Contents
Early in the development of The Architecture of Nonfiction, I described the value of an experienced editor this way:
Before
A good copy editor serves the same function as the senior engineer reviewing a pull request.
A pull request is a proposed code change submitted for review before it becomes part of the shared codebase. The comparison was meant to describe the scale of the review, not equate the two professions.
The need was real. I wanted another informed person to challenge the document’s purpose, structure, evidence, sequence, dependencies, and effect on the reader. Automated grammar tools could find malformed sentences. They could not tell me that a section answered the wrong question.
The professional role was wrong.
Editors Canada’s professional standards and the Editorial Freelancers Association’s service definitions assign that big-picture work to structural, substantive, or developmental editing. Copyediting has a different responsibility. Even the role’s spelling needed correction: copy editor is two words. This essay follows the Editorial Freelancers Association’s one-word spelling for copyediting; Editors Canada uses copy editing.
The parent essay now says:
After
A strong structural or developmental editor behaves more like an experienced engineer reviewing a pull request.
That change did more than replace a job title. It clarified the kind of review the document needed.
Technical practitioners often ask someone to “edit” a draft without defining the risk, stage, or authority involved. That is like requesting a code review without saying whether you need an architectural assessment, a security review, confirmation that existing users and integrations will still work, or proof that the implementation matches the product requirement.
You may receive useful feedback. You may also receive the wrong review at exactly the wrong time.
Editing Is Not One Job
Editors Canada defines editing broadly as reviewing, correcting, improving, and suggesting changes so material works for its audience, medium, and purpose. Its 2024 standards then separate that work into four stages: structural editing, stylistic editing, copy editing, and proofreading.
The Editorial Freelancers Association uses overlapping language. It groups developmental, substantive, structural, and content editing around big-picture content and organization. It describes line editing at the sentence and paragraph level, copyediting around correctness and consistency, and proofreading near the end of production.
The labels vary. The responsibilities remain distinguishable.
The first four rows below summarize those professional editorial responsibilities. Automated checks are a separate supporting function included for comparison.
| Review | Dominant responsibility |
|---|---|
| Structural or developmental editing | Content, organization, argumentative coverage, sequence, audience, medium, and purpose |
| Stylistic or line editing | Meaning, coherence, flow, voice, and language at paragraph and sentence level |
| Copyediting | Correctness, consistency, accuracy, citations, terminology, cross-references, and expected supporting elements in a stable draft |
| Proofreading | Errors, formatting, visual elements, and final-change verification in the intended medium |
| Automated checks | Declared patterns, mechanical errors, and repeatable consistency rules |
These are working boundaries, not walls. Editors Canada says stylistic editing often happens within a structural edit or copy edit. Proofreading may be combined with another stage when time or budget requires it. A single editor may perform several kinds of work across a project.
The author still needs to know which responsibility is active. “This person can do all four” does not make all four appropriate at the same moment.
A Linter Applies Known Rules
In software, a linter examines code for known patterns and reports suspected violations. It can flag an unused variable, a prohibited construction, inconsistent formatting, or a rule the team encoded in configuration. It performs that work consistently and can be run after every change.
Writing tools can serve the same function. A spellchecker catches unfamiliar words. A link checker finds a missing page. A style scanner flags an em dash, a repeated phrase, or terminology that violates a maintained standard. A narrow AI audit can search for the same declared patterns across a draft.
That work matters. Repeatable checks keep an editor from spending expensive attention on defects a machine can identify cheaply.
The limitation is responsibility. A linter applies rules it has been given. It does not decide whether the document asks the right question, contains the evidence its conclusion requires, or exposes the context the reader needs. It can report that a sentence violates a preferred pattern without knowing whether the exception performs necessary rhetorical work.
A draft can pass every automated check and still fail its reader.
A Reviewer Questions the Design
Google’s engineering practices place overall design first in its code-review guidance. Reviewers are asked whether the parts of the change interact sensibly, whether the change belongs in the system, whether it serves users, and whether it introduces unnecessary complexity. Correct syntax and passing tests do not answer those questions.
