Teaching Your AI to Understand You
The Difference Between Sounding Good and Sounding Like You
Imagine you’ve spent an afternoon telling stories.
You’ve talked about childhood adventures, family holidays, funny habits, and all the little things that made someone special.
You’ve laughed at memories that still amuse you.
You’ve remembered conversations.
You’ve described things exactly as you would to a friend sitting across the kitchen table.
Then you ask your AI to turn one of those memories into a short piece of writing.
It produces something like this:
“He possessed an extraordinary spirit of curiosity, bringing boundless joy and unforgettable moments into the lives of everyone fortunate enough to know him.”
You read it.
It’s grammatically correct.
The words are nicely arranged.
It sounds rather impressive.
There’s just one problem.
You would never say that.
Not in a million years.
Perhaps you would have said:
“That boy was into absolutely everything. If there was something he wasn’t supposed to touch, you could pretty much guarantee he’d already taken it apart.”
Now that’s different.
There’s personality.
There’s humour.
There’s a particular way of describing someone.
You can almost hear the storyteller speaking.
And that’s what we’re going to explore in this chapter.
Because one of the biggest challenges in AI-assisted writing isn’t getting the technology to produce words.
It’s getting those words to reflect the person whose story is being told.
Modern AI can generate paragraphs quickly.
It can imitate many styles.
It can produce formal writing, casual writing, dramatic writing, humorous writing, and practically every variation between them.
But it doesn’t automatically know which style is yours.
That understanding has to be developed through the information and feedback you provide.
And even then, you’ll need to check its work.
The good news is that you don’t need to become a professional editor to do this.
You already possess the most important ability.
You know what sounds like you.
Our job is to help you use that judgment.
Your Voice Already Exists
Let’s begin with something important.
You don’t have to invent a writing voice.
You already have one.
You have a particular way of telling stories.
You choose certain words.
You have favourite expressions.
You probably use humour in ways that are familiar to the people who know you.
Perhaps you tell stories with lots of detail.
Perhaps you prefer getting straight to the point.
Maybe you enjoy describing the ridiculous things people do.
Maybe you’re quieter and more reflective.
Perhaps you interrupt yourself because one thought reminds you of another.
And perhaps you sometimes tell a story three different ways before finding the version that captures exactly what you mean.
Those habits are part of how you communicate.
Not all of them need to appear unchanged in a finished manuscript.
A reader may appreciate having repeated sentences removed.
A wandering story may benefit from clearer organisation.
And spoken language sometimes needs a little adjustment to work comfortably on the page.
But we don’t want to erase the personality underneath those habits.
That’s the distinction.
Editing should make your voice easier to understand, not replace it with somebody else’s.
And this is where AI can be both remarkably useful and occasionally frustrating.
It can help organise your words.
But unless you’re careful, it may also smooth away the very things that make those words recognisably yours.
The Problem With Perfectly Polished Nothing
Let’s talk about a particular kind of AI writing.
You’ll recognise it once you’ve seen enough of it.
Everything is beautifully balanced.
Every experience becomes meaningful.
Every ordinary event becomes a cherished moment.
Every personality trait becomes extraordinary.
And every paragraph seems determined to end with a profound observation about the human experience.
For example:
“Through these everyday moments, we discover that life’s greatest treasures are not found in extraordinary achievements, but in the love and laughter we share along the journey.”
There’s nothing grammatically wrong with that sentence.
But does it tell us anything specific about the person we’re remembering?
Not really.
It could appear in thousands of different books.
It doesn’t mention an actual event.
It doesn’t reveal a distinctive personality.
It doesn’t preserve an original expression.
And it doesn’t necessarily reflect what the storyteller meant.
This is what we want to avoid.
A paragraph can sound beautiful while containing almost no personal information.
For a book built around real memories, that’s a serious problem.
We don’t want pages filled with language that sounds meaningful but could describe practically anyone.
We want the details that make this particular person recognisable.
The specific stories.
The actual habits.
The funny expressions.
The unexpected incidents.
The memories that belong to this life and no other.
And we want those memories presented in a voice that feels natural to the person telling them.
That requires more than asking the AI to “make it beautiful.”
Sometimes the better instruction is:
“Make it sound more like me.”
And sometimes the best instruction is:
“Put that sentence back the way I originally said it.”
What Does It Mean to Teach an AI?
Before we go further, there’s something worth clarifying.
When we say we’re teaching an AI to understand you, we’re using ordinary language to describe a practical process.
We’re not suggesting that the system develops human understanding of your life.
We’re not claiming it knows you the way a close friend would.
And we’re certainly not assuming it can remember every conversation indefinitely.
What we’re doing is providing information that helps shape its responses.
You explain your preferences.
You provide examples.
You correct misunderstandings.
You identify language you like.
You reject language that doesn’t fit.
The AI can use those instructions and examples while they’re available in its working context.
Some applications also provide custom instructions, project settings, or memory features that can help carry certain preferences between sessions.
But those features vary.
They have limitations.
And they aren’t a substitute for keeping your own records.
For Project Crystal, the practical goal is straightforward.
