours.
- Preference The hedge list beyond Digital.gov's three words. Digital.gov names "absolutely", "really" and "very"; the rest of the list on the draft check ("just", "I think", "maybe", "sorry to bother" and the others printed there) is ours.
- Preference Read it aloud once, and have one reader say back the ask. A scaled-down version of the published testing methods, sized for one letter. See Cutting it down.
- Preference Apology belongs in a repair, and rarely anywhere else. This manual's rule. See Tone.
If one of these rules does not fit your reader, the reader wins. A rule that is ours can be set aside by you; a deadline in your letter cannot.