AI - I’ll Instantly Know

Once you have read enough AI-generated content, certain patterns become obvious; the polished introductions, perfectly balanced paragraphs, orderly lists, phrases that sound intelligent but somehow say very little, or the conclusion that neatly circles back to the opening idea.

I catch myself thinking far too often...

“That was obviously written by AI.”

But detecting AI-generated writing is rarely that simple.

AI-detection programs can produce conflicting results, human writing can be falsely flagged, and a skilled writer may naturally use many of the phrases now dismissed as ‘AI language.’ OpenAI withdrew its own text classifier in 2023 because of its low rate of accuracy, while independent research has found that detection tools can be unreliable and may excessively misclassify writing by people whose first language is not English.

So, where does that leave authors, editors and publishers?

The answer is not to fear AI, nor completely trust it. It is to understand what it can do, recognise what it cannot, and use professional judgement alongside common sense.

There’s no Doubt: AI Is Changing the Way Writers Work

AI can be a valuable writing companion.

It can help an author organise ideas, develop an outline, identify gaps in a narrative, generate questions, summarise research, or suggest alternative ways to phrase a sentence. It can also assist writers who struggle with confidence, language barriers, or a learning disability.

For independent authors with limited budgets, AI can offer useful support that was once out of reach. That shouldn’t automatically be viewed as unethical.

To be honest, most writers already use technology to assist them in some form. For example, spellcheck, predictive text, grammar tools, speech-to-text software, online dictionaries and digital research databases all influence the finished manuscript. The ethical question is not “was technology used?” it is, “how much creative input has the author contributed?”

How to Avoid Using AI as a Substitute for Creativity

AI becomes problematic when it begins replacing the unique creativity of a writer.

A good example is when I read a manuscript that is polished and well structured, but lacks heart. By heart I mean an obvious lack of lived experience, emotional depth, insight or human voice.

For writers, when AI is used to substitute creativity, it can:

• diminish ownership of their natural voice;

• produce generic or repetitive narratives;

• create inconsistencies in perspective;

• weaken the development of their own writing skills;

• create uncertainty around copyright, originality and disclosure; and

• submit work that does not reflect the author’s genuine knowledge or ability.

There are also broader concerns about how generative AI systems have been trained and whether the work of authors and publishers has been used without permission or compensation.

The Australian Publishers Association has repeatedly emphasised the need for transparency, ethical frameworks and protection for creators and rights-holders as AI becomes more integrated into publishing.

The concern, therefore, is not that AI has entered the writing room, it’s what happens when the author quietly leaves it.

The ‘Obvious’ Signs of AI-generated Writing

There is no single word, punctuation mark or sentence structure that proves something was generated by AI. However, the over-use of metaphors, em dashes and adjectives may prompt a closer look. Here are some of the ‘obvious’ signs I look out for when assessing a manuscript:

1. Polished but strangely impersonal prose

The writing sounds professional, but there is little evidence lived experience or individual perspective. Furthermore, it might explain the subject well, but doesn’t share why the writer cares about it.  

2. Repetition

Many AI-generated passages follow a predictable rhythm:

• introduce the idea;

• present three or four balanced points;

• acknowledge a challenge;

• offer a hopeful reframe; and

• finish with a tidy inspirational conclusion.

There is nothing inherently wrong with this, but when every chapter follows it, the manuscript can begin to feel mechanical.

Another tell-tale sign is repeated information. AI-generated text will often restate the same idea several times using slightly different language.

3. Excessive signposting

Phrases such as these may appear repeatedly:

It is important to remember…In today’s rapidly changing world…At its core…This is not merely about X; it is about Y.By embracing these principles…Ultimately, the journey is one of…

Again, human writers use these phrases too, but the clue is not one phrase, it’s the accumulation of repeated transitions.

6. Lack of evidence

A manuscript may refer to research, experts or studies without properly referencing them. It may make the writing sound authoritative, but ultimately avoids specifics such as names, dates, locations, figures or citations.

7. Sudden changes in voice

A manuscript may shift abruptly from informal dialogue to highly polished professional prose. Vocabulary, sentence length, punctuation and confidence may change dramatically between chapters or even paragraphs.

8. Emotion without specificity

Lived human experience is particular. It remembers the chipped blue cup on the hospital table, the smell of rain through an open window or the sentence someone whispered in a waiting room. Authenticity lives in the detail whereas AI can write about grief, courage, trauma, love and resilience, but the emotion may feel empty.

9. Impossibly high output

Sometimes the strongest clue is in mathematics.

A person may occasionally write thousands of words in a single productive day. But when someone publishes an extraordinary volume of polished material every day, it is reasonable to assume that substantial assistance may be involved. That still doesn’t tell us whether the use was ethical, but does provide context.

Why AI Detectors Are Not Enough

AI detectors generally analyse statistical features of language. However, human writing is not statistically uniform.

A highly formal writer may be flagged, a person writing in a second language may use simpler or more predictable sentence structures, or a heavily edited manuscript may appear less spontaneous than an early draft. On the contrary, AI-generated text can be edited until it passes as human.

