ChatGPT is the one that started it. The one that got deployed in every office in early 2023. The one that showed up in your company's all-hands presentation with a slide that said "the future of work." The one your manager is now cc'ing on emails. The one that generated a four-paragraph response to a question that had a one-sentence answer.

You are living with the consequences of a product launch, not a technological inevitability. And you are allowed to notice that.

But here's the uncomfortable part: ChatGPT is also genuinely good at some things. Not the things it gets credited for — not replacing your job, not running your business, not thinking for you. The things nobody talks about because they're too boring to put in a press release. Things like: drafting the first version of a document you'd otherwise stare at for an hour. Or explaining a concept three different ways until one of them clicks. Or synthesizing forty pages of notes into a readable summary at midnight before a meeting.

This guide draws the line. Clearly. Without apologizing to either side.

THE CHATGPT-SPECIFIC PROBLEM

// worth saying plainly

ChatGPT has a particular failure mode that the other AI tools don't quite match: it is extraordinarily confident when it is wrong. It doesn't hedge. It doesn't say "I'm not sure." It generates fluent, well-structured, completely incorrect answers with the same tone it uses for correct ones. This is not a bug that's been fixed. It is a fundamental characteristic of how large language models work — and ChatGPT, being the most widely deployed, has caused the most damage from it. This guide will tell you exactly when that matters and when it doesn't.

WHAT WE ACTUALLY BELIEVE

The hype was never about you. The ChatGPT rollout of 2023 was a product launch competing for market share, not a considered plan for how to improve your workday. You are under no obligation to adopt it on anyone else's timeline.
Free ChatGPT and paid ChatGPT are genuinely different products. GPT-3.5 (free) and GPT-4o (paid) have meaningfully different capabilities. Many people's opinion of ChatGPT is based on the free tier. That's a bit like judging a restaurant by the bread basket.
Confident wrong answers are worse than no answers. ChatGPT will invent citations, fabricate statistics, and state false information with complete conviction. For tasks where accuracy matters, this is a serious liability. For tasks where you're generating options, not facts, it barely matters at all.
Your data on the free tier trains the model. Unless you've opted out, conversations on the free tier of ChatGPT are used to improve OpenAI's models. This is not a scandal — it's disclosed in their terms. But it's worth knowing before you paste your company's financials into the chat window.
It is a draft machine, not an oracle. Every output is a starting point. The people who get genuine value from ChatGPT are the ones who treat it like a first draft they'll heavily edit — not a finished answer they'll ship. That distinction changes everything.

WHAT THIS GUIDE IS NOT

—
A pitch for ChatGPT Plus. We're not affiliated with OpenAI and we don't get anything if you subscribe.
—
An AI horror story. The risks are real but they're specific. We'll tell you which ones actually matter for your situation.
—
A technical manual. No API keys, no Python, no setup required for anything in this guide.
—
A comparison of every AI tool. We're focused on ChatGPT specifically — what it does, what it doesn't, and when your phone is still better.
—
For people who already love it. If you're already a power user, you don't need this. This is for everyone else.

By the end of this guide you'll be able to walk into any "mandatory AI training" and immediately identify which use cases are real, which are theater, and which tasks your spreadsheet could have handled the whole time.

The goal isn't to make you a ChatGPT user.
The goal is to make you impossible to impress with a demo.

ONTOMATON · NEWBRAND.US

PART 1

The Math of Overkill

Same infrastructure problem, different brand. Every ChatGPT query — free or paid — hits OpenAI's servers, draws power from a data center, and generates heat that gets cooled with water. The model got bigger, the costs went up, and the calculator on your phone still uses essentially zero electricity to add two numbers. That math hasn't changed.

SIDE-BY-SIDE COST COMPARISON

Calculator App

"What's 15% of $47?"

0.001Watt-hours

~0ml water

<1second

ChatGPT (GPT-4o)

"What's 15% of $47?"

~3Watt-hours

~100ml water

3–10seconds

VS

Word / Google Docs

"Make this text bold and centered"

0.005Watt-hours

~0ml water

<1second

Asking ChatGPT

"Make this text bold and centered"

~3Watt-hours

~100ml water

5–15seconds

VS

Google Search

"Who won the 2022 World Cup?"

0.3Watt-hours

~3ml water

<2seconds

Asking ChatGPT

"Who won the 2022 World Cup?"

~3Watt-hours

~100ml water

5–12seconds

The Overkill Ratio

3,000×

More energy than a calculator

1,600×

More energy than a Google search

100ml

Water per average ChatGPT query

THE FREE VS. PAID TIER — A COST YOU ALSO PAY IN QUALITY

Free Tier

GPT-3.5 / limited GPT-4o

Older, less capable model on free
Knowledge cutoff — no live data
No image generation (DALL-E)
Slower responses under load
Your conversations may train the model
No memory between sessions by default

Many people's entire opinion of ChatGPT is based on this tier. It's like judging a car by the rental fleet version.

ChatGPT Plus — $20/mo

GPT-4o + extras

GPT-4o — significantly more capable
Web browsing for current information
DALL-E image generation
Custom GPTs for repeat workflows
Voice mode (genuinely useful)
Opt-out of conversation training

Still has hallucination problems. Paying doesn't fix the confidence-when-wrong issue. It just makes the wrong answers more articulate.