We want to create a clear, reusable description of your voice and preferences that you can provide whenever you’re working on your book.
That way, your writing style doesn’t depend entirely on one long conversation.
You can save the instructions.
Revise them.
Use them with different AI systems.
And improve them as you discover what works.
That’s how we make the collaboration more consistent.
Mark and the Problem of the Disappearing Voice
Let’s return to the project that inspired this guide.
When Mark began turning his accumulated conversations into books, one of the challenges was preserving the way he actually communicates.
Mark’s natural conversational style is direct.
He enjoys humour.
He uses unusual comparisons.
He frequently jumps between technical observations and everyday examples.
He asks questions.
He challenges interpretations.
And he’s not particularly interested in making every idea sound like a formal academic presentation.
Those characteristics are part of what makes his material distinctive.
But when AI begins converting a large collection of conversations into polished chapters, something can happen.
The language becomes smoother.
The sentences become more conventional.
The humour gets softened.
Distinctive expressions disappear.
And the writing begins sounding less like Mark and more like an anonymous professional author.
That’s not necessarily what the book needs.
For The Meistro Manual, maintaining the connection between the original conversational material and the finished prose became an important editorial concern.
The manuscript required review for generic language, repeated ideas, and passages that no longer reflected the intended voice.
That experience provides a valuable lesson for Project Crystal.
The AI can help you write faster, but speed doesn’t guarantee authenticity.
A chapter produced in minutes may still need careful work.
Sometimes a particular phrase needs changing.
Sometimes a paragraph needs to be rewritten.
And sometimes an entire section needs to return to the original source material.
The lesson isn’t that AI-generated writing is automatically bad.
It’s that the author must remain involved.
And the earlier we establish your voice, the easier that involvement becomes.
Step One: Discover How You Naturally Tell Stories
So how do we begin?
Not with a complicated personality questionnaire.
Not with a list of literary styles.
And not by asking you to decide whether you’d prefer to sound like a famous author.
We’re going to start with your own words.
Choose three short recordings or written memories.
Ideally, select stories that feel comfortable and natural.
Perhaps one is funny.
Another is an ordinary family memory.
And the third involves something meaningful to you.
If you only have one recording, that’s enough to begin.
Now examine the material.
Don’t worry about grammar.
Instead, notice how you tell the stories.
Do you begin with a description of the situation?
Do you jump straight into what happened?
Do you use dialogue?
Do you laugh while explaining something?
Do you make comparisons?
Do you frequently say things like “I’ll never forget” or “You wouldn’t believe”?
Do you describe people through their actions?
Or do you prefer explaining what they were like before telling a particular story?
There’s no correct answer.
We’re simply identifying your natural habits.
And remember, people don’t always speak the same way in every situation.
You may tell funny stories one way and serious stories another.
That’s normal.
Our goal isn’t to reduce your entire personality to a rigid writing formula.
It’s to recognise patterns that can help the AI preserve your voice.
Step Two: Build a Voice Profile
Now we’re going to create a practical document.
We’ll call it your Storytelling Voice Profile.
This is a short description of how you prefer your writing to sound.
You can create it yourself or use AI to help examine your original recordings.
But any suggestions should be reviewed by you.
Let’s look at a fictional example.
Storytelling Voice Profile
Preferred Tone:
Warm, conversational, natural, occasionally humorous.
Sentence Style:
Mostly straightforward sentences. Some short sentences for emphasis.
Humour:
Likes gentle teasing, funny observations, and descriptions of everyday absurdities.
Emotional Expression:
Prefers emotions to emerge through events and personal recollections rather than dramatic declarations.
Typical Language:
Casual family expressions, familiar phrases, and direct descriptions.
Things to Preserve:
Nicknames, funny sayings, distinctive expressions, and original descriptions of people.
Things to Avoid:
Overly formal writing.
Exaggerated sentimentality.
Invented dialogue.
Generic inspirational conclusions.
Psychological interpretations that weren’t supplied by the storyteller.
Important Editing Preference:
Improve clarity while preserving the original meaning and personality.
That’s a voice profile.
It doesn’t need to be complicated.
And it doesn’t need to remain unchanged.
As you develop the manuscript, you may discover preferences you hadn’t considered.
Perhaps you realise that you enjoy shorter chapters.
Perhaps you prefer a more reflective tone in certain sections.
Maybe you decide that your original humour is one of the most important elements of the book.
You can update the profile.
The point is to create a reference.
Something you can provide to your AI whenever you’re working on your manuscript.
And something you can use to evaluate whether a draft sounds right.
Step Three: Give the AI Real Examples
Here’s where things become particularly useful.
Instead of simply telling the AI that you prefer natural writing, show it examples.
Suppose you have a transcript containing the following fictional passage:
“He was forever trying to fix things that weren’t broken. I don’t know what it was with that boy. He’d get something working perfectly, then decide he could make it better. And half the time, that’s when it stopped working.”
That passage tells us something about the storyteller’s voice.
It’s conversational.
It’s funny.
It contains affectionate exasperation.
The rhythm is natural.
And it describes the person through a specific recurring behaviour.
Now imagine asking the AI to analyse the style.