The same scale of inquiry belongs in structural editorial review:
- Does this document answer the question its intended reader came to resolve?
- Does this section belong here, somewhere else, or nowhere in this piece?
- Does the reader receive the context each conclusion requires?
- Do the sources support the statements that depend upon them?
- Is the argument more complicated than the subject requires?
- Would a different organization make the reasoning easier to follow?
Those questions can lead to deletion, movement, division, expansion, or complete reconstruction. They can also confirm that the architecture already works and should survive later edits.
The comparison stops there. A structural editor is not a senior engineer. Editors and software reviewers work under different obligations, authority, and methods.
The useful parallel is the scale and kind of question. An experienced reviewer examines whether the complete change belongs and works before debating every line that implements it.
A Copy Editor Still Uses Judgment
Calling automated tools linters does not make a copy editor the human version of one.
Editors Canada’s copy-editing standards begin with grammar, punctuation, spelling, and usage, but they continue into consistency, source documentation, terminology, accuracy, cross-references, missing information, and places where citations are required. A professional copy editor must decide when a variation is intentional, when an assertion needs support, when a quotation requires verification, and when a missing element should be queried rather than supplied.
Those decisions require context. The publication’s house style matters. The author’s purpose matters. The intended reader, subject, medium, and authority granted to the editor matter.
A copyedit does not automatically include comprehensive fact-checking or subject-matter review. If the project requires independent verification, define which facts, quotations, calculations, or sources the editor must check and what specialist knowledge the work requires.
Proofreading also requires more than spotting typos. Editors Canada places it after textual and visual elements have been composed and asks the proofreader to work in the intended medium when possible. The proofreader verifies that requested changes were implemented, that those changes did not create new problems, and that layout, navigation, tables, visual elements, and metadata still work together.
The linter comparison identifies the responsibility of automated checks. It does not set the ceiling on professional editing.
Choose Review by Risk
The familiar service name is not always the service your draft needs. Start with the failure that would matter most if the piece shipped in its current form.
| Dominant risk | Review to request |
|---|---|
| Wrong question, missing argument or evidence that may change the content or organization, weak organization, or broken argument | Structural or developmental review |
| Meaning is present but sentences or paragraphs do not communicate it effectively | Stylistic or line edit |
| Correctness, consistency, citations, terminology, cross-references, labels, or expected supporting elements may be missing from an otherwise stable draft | Copyedit |
| Final layout, links, visual elements, or implemented changes may contain errors | Proofread |
| Known mechanical rules need repeatable application | Automated check |
A draft can carry several risks at once. A white paper may need its argument challenged, its paragraphs refined, its citations checked, and its final PDF proofread. The table tells you where to begin, not where every project must end.
Both structural editors and copy editors can find omissions. The deciding condition is the consequence. If the missing material may change the argument, content, or organization, begin with structural review. If the argument is stable and the missing item is a citation, cross-reference, label, or other expected supporting element, assign it to copyediting.
The stage matters as much as the risk. Structural review is valuable while the structure can still change. Proofreading is valuable when the content and layout are stable enough that a last-stage correction will not trigger another rewrite.
Do not ask a final proofreader to rescue an argument the structural review never examined.
Give the Reviewer a Contract
“Please edit this” does not define a useful assignment. Neither does “tell me what you think.” The reviewer needs enough information to evaluate the draft against its intended result.
A compact review brief can provide it:
Reader: Who is this for, and what may they already know?
Outcome: What should they understand, decide, or do afterward?
Publication: Where will this appear, and what constraints apply?
Stage: Is the structure still open, or is this a final proof?
Risk: What kind of failure should the reviewer prioritize?
Authority: May the reviewer rewrite, restructure, comment only, or flag decisions for the author?
Evidence: Which statements are sourced facts, first-party observations, or inferences? What remains uncertain, and what does the evidence not support?
