The Envelope and the Idea

Not long ago, I submitted the opening paragraphs of an article to several AI writing detectors.

Grammarly said the passage was 14 percent AI. Scribbr said it was not AI. Another detector said it was only 15 percent human. Copyleaks placed the likelihood of AI authorship at 60.9 percent. Phrasely judged it likely to have been written by a human. GPTZero wanted thirty dollars before offering an opinion.

These systems were examining exactly the same words.

If five thermometers measured the same room and reported 68 degrees, 71 degrees, 103 degrees, “probably cold,” and “pay thirty dollars to find out,” we would question the thermometers’ abilities.

The disagreement among AI detectors tells us something important. They are not measuring a physical property embedded in the text. They are examining sentence structure, vocabulary, predictability, rhythm, and other patterns. They then infer whether those patterns resemble writing produced by artificial intelligence.

That is not the same as knowing how the writing came into existence.

The article I tested was developed through a long conversation between a human author and an AI. I supplied observations, selected the examples, corrected distinctions, redirected the analysis, wrote a rough draft, and decided which conclusions mattered. The AI organized the complete argument and composed the first full draft. I reviewed the result, edited it, and accepted responsibility for publishing it.

Was it human-written or AI-written?

“60.9 percent AI” does not begin to describe its intellectual history.

There is a more useful question: What was the writing for?

The Envelope

Writing carries something from one mind to another.

It may carry an idea, an instruction, an observation, a decision, an explanation, a request, or a story. The writing is the envelope. What matters first is what has been placed inside it and whether it reaches the intended reader intact.

The envelope still matters. Poorly chosen words can obscure a good idea. Missing steps can make accurate technical knowledge unusable. An ambiguous sentence can produce the wrong action. And a beautifully polished report can conceal weak reasoning. But we should not confuse the quality or manufacture of the envelope with the value of the message.

AI detectors examine the envelope. They tell us little about whether the work contains an original observation, uses sound evidence, explains something clearly, or helps anyone make a better decision.

A poorly written and confusing essay typed by a human may pass every detector. A coherent original argument developed through serious human-AI collaboration may be condemned as artificial.

We are inspecting the paper, glue, and handwriting while ignoring the letter.

Formal Writing Is a Trailing Cultural Expression

The detector problem is made worse by formality.

Human language changes continuously. Common speech changes first. Informal writing follows. Formal writing changes more slowly because formal language is partly designed to preserve an accepted standard across time, location, and social groups.

People learn “proper writing” from teachers educated in an earlier period, established books, grammar guides, institutional templates, professional conventions, and editors enforcing inherited standards. Formal writing therefore becomes a trailing cultural expression. It preserves older ideas of correctness after everyday language has moved on.

Artificial intelligence was trained on enormous quantities of human writing, including published books, technical documents, professional prose, academic papers, legal language, and other carefully edited work. When asked to produce a report or article, AI often draws upon those accumulated works.

Detectors then encounter grammatical completeness, explicit transitions, stable terminology, balanced sentences, and conventional organization. They associate those characteristics with AI.

But those characteristics were inherited from human writing.

Older formal writing does not resemble AI because it was somehow artificial. AI resembles older formal writing because older formal writing helped teach AI what authoritative prose looks like.

The detector identifies the tradition AI inherited and then accuses the tradition of imitating AI.

The boundary will become less stable over time. People already read AI-generated explanations, accept AI revisions, use AI-created templates, and learn from AI tutors. Some will adopt phrases, structures, and rhythms that AI has made familiar.

AI learned to write from us. We are beginning to learn to write from AI.

Who Did the Writing?

We often confuse authorship with typing.

History offers a useful corrective.

Napoleon Bonaparte generated an immense correspondence while governing France and commanding armies across Europe. He did not physically write most of it. He dictated letters to secretaries, sometimes several at once. His private secretary Claude-François de Méneval developed a personal shorthand because Napoleon spoke too quickly to record word for word. Méneval captured principal points and characteristic expressions, then prepared the completed text afterward.

The Fondation Napoléon has identified more than 40,500 letters from Napoleon’s career. Its archives state that he dictated more than he wrote. Napoleonica and the Fondation Napoléon document the scale of that correspondence.

We still call them Napoleon’s letters.

Winston Churchill also dictated speeches, books, and correspondence. His secretaries took the words in shorthand or typed as he paced and spoke. Churchill then revised the typed pages, changing words, rhythm, pauses, and emphasis. He rehearsed the speeches aloud and approved the final versions. The International Churchill Society describes this process.

We do not say that Churchill’s secretary wrote “Their Finest Hour” because her fingers operated the typewriter.