You might say:
“This is an example of how I naturally tell stories. Please describe the writing characteristics you notice without making assumptions about my personality.”
The AI might identify:
Conversational language.
Direct sentences.
Humour through understatement.
Affection expressed through specific observations.
A natural storytelling rhythm.
Those are reasonable observations based on the text.
But perhaps it also suggests that you prefer sarcastic humour.
You might disagree.
You can correct that interpretation.
“No, I wouldn’t call it sarcasm. I just like telling the funny parts naturally.”
Now you’ve refined the profile.
You can repeat this process with several examples.
And gradually, you’ll develop a clearer understanding of what makes your voice distinctive.
The Difference Between Style and Personality
Let’s make another important distinction.
Your writing style is not a complete psychological description of who you are.
An AI may examine a few passages and begin making broad statements about your personality.
Perhaps it decides you’re adventurous.
Or deeply philosophical.
Or unusually optimistic.
Those conclusions may not be justified by the material.
We’re not conducting a personality assessment.
We’re identifying practical characteristics of your language.
For example:
“You often use short sentences.”
That’s a writing observation.
“You use humour when describing family incidents.”
That’s another.
“You prefer direct explanations over elaborate descriptions.”
That may also be supported by your examples.
But:
“You use humour because you’re emotionally guarded.”
That’s a psychological interpretation.
And we haven’t established it.
You don’t need the AI analysing your inner life to help you write authentically.
You need it recognising how you express yourself.
Teaching writing style is about language and editorial preferences, not allowing AI to define your personality.
Keep that distinction in mind.
It’s one of the ways we maintain a grounded collaboration.
Step Four: Teach Through Comparison
Now let’s introduce one of the most effective techniques in this chapter.
Side-by-side comparison.
You provide the AI with an original memory.
Then you ask it to create a cleaned-up version.
You place the two passages beside each other.
And you compare them.
Let’s look at a fictional example.
Original Spoken Memory
“He had this thing about bringing home animals. I swear, every time he went outside, there was a chance he’d come back carrying something. One day it was a turtle. Another time it was a frog. I finally told him we weren’t turning the house into a zoo.”
AI’s First Attempt
“From an early age, he possessed a deep compassion for the natural world, frequently rescuing small creatures and demonstrating an extraordinary connection with animals.”
Now stop.
What happened?
The original story was funny.
It described a specific habit.
It mentioned turtles and frogs.
It included a memorable line about the house becoming a zoo.
The AI’s version removed those details.
It also introduced unsupported interpretations.
Was he rescuing the animals?
We don’t know.
Did he possess an extraordinary connection with them?
Perhaps, but the storyteller didn’t say that.
The rewritten passage sounds polished but is less informative and less faithful.
Let’s correct it.
You might tell the AI:
“You’ve removed the funny details and added interpretations I never made. Please keep the turtle, the frog, and the joke about turning the house into a zoo. Clean up the grammar if necessary, but don’t change the meaning.”
Now the AI might produce:
“He had a habit of bringing animals home. Every time he went outside, there was a chance he’d return carrying something. One day it was a turtle. Another time it was a frog. Eventually, I told him we weren’t turning the house into a zoo.”
That’s much closer.
But perhaps you still prefer the original opening.
You might say:
“Put ‘He had this thing about bringing home animals’ back. That’s how I would actually say it.”
And now the passage becomes:
“He had this thing about bringing home animals. Every time he went outside, there was a chance he’d come back carrying something. One day it was a turtle. Another time it was a frog. Eventually, I told him we weren’t turning the house into a zoo.”
There we are.
The writing is slightly cleaner.
But the personality remains.
This is the kind of editing we’re aiming for.
Not making every sentence grander.
Making it clearer without losing the person telling it.
The Power of Saying, “That’s Too Much”
As you work with AI, you may notice that it sometimes exaggerates.
You describe someone as curious.
It calls them brilliant.
You describe a funny habit.
It calls the person extraordinary.
You remember a pleasant afternoon.
It describes an unforgettable moment that changed everyone’s lives.
Sometimes a memory really is extraordinary.
Sometimes an afternoon genuinely changes someone’s life.
But those descriptions should come from the storyteller’s understanding.
Not from the AI’s tendency to intensify the language.
You can correct this with a simple instruction.
“That’s too much.”
Then explain what you mean.
“Don’t make him sound perfect.”
“Keep the description ordinary.”
“That wasn’t a profound moment. It was just funny.”
“Don’t turn this into a lesson.”
“Please describe what happened without telling the reader what to feel.”
These corrections can help establish a more grounded style.
And they’re especially important when writing about someone who has died.
Because there’s often a temptation to make every passage emotionally significant.
But the person’s life deserves to be remembered in its full variety.
The ridiculous moments.
The quiet moments.
The frustrating moments.
The ordinary routines.
The genuine accomplishments.
And the complicated experiences.
A personal book can include all of them.
We don’t need to make every sentence carry the weight of the entire life.
Step Five: Preserve the Person’s Voice Too
Until now, we’ve been concentrating on the storyteller’s voice.
But there’s another voice that may be equally important.