Deliverable: Should the reviewer return comments, an editorial letter, tracked changes, a revised file, or a prioritized report?
This brief is not bureaucracy. It prevents a reviewer from spending an hour correcting sentences the author expected to replace. It also prevents the author from treating a structural recommendation as unwanted scope when the assignment never said the structure was closed.
The reviewer can still discover a risk outside the requested scope. The contract tells them whether to fix it, flag it, or recommend another pass.
Review from the Largest Decision Down
Nancy Sommers’s research compared the revision practices of college freshmen and experienced adult writers. The less experienced writers concentrated on words and phrases. The experienced writers reconsidered the shape of the argument and made changes through addition, deletion, substitution, and reordering at several levels.
That distinction explains why review order matters. A default editorial sequence moves from the largest unresolved decisions toward the final composed artifact:
- Structural or developmental review.
- Author revision.
- Stylistic or line review.
- Copyediting.
- Proofreading in the intended medium.
Each stage should receive a draft stable enough for its responsibility. The structural editor needs permission to challenge the outline and content. The copy editor needs an argument stable enough that consistency work will survive. The proofreader needs the composed page, screen, or document whose final behavior the reader will encounter.
Run applicable automated checks before a human handoff and rerun them after substantive changes. Checks for final links, metadata, or composed output must wait until those elements exist. None replaces a proofreader’s inspection of the intended medium.
Reversing the sequence creates waste and risk. A careful copyedit can perfect paragraphs that structural revision later deletes. A late reorganization can break cross-references that were already verified. A new argument introduced during proofreading can escape the research, stylistic, and copy checks applied to the rest of the work.
Small articles and constrained projects may combine stages. The order of decisions still matters even when one person performs them in one engagement. Resolve whether the section belongs before polishing every sentence inside it.
Feedback Should State Why It Matters
Review comments become harder to use when every note looks equally mandatory. A missing citation, a confused reader question, and a preference for another adjective do not require the same response.
Felicity Brand’s practitioner guide to prose review separates corrections, questions, and suggestions. Her model also advises reading the whole piece before marking it and explaining why a change is being proposed. We can extend those labels for an editorial workflow:
| Label | Meaning |
|---|---|
| Correction | An external rule, fact, or requirement is violated, and a fix is required |
| Question | The reviewer lacks context or cannot follow the reasoning |
| Recommendation | The reader’s outcome is at risk, so the issue requires a response even if the proposed remedy changes |
| Suggestion | A defensible improvement remains the author’s choice |
| Praise | A working decision should survive later revision |
Recommendation and praise are additions to the model used here. The first distinguishes fitness-for-purpose problems from discretionary improvements. The second protects strong choices from being removed during a later pass that sees only defects.
The reason behind a comment matters more than the proposed replacement. “Move this paragraph” is an implementation. “The evidence arrives three sections after the conclusion that depends upon it” is a diagnosis the author can evaluate.
Good feedback tells the author which kind of problem the reviewer experienced and why it affects the reader.
AI Can Review More Than Mechanics
A general AI system is capable of more than linter-like checking. The level of review depends heavily on the context, responsibility, and degree of freedom supplied in the request.
Compare these assignments:
Mechanical review
Correct grammar and punctuation. Report violations of the supplied style rules. Do not change structure or meaning.
Structural review
Review this document for the declared reader and outcome. Identify sections that are missing, misplaced, duplicated, or unnecessary. Explain the reader consequence. Do not rewrite until the structural findings are resolved.
The first request deliberately constrains the AI to local, declared problems. The second allows it to question whether components should remain and how they compose. Neither request guarantees good judgment, but they ask for different work.
Using the same AI collaboration throughout a project can create an independence problem. A collaborator that helped research, outline, and draft the piece may share the author’s assumptions because those assumptions are present in the same working context. It can understand why a phrase made sense to the author and therefore miss that the phrase will not make sense to anyone else.
