A Prompt for Making Writing Sound Less Like AI

Published on November 19, 2026

A Prompt for Making Writing Sound Less Like AI

I find that newer language models write better code, but their prose can be harder to follow. Even when each sentence reads well, a model may break up a short explanation with headings or repeat the point it just made with an example.
These models have also gotten better at following editing instructions. Here’s the prompt I use to revise articles on this site.

Using the prompt

A request like “make this sound natural” leaves the model to guess what needs work. The rules I use ask it to cut repeated explanations, give reasons for judgments like “better,” and preserve both what a claim says and the conditions under which it holds.
To use the English editing rules, send this prompt along with your draft.
Read https://imoz.jp/humanizer.en.txt carefully, then repeatedly review and revise every part until it fully complies with the rules.
For a Japanese draft, this version points to the Japanese editing rules.
https://imoz.jp/humanizer.ja.txt を熟読し、ルールに完全に適合するように隅々まで何度も確認し修正してください
To see how the rules affected repetition and the connections between ideas, I gave GPT-6.1 Sol three drafts in English and three in Japanese. Each draft went through three passes with the prompt for its language. On the second and third passes, the model revised its previous output. Each pair below shows the original draft and the model’s third revision, quoted verbatim.

English examples

Why a $10 bill has value

The original introduces trust, government support, and ease of trade, gives each a paragraph, and lists all three again at the end.
Before
Why does a $10 banknote have value? After all, it is just a piece of paper that costs very little to produce. The answer lies in trust, government support, and usefulness.
First, money has value because people trust it. When you receive a $10 banknote, you believe that someone else will accept it in exchange for goods or services. For example, you can use it to buy lunch because the restaurant owner knows they can spend it elsewhere.
Second, governments and central banks help maintain money's value. Governments recognize money as legal tender, while central banks manage the supply of money and work to keep prices stable. This helps people feel confident that their money will remain useful over time.
Finally, money makes trade much easier. Without money, you might have to exchange something you own directly for food, clothing, or other necessities. Finding someone willing to make that exchange could be difficult.
Ultimately, a $10 banknote has value not because of the paper itself, but because people trust it, institutions support it, and everyone can use it to buy things.
After (three passes)
Why does a $10 banknote have value when it costs so little to produce? People accept it because they trust that others will accept it too. You can use it to buy lunch because the restaurant owner expects to spend it elsewhere.
This shared acceptance makes trade easier. Without money, you might have to find a restaurant owner who wants something you can offer in exchange for a meal. Money lets you sell to one person and buy from another.
Governments and central banks help maintain that trust. Governments recognize currency as legal tender for settling debts, while central banks manage the money supply and work to keep prices stable. Stable prices help people judge what their money will buy over time.
The revision moves barter ahead of the paragraph on governments and central banks and stays with the restaurant example. The owner takes your $10 because they expect to spend it elsewhere. To barter for the same lunch, you’d have to offer something the owner wanted.
The model drops “First,” “Second,” and “Finally,” along with the recap that begins “Ultimately.” The paragraph on institutions keeps “legal tender” and adds “for settling debts.”

An email about shorter meetings

The original email explains the goal of running more efficient meetings both at the start and near the end. Each of the four changes has its own bold label.
Before
Subject: Changes to Recurring Team Meetings – Starting Next Month
Hi Team,
Starting next month, we will be introducing a new format for our recurring team meetings to make them more efficient, focused, and productive.
Here’s what will change:
  • Shorter meetings: All recurring team meetings will be reduced from 60 minutes to 30 minutes.
  • Progress updates in advance: Everyone should record their progress, key updates, and any blockers in our shared document no later than the day before each meeting.
  • Discussion-focused meetings: Rather than spending time on routine status updates, meetings will focus on discussing important issues, resolving blockers, and making decisions.
  • Clear decision points: Each agenda owner should include a one-line statement outlining the specific decision that needs to be made during the meeting.
One-month trial period
We will test this new approach for one month and gather feedback from the team to assess its effectiveness. Based on your input, we may make further adjustments to improve the process.
Our goal is to reduce meeting time while ensuring meaningful collaboration and timely decisions.
Thank you for your cooperation and support in making this change a success!
After (three passes)
Subject: Shorter recurring team meetings starting next month
Hi team,
Starting next month, we’ll run a one-month trial of a new format for all recurring team meetings. The goal is to spend less time in meetings while still working through issues together and making timely decisions.
  • Meetings will last 30 minutes instead of 60.
  • Please record your progress, key updates, and any blockers in our shared document by the day before each meeting.
  • We’ll use meetings to discuss important issues, resolve blockers, and make decisions. Routine status updates will go in the shared document.
  • Each agenda owner should include one line stating the specific decision needed at the meeting.
We’ll gather your feedback to see how well the format works and decide whether to adjust it.
Thanks for your help with the trial.
The revised subject tells readers that meetings will be shorter, and the first sentence says the one-month trial starts next month. The model keeps all four bullets and drops their bold labels. The change from 60 minutes to 30 is clear without “Shorter meetings” at the start of the line.
“Everyone should record” becomes “Please record,” a direct request to post an update in the shared document by the day before each meeting. Agenda owners still have a separate bullet asking them to state in one line what the team needs to decide. The email closes with one sentence about using feedback to decide whether to adjust the format and a short thank-you.