Research has repeatedly shown that detectors can perform poorly outside the conditions in which they were developed, and relatively modest editing can reduce their ability to identify machine-generated text.

A detector score should therefore be treated as a signal to investigate, not a verdict. No author should be accused of deceptive AI use solely because a website produced a percentage.

How Publishing Professionals Detect AI-Assisted Narrative

As an editor and publisher, I’m not running every chapter through a detector and accepting the result as fact. Professional assessment is more relative and I take a well-rounded approach to detecting AI-generated writing:

1.    I compare the manuscript’s voice

Does the writing remain consistent across chapters?

If an author’s natural voice is warm and imperfect, but sections of the manuscript suddenly read like a business report, I will look more closely.

2.    I examine the author’s relationship with the material

Can the author discuss the ideas naturally and explain why a scene matters, where a claim came from or how they reached a conclusion?

Writers usually understand and defend work they have genuinely developed, even when they need help expressing it.

3.    I check factual reliability

Names, dates, legislation, medical claims, historical events, quotations and references must all be verified. A reference list that contains non-existent books, journals or page numbers is a significant warning sign.

4.    I look for narrative fingerprints

Human writers have habits. We repeat certain words, favour particular sentence patterns, tell stories in recognisable ways and return to personal themes. Those habits create the ‘author voice.’

Large blocks that lack those fingerprints, or display entirely different ones, may indicate outside generation or heavy intervention.

5.    I review the drafting history

Not always, but when I’m uncertain about the use of AI, I will ask the author for manuscript notes, outlines, tracked changes and earlier manuscript files. These are all evidence of how a manuscript was developed.  

Evidence such as deleted paragraphs, re-written chapters, unfinished ideas, questions, contradictions and gradual refinement, will strongly prove authenticity.

6.    I ask the author directly

An honest conversation is often more valuable than software, and always my first point of call. A simple conversation around:

·         If AI was used, in what way and with which program?

·         Was it used to brainstorm, rewrite, research or generate entire passages?

·         Were all facts checked?

·         Does the author understand and approve every sentence?

·         Has the work been sufficiently rewritten in their own voice?

Transparency allows an editor to determine what further work may be required without treating AI use as an automatic moral failure.

An Ethical Way for Writers to Use AI

The healthiest approach is to treat AI as an assistant, not an author.

A practical principle is: Use AI to support your thinking, not to avoid thinking.

Writers can use it ethically by following several safeguards.

1.    Begin with your own ideas

Record your memories, research and intentions before asking AI for assistance. The creative foundation should come from you.

2.    Use it for questions, not just answers

Ask AI to identify gaps, challenge assumptions or suggest questions you have not considered to encourage deeper thinking, rather than replacing it.

3.    Never accept facts without verification

Confirm all factual material through reliable primary or authoritative sources. AI-generated citations should be treated as unverified until checked manually.

4.    Rewrite in your own voice

Don’t simply copy & paste generated paragraphs into your manuscript. Reshape the material, add personal examples, replace generic wording, and question whether every sentence reflects what you truly think and how you would naturally say it.

5.    Protect confidential material

Authors should be cautious when entering unpublished manuscripts, client information, private correspondence, culturally sensitive stories or identifying personal details into external AI systems.

6.    Keep records

Retain your outlines, drafts, research and revision history. These demonstrate the evolution of your work and help protect your authorship.

7.    Understand your publishing agreement

Publishers, competitions, educational institutions and funding bodies may have different requirements regarding AI-generated or AI-assisted work. Authors should check the relevant terms and disclose their use when required.

8.    Be transparent where the use is substantial

There is a meaningful difference between using AI to generate ten possible titles and asking it to re-write half a manuscript.

Where AI has materially shaped the narrative, structure or wording, disclosure may be appropriate. Transparency protects trust between the author, publisher and reader.

The Question Is Not “Human or AI?”

The writing world may never receive a perfect stamp that reliably separates human-created work from AI-generated content.

Even if detection technology improves, the line will continue to blur. Authors can directly dictate into software, use AI to organise research, employ grammar tools, collaborate with editors and then manually rewrite every sentence. But is that human writing, AI-assisted writing, or edited writing? It’s almost impossible to decide.

Maybe, better questions would be:

• Who supplied the ideas and experiences?

• Who made the creative decisions?

• Who verified the information?

• Who accepts responsibility for the finished work?

• Does the manuscript represent the author’s genuine voice?

• Has technology been used transparently, legally and ethically?

AI can help writers begin, organise, experiment and improve. What it cannot replace is the inner life of the writer: the memories, doubts, contradictions, observations and hard-won understanding that make one voice different from every other.

The words on the page are only the visible part of writing. Behind them are reading, research, reflection, questioning, revision and the willingness to change one’s mind.

That is why the future of authorship should not be built on suspicion alone, but on discernment.

Use the tools, understand their limitations, and protect your unique author voice. Be honest about the process, and be certain that, when the book finally carries your name, it also carries something unmistakably yours.

I’d like to acknowledge The Medium Daily for the inspiration for this blog. To see the original article, go to: How To Identify If Someone Used ChatGPT | The Daily Draft