Use a cold-start review when independence matters. Give a fresh human or AI reviewer only the reader-facing draft, intended audience, desired outcome, declared prerequisites, and applicable public standards. Questions that the reviewer cannot answer from that material expose dependencies the authoring context may have hidden.
AI can make broader review inexpensive and repeatable. An independent reader still contributes something different: an actual encounter with the draft outside the environment that produced it.
The Author Owns the Merge
An editor reports how the document works for another informed reader. The editor may also propose new wording, a different sequence, a missing explanation, or a smaller scope. Those implementations are valuable options, not automatic commands.
You do not have to accept the patch to accept the defect.
In editorial terms, you can accept the problem without accepting the proposed edit.
An editor may correctly identify confusion and propose a sentence that does not sound like you. A reviewer may detect that a section breaks the argument and recommend deletion when movement would preserve useful material. The diagnosis can be right while the first remedy is wrong.
The opposite risk matters too. Accepting every suggested change can flatten voice, remove intentional rhetoric, or introduce a factual error. The editor brings informed reader experience. The author retains accountability for the purpose, evidence, tradeoffs, and published result.
Respond to the reason behind the feedback. Then inspect every component that depended on the changed material. Moving a definition affects the first paragraph that uses it. Deleting an example may leave a later reference without an antecedent. Replacing evidence may require a different conclusion.
The author owns the merge, the decision to incorporate a proposed change, and the effects that follow it.
Build a Review Pipeline
A useful review process does not begin by choosing a tool or job title. It begins by naming the document’s present risk.
- State the document contract. Name the reader, outcome, publication, scope, and evidence boundaries.
- Identify the dominant risk. Decide whether the unresolved problem calls for structural editing, stylistic editing, copyediting, proofreading, or an automated check.
- Choose the matching review. Select the person or tool whose responsibility fits that risk.
- Define the assignment. Give the reviewer the stage, authority, priorities, and desired deliverable.
- Resolve large decisions first. Settle purpose, content, and organization before investing in smaller corrections.
- Evaluate feedback by reason. Accept, reject, or adapt the implementation without losing the diagnosis.
- Recheck affected relationships. Verify the context, references, evidence, and transitions that depended on the change.
- Proofread the composed result. Inspect the final medium after the text and visual elements are in place.
Automated checks apply standards the process already knows how to state. Editors contribute judgment about the document, the reader, and the risks no rule has captured yet.
A mature publishing process needs both functions and gives each one work it can actually perform.
Receipts
- Professional editorial stages: Editors Canada’s Professional Editorial Standards 2024 defines editing in relation to audience, medium, and purpose and distinguishes structural editing, stylistic editing, copy editing, and proofreading. Its detailed standards establish the responsibilities and overlap described in this essay.
- Editorial service terminology: The Editorial Freelancers Association’s “Editorial Service Definitions” describes developmental editing as big-picture content and organization work, line editing as sentence- and paragraph-level language work, copyediting as correctness and consistency work, and proofreading as a late production stage. Terminology and service combinations vary across professionals and projects.
- Code-review responsibilities: Google’s “What to look for in a code review” places overall design first and directs reviewers to consider functionality, users, integration, complexity, tests, naming, comments, style, and documentation. Applying that scale of review to writing is this project’s analogy.
- Revision beyond wording: Nancy Sommers’s “Revision Strategies of Student Writers and Experienced Adult Writers” distinguishes word- and phrase-level changes from experienced revision that reconsiders the argument through addition, deletion, substitution, and reordering. The study concerns writers’ revision practices, not professional editorial roles.
- Developer-led prose review: Felicity Brand’s “From code reviews to prose reviews” advises reviewers to read the whole piece first, advocate for the reader, explain the reasons behind feedback, and distinguish corrections, questions, and suggestions. The recommendation and praise labels in this essay extend that practitioner model.
- First-party terminology correction: The before-and-after opening records a real correction made during the development of The Architecture of Nonfiction. It demonstrates how professional standards refined this project’s analogy and terminology.