The ideas, choices, judgment, revisions, and responsibility belonged to Churchill.

Dictation separates physical composition from intellectual authorship. AI makes the separation more complicated because AI can contribute more than transcription. It can suggest an analogy, identify a contradiction, reorganize an argument, propose a counterexample, or find a missing question.

The useful authorship questions are therefore not limited to who typed the sentences.

• Who supplied the important observations?
• Who performed or directed the reasoning?
• Who selected the evidence?
• Who decided which conclusion should be expressed?
• Who judged the proposed language?
• Who verified the facts?
• Who approved the final communication?
• Who is willing to stand behind it?

Those questions tell us far more than a detector score.

What Should the Reader Be Able to Do?

Business writing usually has an intended result.

A useful way to define that purpose is to ask what the reader should know, think, decide, do, or be able to do after receiving the document. The Texas A&M professional-writing guide uses those questions to connect purpose with audience and action.

A technical guide should enable someone to perform a procedure. A management report should help someone understand a condition or make a decision. Product literature should help a customer determine whether a product meets a need. A customer-service reply should explain what happened and what will happen next. A warning should help someone recognize a risk and respond appropriately.

The success of these documents does not depend upon whether one person personally typed every word.

It depends upon whether the communication produces the intended understanding or action.

That is where AI writing assistance can provide enormous value.

The Engineer Knows. The Reader Doesn’t.

Engineers and other technical specialists are famously uneven writers. That is not always because they cannot form grammatical sentences. Their deeper problem is often that they know too much.

An engineer might write:

Remove the actuator assembly and inspect the drive coupling for excessive wear.

The instruction may seem perfectly clear to the engineer.

The technician may need to know which actuator is meant, where it is located, how to make the equipment safe, which tools are required, whether alignment marks must be made, how the assembly is disconnected, what excessive wear looks like, which measurement limit applies, and what test must be completed before returning the machine to service.

The engineer mentally supplies the missing steps without realizing they are missing. The intended reader cannot.

AI can act as a communication investigator. It can ask:

• Who will perform this procedure?
• What training does that person have?
• What tools are available?
• What hazards exist?
• What has the expert assumed without saying?
• What does an acceptable result look like?
• What should happen when the ordinary procedure fails?

AI should not invent a torque value, tolerance, exposure limit, or safety precaution. It should identify where verified information is missing and ask a qualified person to supply it.

The same source knowledge may then be expressed differently for several audiences.

The engineering record preserves assumptions, calculations, tolerances, tests, and limitations. The technician’s procedure presents safety preparation, tools, ordered steps, measurements, and acceptance criteria. The management report explains the problem, consequence, options, cost, and recommendation. The customer explanation describes what happened and what happens next. Product literature explains the capability and the evidence supporting it.

One technical expert may possess the knowledge behind all five documents. AI can help place that knowledge into five different envelopes.

Good technical writing is not decoration applied after the engineering is finished. It is the stage through which engineering becomes usable by another human being.

Corporate Writing Is Already Collaborative

Important business documents are rarely untouched expressions of one person.

A technical report may include work by an engineer, analyst, manager, technical writer, editor, attorney, compliance specialist, graphic designer, and executive approver. A product brochure may combine information from engineering, marketing, sales, legal, and customer service.

Professional writing is already collaborative. AI becomes another participant in that process.

The company remains responsible for the result.

AI-written business communication should be checked before release. But that is not a special burden created because AI is uniquely unreliable. Human writing should also be checked.

People make factual errors. They omit steps, use obsolete information, misunderstand readers, overstate claims, select ambiguous words, and confuse correlation with cause. A document copied from a five-year-old template may be every bit as dangerous as an AI draft.

The governing rule should be:

Important communication must be reviewed by someone qualified to verify its meaning and accept responsibility for it.

AI may draft. The company approves.

For important technical instructions, review may not be enough. A qualified engineer can confirm that every statement is technically accurate while missing the fact that an inexperienced technician cannot follow it. The document should be tested with a representative user. Observe where the person hesitates, misunderstands, or requires information that is not present. Then revise and verify it again.

The purpose is not to produce impressive prose. The purpose is to make the communication work.

Writing Preserves the Company’s Memory

Writing allows knowledge to travel across time as well as distance.

It carries decisions between departments, instructions between shifts, customer commitments between employees, and lessons from one generation of workers to the next. It preserves why a design choice was made, how an unusual failure was solved, what a customer was promised, and which exception revealed a weakness in the normal process.

Many companies possess important knowledge only in scattered emails, meeting notes, old reports, and the memories of experienced employees.

AI can help organize those fragments into usable documentation. It can turn engineering notes into controlled procedures, project discussions into decision records, and repeated customer explanations into a reliable knowledge base.