The voice of the person being remembered.
Their expressions.
Their jokes.
Their nicknames.
The way they answered questions.
The things they always said.
Perhaps your son had a favourite phrase.
Perhaps he had an unusual way of describing things.
Maybe he used a particular expression so frequently that the entire family began repeating it.
Those details can be wonderful additions to a book.
But they require care.
We need to distinguish between exact remembered wording and the general meaning of a conversation.
Suppose you remember him regularly saying:
“That’s not how I’d do it.”
If you’re confident those were his actual words, you can preserve the expression.
But suppose you only remember that he often disagreed with the way people approached a task.
You shouldn’t ask the AI to invent a collection of witty quotations that sound like something he might have said.
That’s not preserving his voice.
That’s manufacturing one.
Instead, you can write:
“He had a habit of questioning how things were done. If he thought there was a better way, he wasn’t shy about saying so.”
That communicates the behaviour without inventing a quotation.
If you later find a recording, letter, text message, or other reliable source containing his actual words, you can include them appropriately.
The principle remains the same.
A real person’s voice deserves more than plausible imitation.
We want to preserve the words that actually exist or are genuinely remembered.
Creating a Collection of Favourite Expressions
Here’s a practical exercise.
As you collect memories, begin keeping a list of distinctive expressions.
Not just your own.
Also the expressions used by the person you’re remembering.
For example, your collection might include:
A favourite family saying.
A nickname.
A joke repeated at gatherings.
A phrase used when someone was frustrated.
An expression associated with a particular hobby.
A funny response to a familiar situation.
For each expression, note its source.
Was it recorded in an interview?
Remembered by a family member?
Found in an old message?
Or preserved in an original recording?
This helps establish how confidently you can use it.
You might even create a document called:
Family Expressions and Sayings
Over time, it could become a useful reference for the manuscript.
And perhaps you’ll discover that several expressions belong together in a chapter about humour, family traditions, or everyday life.
But there’s no need to decide that immediately.
For now, we’re simply preserving language that might otherwise disappear.
What the AI Should Never Do With Your Voice
Let’s establish a few boundaries.
The AI should not invent memories to make your writing more interesting.
It should not create dialogue and present it as genuine recollection.
It should not assign emotions you haven’t expressed.
It should not replace your humour with generic sentimental language.
It should not transform ordinary people into idealised characters.
And it should not ignore corrections you’ve made within the working material.
Let’s examine that last point.
Suppose you’ve explained several times that your son wasn’t particularly interested in formal engineering, but loved experimenting with objects.
Later, the AI produces a chapter describing him as a gifted young engineer.
That’s a problem.
It has replaced your more precise description with a broader label.
Perhaps it remembered an earlier interpretation and lost track of your correction.
Perhaps it generated language that seemed appropriate in context.
Either way, you need to fix it.
And rather than relying on the AI to remember the correction forever, you can preserve it in your source material or project instructions.
For example:
Editorial Note: Do not describe him as an engineer based solely on childhood experimenting stories. The storyteller describes him as someone who enjoyed trying unusual ideas and seeing what happened.
Now the clarification has a place in your records.
You can provide it during future writing sessions.
This is one way to make your collaboration more consistent across conversations.
The Voice Guide Is Not the Memory Library
Let’s pause to distinguish two important parts of Project Crystal.
Your Memory Library contains the stories.
Your Voice Guide contains instructions about how those stories should be expressed.
They’re connected.
But they serve different purposes.
The Memory Library answers questions such as:
What happened?
Who was involved?
Where did the information come from?
What details are uncertain?
Which original expressions were used?
The Voice Guide answers different questions:
How formal should the writing sound?
How should humour be handled?
What kinds of language should be avoided?
How closely should original expressions be preserved?
What does the storyteller consider authentic?
This distinction is useful because you can apply the same Voice Guide to many different chapters.
You don’t need to explain your preferred writing style from scratch each time.
You can provide the guide, along with the source material for the chapter you’re developing.
Then ask the AI to work within those boundaries.
And if the result doesn’t sound right, correct it.
The guide is a reference.
Not a guarantee.
Your judgment remains essential.
A Working Example: The Same Memory, Three Different Styles
Let’s examine how much the style of a passage can change its character.
We’ll use another fictional memory.
The reported details are simple.
A boy tried to repair a broken lawnmower.
He spent an afternoon working on it.
He eventually got it running.
But when the machine started, a loose part caused an unexpected noise.
Everyone nearby laughed.
Now let’s look at three possible versions.
Version One: Formal Biography
“During his youth, he demonstrated considerable perseverance while attempting to repair a malfunctioning lawnmower. After several hours of mechanical work, he successfully restored its operation, although the outcome included an unexpected complication.”
That’s fairly dry.
It might belong in a technical report.
But it doesn’t sound much like someone telling a family story.
Version Two: Overly Dramatic AI Writing
“With unwavering determination and an indomitable spirit, he confronted the broken machine, transforming an ordinary afternoon into a testament to perseverance, ingenuity, and the beautiful unpredictability of life.”
Good grief.
We’ve somehow turned a lawnmower into a philosophical monument.
