Claude guide offers five hidden techniques to reduce AI‑written feel
This tutorial explains how to use Claude’s built‑in features to make its output sound less like generic AI text. The author suggests first storing a personal writing sample in Claude’s Memory so the model can adopt the writer’s tone automatically. Next, a list of forbidden phrases can be saved in a Project to stop the model from using common AI‑style wording.
What you can do with it
AI at Work →generates text and lets users store style preferences, forbidden phrases, and iterative editing settings
- Who for
- marketers, founders and operations
- Cost
- Cost not stated
- Effort
- minutes
Use it for
- store personal writing style for future prompts
- block generic AI phrases in drafts
- iteratively refine drafts with read‑aloud checks
Watch out requires careful prompt design; Claude may still default to a safe style if not guided
Key points from the news
- Store personal writing samples in Claude’s Memory to set a consistent tone.
- Create a forbidden‑phrase list in a Project to block generic AI wording.
- Iteratively refine drafts with Artifacts, style profiles, and read‑aloud checks.
The guide then shows how to use the Artifacts view as a workbench, adding headings first and fleshing out sections step by step, while applying a custom style profile that records individual speech habits. Finally, a read‑aloud check can be asked of Claude to flag awkward phrasing that sounds stiff when spoken. Combining these five techniques creates a workflow where most of the “AI feel” is avoided from the start, reducing later editing effort for marketers and small‑business writers.
Does the text you write with Claude still sound like 'AI'? 5 hidden techniques to master its features
note.com · 20 September 2026
Does the text you write with Claude still sound like 'AI'? 5 hidden techniques to master its features
“Claude writes natural-sounding text.”
You started using it because you heard that, but when you look closely at the finished text, you still find that “AI-written feel” somewhere—have you ever had that experience?
In a previous article, I introduced a “10-point checklist” to remove the mechanical feel from AI-written text. This time, I’ll go a step further and introduce a hidden technique to prevent that “AI feel” from emerging in the first place by fully utilizing the specific features that Claude possesses.
Instead of fixing it later with a checklist, create a “state that is less likely to sound like AI” from the start. This shift in perspective will significantly reduce the effort required for revisions.
What you will learn in this article:
- Why Claude’s standard output still sounds like “AI”
- 5 hidden techniques using Claude’s unique features
- Concrete steps you can set up starting today
Why does even Claude sound like “AI”?
It is often said that Claude is better at writing natural text compared to other AIs. While this is true, there is a reason why that “AI feel” still remains.
It is because Claude is guessing “what tone it should write in” from scratch every time. If you don’t specify anything, Claude chooses a “safe, polite, and universally applicable style.” This is by no means wrong, but that very “safeness” is the true identity of the “AI feel” for the reader.
In other words, it’s not that you aren’t bringing out Claude’s full potential, but rather that in many cases, the cause is that you are using it while not teaching Claude “your personal style.”
Hidden Technique 1: Cultivate your writing style samples in the “Memory” feature
Claude has a memory feature, a mechanism that remembers your preferences and writing style tendencies as you continue your conversations.
Actively utilizing this is the first hidden technique. Create opportunities within your conversations to intentionally convey that “this way of writing is like me.”
これまで自分が書いたnote記事の中で、一番「自分らしい」と思う文章を貼ります。
今後の文章作成では、この文体の癖(語尾、間の取り方、比喩の使い方)を
できるだけ再現してください。
【自分の過去記事サンプル】
(実際に書いた文章を貼り付ける)
Once you have communicated this, it becomes easier for Claude to operate under the premise that “this person writes in this way” in subsequent conversations. This reduces the hassle of specifying your writing style every time, and as a result, the “AI feel” is less likely to appear.
Hidden Technique 2: Keep a list of forbidden words resident in the “Projects” feature
By using Claude’s Projects feature, you can keep common instructions and materials within a specific project. The second hidden technique is to use this to keep a “forbidden list” of standard phrases that create an AI feel resident in the project.
Register a list like the one below in the project’s instruction field.
このプロジェクト内での文章作成では、以下のフレーズを使用しないでください。
・「まず初めに」「さらに」「また」の多用
・「〜することが重要です」「〜と言えるでしょう」
・「いかがでしたか」「参考にしてみてください」
・「〜ではないでしょうか」の多用
代わりに、具体的な理由や体験に基づいた自然な言い回しを使ってください。
Once you register it in a project, Claude will construct text based on these rules every time you start a new conversation within that project. This eliminates the hassle of pasting the same instructions every time and allows you to maintain a consistent quality.