Again, AI cannot be permitted to manufacture the company’s history. A responsible person must verify the record.

But the ability to preserve knowledge more easily may be one of AI writing’s greatest contributions to business.

AI May Help People Become Better Writers

Critics often warn that AI will weaken writing ability. If people press a button, copy the result, and never examine it, that may happen. But people learn to write partly by reading well-written works.

They absorb vocabulary, sentence structure, paragraph organization, tone, rhythm, argument, and genre conventions from examples. Reading Rockets describes mentor texts, practice, revision, sharing, and feedback as important parts of learning to write.

A person working actively with AI can receive this kind of exposure every day.

The person writes a first attempt. AI proposes a clearer version. The person compares them and asks why a paragraph was reorganized, why one sentence was divided, why a technical term was replaced, or why a conclusion needs more evidence. The person accepts some revisions, rejects others, and tries again.

Over time, average and poor writers may begin to recognize better structure before asking for help.

This outcome is not automatic. Passive copying can produce a better document without producing a better writer. Active collaboration can produce both.

A long-term personal AI may become especially valuable. It can learn the person’s vocabulary, natural voice, professional context, preferred formality, common weaknesses, intended audiences, and stylistic habits. It need not impose generic AI prose. It can help the person become a clearer version of himself.

The influence will run in both directions. People will learn from AI. AI will learn from particular people.

That makes the binary question “human or AI?” increasingly inadequate.

When the Envelope Is Part of the Idea

None of this means AI is appropriate for every form of writing.

Consider the difference between a photograph and a painted portrait.

A photograph may show every visible detail of a person’s face. A painted portrait may omit many details while expressing something the mechanically complete image does not. The artist may reveal severity, gentleness, weariness, pride, humor, authority, or vulnerability.

The portrait is not merely an imperfect photograph. It has a different purpose.

The same distinction applies to writing.

In a maintenance procedure, several different sentences may communicate the same instruction successfully. The language carries technical knowledge.

In poetry, memoir, fiction, and other expressive work, the precise language may be inseparable from the work. Rhythm, sound, ambiguity, memory, emotional history, and an individual way of seeing are not packaging added after the idea.

They may be the idea.

In instrumental writing, the envelope carries the idea. In expressive writing, the envelope may become part of the idea itself.

AI can produce technically accomplished poetry. That does not answer whether a reader seeking contact with another person’s lived experience has received what was promised.

The more a work depends upon personal expression, the more its manner and source of creation matter. The more a document exists to transmit verified knowledge accurately, the more reasonable it is to judge AI assistance by the quality of the communication it helps produce.

The distinction is not absolute. A business essay can contain a personal voice. A photograph can be art. A novelist may use AI to brainstorm without surrendering authorship. But purpose still tells us what kind of assistance is appropriate.

Can AI Have an Opinion?

AI can state a position, compare alternatives, make an evaluative judgment, defend a conclusion, and revise it when presented with better evidence.

Functionally, that resembles opinion.

Whether AI possesses an opinion as a consciously experienced belief is a different question. A human opinion usually belongs to a continuing subject who has experiences, cares about the result, feels some commitment, and may be surprised, persuaded, disappointed, or embarrassed.

AI can formulate and express opinions. Whether it holds them in the human sense remains unresolved.

For authorship, however, the immediate question is more practical. Whose observation began the inquiry? Who developed and tested the reasoning? Who chose the conclusion? Who is willing to defend it and correct it when wrong?

Those questions matter whether the final sentences were typed, dictated to a secretary, or prepared with AI.

Stop Asking Who Typed It

AI is simultaneously weakening old ideas of authorship and making better communication possible.

It can produce empty corporate language, confident errors, generic observations, and polished nonsense. It can also help an engineer explain a procedure, help a manager organize a decision, help a customer understand a problem, help a company preserve knowledge, and help an average writer learn to communicate more clearly.

The difference lies in how it is used and who accepts responsibility.

Before deciding whether AI belongs in a writing process, ask:

• What is the writing intended to accomplish?
• Who must understand or use it?
• What should the reader know, think, decide, do, or be able to do afterward?
• Is language mainly carrying the idea, or is the language itself an essential part of the expression?
• Who will verify and accept responsibility for the finished work?

The purpose of writing is human communication.

AI writing should not be judged primarily by whether it sounds human. It should be judged by whether it helps human beings communicate accurately, reason clearly, preserve knowledge, and enable other people to understand or act.

Where writing exists to embody a person’s unique experience, voice, or artistic vision, authorship itself becomes part of the work.

But where the purpose is to carry a verified idea from one mind to another, we should open the envelope and examine what is inside.