And we’ve added an enormous amount of interpretation.
Version Three: Conversational Storytelling
“He spent practically the whole afternoon messing with that old lawnmower. I don’t know how many times he took the thing apart, but he finally got it running.
“And wouldn’t you know it, the minute he started it, something came loose and made the most ridiculous noise.
“We all started laughing.
“After all that work, I think the lawnmower was trying to have the last word.”
Now we have something recognisably conversational.
But there’s an important detail.
That third version includes creative wording about the lawnmower having the last word.
If the storyteller actually said something similar, wonderful.
If not, we need to ask whether they want that kind of humorous addition.
Because even harmless embellishment is an editorial choice.
For Project Crystal, the safest starting point is to preserve the source language first.
Then, if the author wishes, we can explore stylistic changes and approve them explicitly.
The purpose of this example isn’t to declare one version universally superior.
It’s to show how writing style changes the experience of reading the same basic information.
And why your preferences matter.
Teaching Through Corrections
Now let’s introduce a practical habit that can improve the collaboration over time.
Whenever you make a significant correction, consider recording the lesson.
Not every small wording change needs documentation.
But recurring problems are worth capturing.
Perhaps the AI repeatedly makes your writing too formal.
You could add this to your Voice Guide:
Avoid: Formal biographical language when the original storyteller used casual, conversational wording.
Perhaps it repeatedly turns funny incidents into emotional reflections.
Add:
Avoid: Ending humorous stories with moral lessons or sentimental conclusions unless the storyteller specifically requests them.
Perhaps it keeps inventing dialogue.
Add:
Never: Present invented dialogue as remembered speech.
Perhaps it exaggerates personality traits.
Add:
Accuracy Rule: Describe behaviour through source-supported events. Don’t convert isolated incidents into lifelong personality claims without approval.
These notes become part of your editorial preferences.
And when you begin working on a new chapter, you can provide the updated guide.
This reduces the need to repeat the same instructions constantly.
It won’t eliminate every mistake.
But it gives the AI a clearer reference.
And it gives you a way to maintain consistency.
A Simple Correction Log
Let’s create another small working document.
Call it:
Writing Corrections and Lessons
It might look like this:
| Problem | Correction | General Rule |
|---|---|---|
| Writing too formal | Restored casual language | Prefer conversational phrasing |
| Humour removed | Reintroduced original joke | Preserve distinctive humour |
| Emotion invented | Removed unsupported description | Don’t assign feelings |
| Events combined | Separated two memories | Preserve event boundaries |
| Dialogue invented | Changed to reported speech | Never invent quotations |
You don’t need to maintain an elaborate log.
A few short notes can be enough.
The purpose is to recognise recurring problems.
And once you notice them, you can update your Voice Guide.
This is another example of a principle from Mark’s working style.
Notice what doesn’t work.
Identify the problem.
Improve the process.
Then try again.
The goal isn’t perfection on the first attempt.
It’s a method that becomes more dependable through use.
How Much Should You Let the AI Change?
This is an important editorial decision.
Some people prefer very light editing.
They want the manuscript to preserve their natural words as closely as possible.
Others are comfortable with more substantial restructuring.
They may wish to combine several conversations into a smooth narrative.
Neither preference is automatically wrong.
The important thing is that you understand the changes being made.
Let’s introduce three editing levels.
Level One: Light Cleanup
At this level, the AI may correct obvious grammar and punctuation, remove accidental repetition, and improve readability without changing the structure significantly.
The original language remains largely intact.
This can be useful when preserving the storyteller’s natural voice is the highest priority.
Level Two: Organised Narrative
At this level, the AI may rearrange sentences, clarify chronology, and create transitions between parts of the same memory.
The goal is to produce a coherent passage while preserving all important facts and distinctive expressions.
This may be useful when a recording contains a complete story told out of order.
Level Three: Manuscript Development
At this level, the AI may help combine several approved memories, organise them into sections, develop chapter openings, and improve overall narrative flow.
This involves more significant editorial decisions.
And it requires more careful review.
Because the greater the transformation, the greater the opportunity for misunderstanding or unsupported additions.
For Project Crystal, we’ll generally begin with light cleanup.
Then move toward more substantial editing when you’re comfortable.
You can always ask for a more polished version.
But it’s much harder to recover distinctive language after the original has been discarded.
That’s why we preserve it.
Learning to Review Without Becoming an Editor
You may be thinking:
“This is all very well, but I don’t know how to edit a book.”
That’s okay.
You don’t need to learn every professional editorial technique before you begin.
We’re going to start with a few straightforward questions.
When you read an AI-generated passage, ask yourself:
Does this sound like me?
If not, identify which words or sentences feel wrong.
Is this what actually happened?
Check important details against the source.
Did the AI add anything?
Look for invented descriptions, dialogue, motivations, or emotions.
Did it leave anything important out?
Sometimes the best part of the story disappears during summarisation.
Would someone who knew this person recognise the story?
This is not a guarantee of factual accuracy, but it can be a useful test of whether the writing has retained the person’s distinctive qualities.
You can answer these questions without using formal editorial terminology.