A stale price in an online store

The original uses a price cut and a sold-out item to show what shoppers might see when cached data gets out of date. Its advice about updating or deleting a cache entry comes later, in general terms.
Before
Web application caching is a technique used to make websites load faster by temporarily storing frequently requested data. Instead of fetching the same information from a database every time, a web application can reuse previously stored data, called a cache.
For example, imagine an online store displaying a product listing page with product names, images, prices, and stock levels. Without caching, the server must query the database whenever a customer visits the page. With caching, the server can quickly retrieve stored product information, reducing database requests and improving page loading speed.
Caching also reduces server load, allowing the website to handle more visitors efficiently, especially during busy shopping periods.
However, cached information can become outdated. For example, if a product's price changes from $50 to $40, customers might still see the old price. Similarly, a sold-out product might incorrectly appear available.
To prevent this, developers use cache invalidation, which means updating or deleting outdated cached data. When product prices or stock levels change, the application can refresh the relevant cache entries or remove them so that the next request retrieves updated information.
This keeps the website fast while ensuring customers receive accurate product details.
After (three passes)
Caching can help web applications load pages faster by reusing frequently requested data. The application keeps copies in a cache, where it can retrieve them without querying the database each time.
For example, an online store’s product listing page might display product names, images, prices, and stock levels. Without caching, the server queries the database for that information whenever a customer visits the page. With caching, it can store the results and reuse them for later visits. This reduces database work and can help the store handle more visitors during busy shopping periods.
Cached information can become outdated when the underlying data changes. If a product’s price drops from $50 to $40, customers might still see the old price. A sold-out product might also appear available if the cache still holds its previous stock level.
When prices or stock levels change, the application can update the affected cache entries or invalidate them by removing them. For example, it can replace the cached $50 price with $40. If it removes the price entry instead, the application must fetch the current price from the database the next time that price is requested. It can then cache the result for later requests.
In the revised stock example, a sold-out item can still look available because the cache holds the previous stock count.
The model follows the price cut through both fixes. The application can replace the cached $50 with $40. If it deletes the entry instead, the next request for the price sends the application back to the database. Both drafts mention that lookup. The revision goes one step further by showing that the application can cache the price it just fetched for later visits.

Japanese examples

I used separate Japanese drafts on the same three subjects.

The value of a 10,000-yen note

Trust appears in the opening sentence of the original and again at the top of a three-item list. A later formula sums up the value of money as trust plus a way to use it. The revision drops the list and the formula and brings the 10,000-yen note forward. Someone accepts the note because they expect to spend it later. The next paragraph gives two reasons for that trust. Governments and central banks manage the currency, and people can pay their taxes with it.
The revision also spells out why money makes trade easier. A barter partner must have what you want and want what you have. With money, you can sell to one person and buy from another.
The last sentence changes the strength of a claim. In the original, the currency loses value if people lose trust in it. In the revision, its value can fall, in which case the same amount buys less.
Before
通貨(お金)が価値を持つ最大の理由は、「みんながそれに価値があると信じ、受け取ってくれるから」です。
主に3つの理由があります。
  1. 信用があるから
    みんなが「1,000円で商品を買える」と信じているため、お金として使えます。
  2. 国が制度を支えているから
    政府や中央銀行が通貨を管理し、税金の支払いなどにも使えるようにしています。
  3. 交換に便利だから
    お金があれば、物々交換をしなくても、商品やサービスを自由に交換できます。
例えば
1万円札は、ただの紙です。紙そのものに1万円の価値はありません。
しかし、社会全体が「この紙で1万円分の買い物ができる」と認めているため、価値を持ちます。
つまり、
通貨の価値 = 社会の信用 + 使える仕組み
ということです。
そして重要なのは、通貨の価値は絶対的なものではなく、人々の信用が失われれば下がるという点です。
After (three passes)
通貨(お金)が価値を持つのは、支払いに使えると人々が信じ、受け取ってくれるからです。
例えば、1万円札の紙そのものに1万円の価値はありません。それでも買い物に使えるのは、受け取った人も、そのお札で別の商品やサービスを買えると信じているからです。
この信用は、政府や中央銀行による通貨の管理や、税金をその通貨で納められる制度に支えられています。
取引のしやすさも、お金が広く使われる理由です。物々交換では、自分の持ち物を欲しがり、こちらが欲しい物を持っている相手を探す必要があります。お金を使えば、物を売る相手と商品やサービスを買う相手を別々に選べます。
通貨への信用が失われると、その価値が下がり、同じ金額で買える商品やサービスの量が減ることがあります。