Secret Technique 3: Use the 'Artifacts' feature to get a sense of 'nurturing' your writing
Claude has a feature called 'Artifacts' that displays text or code in an independent, editable format. The third secret technique is to use this not just as an output destination, but as a workbench to gradually 'nurture' your writing.
Instead of seeking a finished product all at once, follow these steps:
Step 1: Have Artifacts output only the framework (headings and key points only) Step 2: Give partial instructions like, 'Flesh this out under this heading, incorporating my personal experience' Step 3: Make pinpoint corrections like, 'Change only this paragraph to sound more casual'
If you try to complete it all at once, AI tends to produce text that is a 'safe, lowest common denominator.' By repeating partial corrections in stages, 'traces of human intervention' naturally remain throughout the text, and as a result, the mechanical uniformity disappears.
Secret Technique 4: Use the style feature to lock in your 'speech habits'
Claude has a feature that allows you to set the style (tone and manner) of its responses. The fourth secret technique is to use this to create your own 'tone profile'.
It is effective not only to use existing generic styles like 'polite' or 'casual,' but also to verbalize and register the characteristics of your own speech patterns as follows.
文末は「〜です」「〜ます」を基本としつつ、
たまに体言止めを使ってリズムを作る。
専門用語を使うときは、必ず一言かみ砕いた説明を添える。
読者に問いかける一文を、記事の中盤に最低1つ入れる。
By verbalizing and registering your 'writing habits' rather than just a 'tone' in this way, the text Claude produces will move away from feeling like an existing template.
Secret Technique 5: Have Claude itself perform a 'read-aloud check'
The final secret technique is to have Claude check the text it wrote for 'unnaturalness when read aloud'.
以下の文章を、声に出して読んだときに不自然に聞こえる箇所を指摘してください。
「書き言葉としては正しいが、話し言葉として読むと違和感がある表現」
に絞って教えてください。
【文章】
(チェックしたい文章を貼り付ける)
Even if the text is well-structured, when you actually read it aloud, you may notice strangely stiff phrasing or unnatural pauses. By having Claude check from this perspective, you can pick up on 'awkwardness when read aloud' that the human eye might overlook.
The flow of using all 5 secret techniques together
When incorporated into an actual article production flow, it looks like this:
- Tell the memory your writing style samples (①) → Initial foundation building
- Register a list of forbidden words in the project (②) → Elimination of standard phrases
- Flesh out gradually with Artifacts (③) → Leave traces of human-like corrections
- Lock in speech habits with the style feature (④) → Maintain consistent 'personality'
- Finally, a read-aloud check (⑤) → Finishing check for awkwardness
①, ②, and ④ are 'preparation' techniques that, once set, will continue to be effective thereafter. ③ and ⑤ are 'practical' techniques used for every article production. If you finish the preparation first, the work each time will become much easier.
Summary: One thing to do today
Removing the 'AI feel' from Claude isn't just about fixing things later with a checklist. By fully utilizing the features Claude already has, you can create a state where mechanical-sounding text is less likely to be generated in the first place.
What to do today: Register a sample of your writing style or a list of forbidden words in either the Memory or Projects feature.
Once you've done this 'prep work,' creating articles from then on will become surprisingly easy. You'll be able to feel for yourself how much of a difference knowing these features makes.
If you found this article helpful, please click the 'Like' button; it encourages me. If you follow me, I will deliver the latest information on AI side hustles.
① Readers can tell when AI wrote your text. 10 checklists to finish articles that won't be detected
→ https://note.com/ai_do0419/n/nf9fb04dd5c1f
② It won't be detected as AI-written. 7 rewriting techniques that high-quality writers use in secret
→ https://note.com/ai_do0419/n/n512f9cdce493
③ When I used Claude as a 'sounding board,' my side hustle ideas started taking shape in 30 minutes every night
This text was published by note.com and written by もちまる@副業. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.
More in Marketing & Small Business
All →- Google AI Studio lets non‑engineers build personal apps for free · 4 src
- OpenAI tests Sponsored Agents to let ChatGPT users chat with brand AIs · 1 src
- Opinion: CrowdWorks Academy’s beginner Claude Code course teaches non‑engineers AI file automation · 1 src
- AWS and Salesforce expand partnership to embed AI agents and CRM data for small businesses · 1 src
- Opinion: limit ChatGPT to three options and stop at 80% to keep work fast · 2 src
Comments
via GitHub Discussions