You might simply say:
“That paragraph doesn’t sound like me.”
Or:
“You left out the funny part.”
Or:
“He didn’t say that.”
Or:
“This is making the whole thing sound far more dramatic than it was.”
Those are useful editorial instructions.
And the AI can help you revise accordingly.
The Importance of Keeping Different Voices Distinct
A personal memoir may contain several different voices.
Your own narration.
Actual quotations from the person you’re remembering.
Stories contributed by relatives.
Letters.
Messages.
Interviews.
Perhaps even excerpts from old recordings.
Those voices should not automatically become identical.
Imagine your brother contributes a story.
He tells it in his own particular way.
You decide to include his recollection in your book.
You might choose to paraphrase it in your own voice, with attribution.
Or you might include a direct quotation.
Those are different editorial choices.
If you include his exact words, preserve them accurately.
If you paraphrase, make it clear that you’re describing his recollection rather than claiming firsthand memory.
And if the source contains sensitive information, ensure that its use is appropriate.
This becomes particularly important when different people remember the same event differently.
We don’t want the AI silently combining several voices into one supposedly definitive account.
Instead, we can preserve the perspectives.
Perhaps the book says:
“My brother remembers that afternoon a little differently.”
Or:
“According to his childhood friend, the story began several hours earlier.”
Those distinctions can add depth.
They can also help maintain honesty.
And they remind readers that a person’s life may be remembered through several different perspectives.
What Happens When Your Style Changes?
Here’s another useful point.
You don’t always tell stories the same way.
Perhaps you’re in a playful mood one day.
Another day, you’re more reflective.
Maybe you’re describing something difficult.
Perhaps you’re recalling an ordinary moment that doesn’t require much explanation.
Your language naturally changes with the subject.
That’s not inconsistency.
It’s part of being human.
So your Voice Guide shouldn’t force every chapter into exactly the same emotional tone.
Instead, it should describe your general preferences while allowing individual stories to retain their character.
For example:
“Use a natural conversational voice. Allow humour where it appears in the source material. Treat serious memories with appropriate restraint. Don’t force either humour or sentimentality into a story that doesn’t call for it.”
That’s a useful instruction.
It gives the AI guidance without making the writing mechanical.
And it helps preserve the emotional variety of the memories.
Because someone’s life wasn’t lived in one consistent tone.
A book about them shouldn’t have to be either.
The AI’s Memory Is Not Your Style Guide
Let’s return briefly to the technical side.
You may spend several weeks working with an AI.
It begins producing writing that feels increasingly comfortable.
Perhaps you’ve corrected it many times.
Perhaps it seems to understand your preferred style.
Then you start a new conversation.
And suddenly, the writing sounds generic again.
What happened?
The new conversation may not have access to the instructions, examples, or corrections from your earlier work.
Some services provide saved preferences or project-level instructions.
Others may not.
And even when memory features are available, they don’t guarantee perfect continuity.
This is why we’re creating a separate Voice Guide.
You can save it in your Memory Library.
You can provide it at the beginning of a writing session.
And you can update it whenever you discover something useful.
The guide belongs to you.
It isn’t dependent on one application remembering everything.
If you eventually change AI services, you can take the guide with you.
That’s an important part of maintaining control over your project.
Your Project Crystal Voice Guide
Now let’s build a reusable template.
You can fill it in gradually.
You don’t need to complete every section immediately.
PROJECT CRYSTAL: PERSONAL WRITING VOICE GUIDE
Author or Storyteller:
[Your name or preferred identifier]
Purpose of the Book:
A personal collection of genuine memories and stories.
Preferred Writing Tone:
[Describe how you wish the writing to feel.]
Natural Storytelling Style:
[Describe how you usually tell stories.]
Humour Preferences:
[Explain whether and how humour should be preserved.]
Emotional Tone:
[Describe how you prefer emotional material to be handled.]
Favourite Expressions:
[List phrases or expressions that reflect your natural voice.]
Language to Avoid:
[List words, phrases, or writing habits you dislike.]
Editing Preferences:
[Explain how much rewriting you’re comfortable with.]
Accuracy Rules:
Do not invent memories, events, dialogue, dates, emotions, or motivations.
Keep interpretations separate from source-supported details.
Preserve uncertainty when information is incomplete.
Source Rules:
Use approved memory records and transcripts as the foundation.
Preserve distinctive original wording where appropriate.
Identify unsupported additions.
Review Process:
Present drafts for approval.
Accept corrections.
Do not treat unapproved changes as established facts.
Additional Notes:
[Any other preferences you discover during the process.]
That’s your starting document.
You can make it as simple or detailed as you like.
You may discover that only a few lines are necessary.
Or perhaps you’ll gradually build a more substantial reference.
Either approach is fine.
The point is to make your preferences visible and reusable.
A Reusable Prompt for Teaching Your AI Your Voice
Now let’s create a practical prompt.
This is something you can use when you have a few genuine writing samples or transcripts available.
PROJECT CRYSTAL: LEARN MY WRITING VOICE
I’m developing a personal memoir using genuine memories, recordings, and conversations.
I want you to help me preserve my natural storytelling voice.