A notice about shorter meetings

This fictional staff notice gives the change from 60 minutes to 30 in both the opening and the list. The revision gives the new length once, then orders the list from preparation to discussion. Participants should post updates in the shared document by the day before the meeting, and agenda owners should add one line stating what they want the group to decide. The last bullet turns to the meeting itself, where the team will use those updates to focus on issues that need discussion or a decision.
The notice still calls for a one-month trial. Where the original looks forward to more fruitful meetings and asks for understanding and cooperation, the revision ends with a specific request for help preparing.
Before
皆さま
いつも定例会議へのご協力ありがとうございます。
来月より、会議の効率化と意思決定の迅速化を目的として、定例会議の進め方を変更します。
これまで60分としていた会議時間を30分に短縮し、進捗報告を中心とした形式から、相談や意思決定に集中する形式へ切り替えます。
具体的な運用方法は以下のとおりです。
・会議時間: 60分から30分に短縮します。
・進捗共有: 各自、会議の前日までに共有ドキュメントへ記入してください。
・会議で扱う内容: 事前に共有された進捗を踏まえ、相談事項や意思決定が必要な議題を優先します。
・議題の準備: 各議題の担当者は、内容に加えて「会議で決めたいこと」を1行で明記してください。
この運用は、まず1か月間の試行とします。試行期間中に出た課題や皆さまからのご意見を踏まえ、必要に応じて進め方を見直します。
限られた時間を有効に活用し、より実りある会議にしていきたいと考えています。事前準備などでご協力をお願いすることになりますが、ご理解とご協力のほど、よろしくお願いいたします。
After (three passes)
皆さま
いつも定例会議へのご協力ありがとうございます。
来月から、定例会議の時間を60分から30分に短縮します。会議を効率よく進め、意思決定を早めるため、進捗報告を中心とした形式から、相談や意思決定に集中する形式へ切り替えます。
  • 各自、会議の前日までに共有ドキュメントへ進捗を記入してください。
  • 各議題の担当者は、議題の内容に加えて「会議で決めたいこと」を1行で明記してください。
  • 会議では、事前に共有された進捗を踏まえ、相談事項や意思決定が必要な議題を優先して話し合います。
まずは1か月間試行します。試行中に出た課題や皆さまからのご意見を踏まえ、必要に応じて進め方を見直します。
事前準備へのご協力をよろしくお願いいたします。

A cached product listing

The original goes straight from querying the database to reusing saved product details. The revision explains where that copy comes from. If the server caches the names, prices, and images it just fetched, it can reuse them the next time someone opens the listing. Skipping the database query reduces the server’s workload and can help the page load faster.
The model drops the invitation to imagine a store and the general warning about caching. The paragraph on stale data now begins with the problem a shopper might see. If a price or stock level changes, the listing may still show the old value from the cache. The revision also cuts the closing reminder to balance speed and accuracy, leaving readers with two remedies the original already gave. The application can refresh the cache after a set interval or delete the affected entry when the data changes.
Before
Webアプリケーションのキャッシュとは、一度取得したデータを一時的に保存し、次回以降に再利用する仕組みです。毎回同じデータを取得する必要がなくなるため、画面の表示を速くできます。
例えば、ネットショップの商品一覧ページを考えてみましょう。キャッシュがない場合、ページを開くたびにサーバーがデータベースから商品名や価格、画像などの情報を取得します。一方、キャッシュがあれば、保存済みの商品情報を使えるため、データベースへのアクセスが減り、表示速度が向上します。また、サーバーの負担を軽減する効果もあります。
ただし、キャッシュには注意点があります。商品の価格や在庫数が変更されても、古い情報が保存されていると、実際とは異なる内容が表示される可能性があります。そのため、一定時間が経過したらキャッシュを更新したり、価格や在庫が変更されたタイミングで古いキャッシュを削除したりする工夫が必要です。
キャッシュは、表示速度と情報の正確さのバランスを考えて利用することが大切です。
After (three passes)
Webアプリケーションのキャッシュとは、一度取得したデータを一時的に保存し、次回以降に再利用する仕組みです。毎回同じデータを取得し直さずに済むため、画面の表示を速くできます。
例えば、ネットショップの商品一覧ページでキャッシュを使わない場合、ページを開くたびにサーバーがデータベースから商品名や価格、画像などの情報を取得します。取得した商品情報をキャッシュに保存しておけば、次回以降はその情報を使えます。データベースへのアクセスが減るため、表示が速くなり、サーバーの負担も軽くなります。
ただし、商品の価格や在庫数が変わっても、キャッシュに古い情報が残っていると、変更前の内容が表示されることがあります。そのため、一定時間が経過したらキャッシュを更新したり、価格や在庫数が変わったときに古いキャッシュを削除したりする必要があります。

Closing thoughts

In both languages, the model cut repeated explanations and formulaic lead-ins. I also found the revisions easier to follow where they spelled out a reason or a step. The money examples show why it can be hard to find someone to barter with. In the Japanese store example, the server saves product data for reuse. In the English one, deleting a cached price sends the application back to the database the next time someone requests that price.
I still have to decide what I want readers to take away. That’s how I choose where to go into detail and what I can leave out. When the model chooses the emphasis, I tend to get a little about every related topic, all in the same cautious tone.
For the store example, you might want to show how a cache saves the server from looking up the same product for every shopper. Or you might want someone who sets prices to understand why a customer can still see the old price. When you send your draft, tell the model which point you care about and why.