I will provide several examples of how I normally speak or write.
Please study the examples and identify practical characteristics of my language.
Focus on:
- Sentence length and rhythm.
- Level of formality.
- Use of humour.
- Typical expressions.
- How I describe people and events.
- How I communicate emotions.
- How I introduce and conclude stories.
- Distinctive wording worth preserving.
Do not make psychological claims about my personality based on these examples.
Do not invent biographical details.
Do not assume every sample represents how I speak in every situation.
After examining the examples, create a proposed Storytelling Voice Profile.
Then ask me to review it.
I may correct your observations.
Use my corrections to revise the profile.
When the profile is approved, help me save a reusable version that I can provide during future writing sessions.
When editing my memories:
- Preserve my intended meaning.
- Avoid generic sentimental language.
- Don’t invent events or dialogue.
- Keep my distinctive expressions when appropriate.
- Don’t exaggerate personality traits or emotions.
- Improve clarity without erasing humour or individuality.
- Identify significant changes for my approval.
- Never treat a polished sentence as more important than an accurate memory.
The goal is not to make me sound like a professional author.
The goal is to make the writing clear while still sounding like me.
This prompt gives your AI a specific job.
It isn’t being asked to write your book yet.
It’s helping you develop an editorial reference.
And once that reference exists, it can support the later stages of manuscript construction.
The Side-by-Side Voice Exercise
Now let’s put the ideas into practice.
Choose one memory from your collection.
Ideally, select something fairly short.
A few paragraphs of original transcript will do.
First: Preserve the Original
Make sure the source is saved.
Don’t alter or overwrite it.
Second: Request a Light Edit
Ask your AI:
“Please make this easier to read while preserving my natural voice. Don’t invent anything or remove distinctive details.”
Third: Compare the Versions
Read the original.
Then read the edited version.
Pay attention to what changed.
Did the AI remove humour?
Did it change the order of events?
Did it replace familiar expressions with formal language?
Did it add emotions or descriptions?
Did it improve clarity?
Fourth: Give Specific Feedback
Tell the AI what you like and dislike.
For example:
“Keep that sentence.”
“Remove the dramatic ending.”
“That’s not how I speak.”
“You changed the meaning.”
“That part is much clearer.”
Fifth: Request Another Version
Ask the AI to revise the passage using your feedback.
Then compare again.
Sixth: Record the Lessons
If you notice a recurring preference, add it to your Voice Guide.
For example:
“Preserve short humorous sentences.”
Or:
“Don’t add reflective conclusions to funny stories.”
Seventh: Save the Approved Version
Once the passage feels right, save it separately.
Keep the original source.
Keep the approved version.
And, when useful, retain a note about the editorial decisions.
You’ve now completed a practical exercise in teaching your AI how you prefer to communicate.
And you’ve done it without needing to learn a complicated technical process.
You simply compared two versions and explained what felt right.
When to Trust Your Own Ear
There’s something I’d like you to remember throughout this process.
You may not always be able to explain why a sentence sounds wrong.
Perhaps the grammar is fine.
Perhaps the meaning is technically accurate.
Perhaps the AI insists that the passage is clear and polished.
But something about it doesn’t fit.
That’s worth paying attention to.
You can say:
“I can’t explain exactly what’s wrong, but that’s not how I’d tell the story.”
Then ask the AI to offer a few alternatives that remain faithful to the source.
You can compare them.
Perhaps one is closer.
Perhaps the original wording is better.
Or perhaps you decide to rewrite the sentence yourself.
All of those approaches are valid.
You don’t need to justify every editorial preference with a formal rule.
This is your writing.
And while AI can help you evaluate language, you retain the final decision about whether it represents your voice.
The AI can offer alternatives. It doesn’t get the deciding vote.
What If You Prefer Someone Else to Narrate?
This raises an interesting question.
Perhaps you don’t want the entire book written in your natural speaking style.
Maybe you’d prefer a more traditional biography.
Perhaps you want some sections written as personal recollections and others presented in a more formal narrative voice.
That’s possible too.
The key is making the choice deliberately.
For example, you might decide:
“Personal memories should remain in my first-person voice.”
“Historical background can use a more neutral third-person style.”
“Direct quotations should preserve the speaker’s original words.”
“Photograph captions should be concise and factual.”
These are editorial choices.
And your Voice Guide can include them.
We’re not insisting that every book must sound like a transcribed conversation.
We’re protecting the storyteller’s ability to decide how the material should be presented.
The finished manuscript may involve considerable editing.
But the editing should remain accountable to the source and the author’s intentions.
That’s the principle.
A Good AI Collaborator Should Make You More Independent
Let’s consider the larger purpose of this chapter.
We’re not trying to make you dependent on one particular AI service.
We’re not trying to create a mysterious relationship where the machine is the only thing that understands your writing.
And we’re not suggesting that you need the technology to tell you who you are.
Quite the opposite.
We’re helping you recognise and document your own preferences.
Your natural language.
Your humour.
Your storytelling habits.
Your editorial boundaries.
Your ideas about what makes a memory worth preserving.
Once those preferences are written down, you can use them with different tools.
You can share them with a human editor.
You can refer to them while reviewing your own writing.
You can change them.
You can discard instructions that no longer fit.
The process should make you more confident in your decisions.
Not less.
And that’s something worth carrying forward into every stage of Project Crystal.
The Most Important Editorial Question
Let’s return to the beginning.
You tell a memory.
The AI produces a polished passage.
You read it.
It sounds impressive.
But perhaps something is missing.
Now you know what to do.
Return to the original.
Compare the versions.
Identify what changed.
Correct anything inaccurate.
Restore the distinctive language.
And decide whether the result actually communicates what you intended.
Sometimes the AI’s version will be better.
Sometimes your original will be better.
Often, the strongest result will combine the clarity of careful editing with the personality of your natural storytelling.
But the decision remains yours.
And when you’re writing about someone you love, there’s one question that matters more than whether the prose sounds impressive.
Does this actually sound like the way I remember them?
That question can guide an enormous amount of editorial work.
It doesn’t eliminate the need to check facts.
It doesn’t guarantee that every memory is historically accurate.
But it helps keep the writing connected to the person and experiences you’re trying to preserve.
And that connection is the reason this project exists.
One Final Thought: The Words That Only You Would Choose
Imagine two people describing the same afternoon.
They were both there.
They both remember what happened.
But they tell the story differently.
One remembers the weather.
The other remembers the conversation.
One focuses on the sequence of events.
The other remembers the ridiculous expression somebody made.
Neither account necessarily contains the entire experience.
But each preserves a perspective.
That’s part of what makes personal storytelling meaningful.
Your book doesn’t need to pretend that it contains every possible perspective on someone’s life.
It can acknowledge that you’re telling the stories as you remember them.
And your voice is part of that perspective.
The way you describe your son.
The things you find funny.
The details you notice.
The memories that remain vivid.
The expressions that come naturally when you speak about him.
Those choices are part of the book’s character.
If we replace them all with generic language, something important disappears.
We may still have a collection of facts.
We may still have a readable manuscript.
But we’ve lost part of the person telling the story.
Project Crystal is designed to protect against that.
Not by refusing to edit.
Not by insisting that every spoken sentence remain unchanged.
But by making the storyteller an active participant in every significant transformation.
Your original memories remain available.
Your voice becomes a reference.
Your corrections become editorial guidance.
And your approval determines what belongs in the final manuscript.
The AI can help with the writing.
But it shouldn’t become the author of a personality that never existed.
That’s the difference.
And it’s a distinction worth protecting from the very first paragraph to the final page.
Before You Turn the Page
Let’s review the most important lessons from this chapter.
1. Your natural voice already exists.
You don’t need to invent a writing style. Begin by examining how you actually tell stories.
2. Polished writing isn’t automatically authentic writing.
A grammatically beautiful paragraph can still remove personality or introduce unsupported interpretations.
3. Give the AI real examples.
Original recordings and writing samples provide a much stronger foundation than vague instructions such as “make it sound natural.”
4. Create a Storytelling Voice Profile.
Record your preferences for tone, humour, sentence style, emotional expression, and editing.
5. Compare original and edited versions.
Side-by-side review helps identify when the AI has changed meaning, removed important details, or replaced your natural expressions.
6. Teach through corrections.
When you identify recurring problems, record the lesson and update your Voice Guide.
7. Preserve the voice of the person you’re remembering.
Keep genuine sayings, quotations, and distinctive expressions, but never invent dialogue and present it as remembered speech.
8. Separate style instructions from memory records.
Your Voice Guide explains how to write. Your Memory Library preserves what happened.
9. Keep your own copy of the Voice Guide.
AI memory features can be useful, but they shouldn’t be the only place your editorial preferences exist.
10. You remain the final editor.
The AI can suggest improvements. You decide whether those improvements are accurate and appropriate.
We’ve now developed another important part of the Project Crystal process.
You know how to collect memories.
You know how to preserve and organise them.
You know how to explore them through conversation.
And now you know how to begin teaching your AI to work with your particular style of communication.
But there’s another fascinating part of conversational discovery that we haven’t fully explored yet.
Sometimes the most important story isn’t the one you originally sat down to tell.
Sometimes one memory leads to another.
A photograph opens a conversation.
A conversation reveals an unexpected connection.
And suddenly, you begin noticing recurring people, places, experiences, or interests across your entire collection.
That’s where we’re going next.
In Chapter 12: Following the Threads, we’ll explore how separate memories can begin connecting, how to recognise useful patterns without inventing them, and how to use those discoveries to understand the larger collection.
We’ll also examine the difference between a genuine connection and an attractive interpretation that the evidence doesn’t support.
Because as your library grows, you may discover that the stories aren’t quite as separate as they first appeared.
But we won’t force them together.
We’ll follow the evidence.
We’ll listen to the memories.
And we’ll let the storyteller decide which connections matter.
The AI can help you find the words.
It can help you recognise patterns.
It can help you construct the manuscript.
But the story should still sound like it came from the person who lived it.
And when you reach the final page, that voice should still be there.
Recognisable.
Natural.
Yours.