None of this means ChatGPT is broken. It means it's a powerful tool with a real cost — and that cost is only justified when the task genuinely requires it. The escalation ladder in Part 3 tells you when that is. Most tasks don't clear the bar.

THE ESCALATION LADDER — PREVIEW

1

Simple math, conversion, timer

Phone calculator / clock app

0.001 Wh

2

Look up a fact, address, score, price

Google search

0.3 Wh

3

Sort, filter, format structured data

Spreadsheet or Word

0.005 Wh

4

Repeat task you do every week

Template or Custom GPT (once built)

0.001 Wh

5

Draft a complex document, synthesize info, work through judgment calls

Now ChatGPT earns its keep

3–50+ Wh

Energy and water figures are estimates based on published research into LLM inference costs. GPT-4o figures are higher than GPT-3.5 due to model size. Actual costs vary by query complexity and data center efficiency. The point is the order of magnitude.

PART 2

Your Phone Can Already Do That

ChatGPT launched in November 2022. The smartphone launched in 2007. In the fifteen years before ChatGPT existed, your phone developed a calculator, a search engine, a weather app, a camera scanner, a unit converter, a voice assistant, a spreadsheet, a notes app, and about forty other tools. None of them require a data center. None of them are wrong with complete confidence. Start there.

Calculations & Conversions

Never needs ChatGPT

What's 18% tip on $63.40?

Calculator app

63.40 × 1.18. Four taps. Zero watts.

How many kilometers is 8 miles?

Google / Siri search bar

Type "8 miles in km." Two seconds. Exact answer.

What's today's exchange rate?

Google search / XE.com

Live data. ChatGPT's free tier can't access this at all.

How many days until my deadline?

Calendar app / timeanddate.com

Your calendar has this. It's been there for years.

Information Lookups

Google is faster and more accurate

What are the store hours?

Google search

Live data in the knowledge panel. ChatGPT free tier doesn't have this. ChatGPT Plus may browse — but Google is still faster.

What's the weather this weekend?

Weather app

ChatGPT free cannot access live weather. At all. Your phone's weather app has done this since 2010.

Did the news happen yet?

Google News / any news app

ChatGPT free has a training cutoff. It does not know what happened last week. Google does.

Track my package

Carrier app / Gmail auto-track

ChatGPT cannot track packages. It's not connected to UPS. Gmail can see your shipping emails automatically.

Formatting & Document Tasks

Word has done this since 1995

Make this text bold and larger

Ctrl+B, font size dropdown

This is not a task. This is a menu option. In every word processor. For thirty years.

Sort this list alphabetically

Excel / Sheets: Data → Sort

One click in any spreadsheet. ChatGPT will sort it correctly — after you paste it, wait, and copy it back out.

Find and replace a word throughout

Ctrl+H in Word / Docs / Notepad

Literally a keyboard shortcut. Has been since the 90s.

Convert this Word doc to PDF

File → Export → PDF (built-in)

Every version of Word and Google Docs has this. No AI required, requested, or welcomed.

// ChatGPT-specific overkill — things people actually do

"Paste my meeting notes and ask ChatGPT to format them"

Use Word's outline view or Notion. Formatting is a menu, not a language model.

"Ask ChatGPT to write a formula for my spreadsheet"

Google Sheets now has built-in AI formula help. Type your question directly in the formula bar. Zero extra cost.

"Use ChatGPT to proofread a two-sentence email"

Grammarly's free tier. Works inline, no copy-pasting, catches typos as you type.

"Ask ChatGPT what a word means"

Double-click the word, right-click, Look Up. Built into every Mac and most browsers. Instant, offline, correct.

Spreadsheet formulas that replace ChatGPT for 80% of office math

Copy these into any spreadsheet. No AI required.

Sum a column

=SUM(A1:A100)

Average a range

=AVERAGE(B1:B50)

Count matching values

=COUNTIF(C1:C100,"Yes")

Look up a value in a table

=VLOOKUP(D2,Sheet2!A:B,2,0)

Flag overdue items

=IF(E2<TODAY(),"OVERDUE","OK")

Combine first and last name

=A2&" "&B2

Google vs. ChatGPT — the honest comparison

Use Google when you need

A specific fact with a known answer

Current information (live data)

Something that happened recently

A direct link to verify the source

Images, maps, or local results

Use ChatGPT when you need

To understand something, not just find it

To draft or restructure text

Judgment, synthesis, or explanation

A starting point, not a final answer

Many options generated fast

Free tools at newbrand.us that replace ChatGPT for specific jobs

Invoice Generator

Professional invoices in under two minutes. No account required.

newbrand.us/invoice

QR Code Maker

Generate a QR code for any URL or text. Download instantly.

newbrand.us/qr

Email Signature Builder

Your details, your design, copy the HTML. Done.

newbrand.us/signature

Privacy Policy Generator

Ten questions. One compliant privacy policy for your site.

newbrand.us/privacy

Business Name Checker

LLC and domain availability in one search.

newbrand.us/names

PDF Merge & Split

In your browser. Nothing leaves your device.

newbrand.us/pdf

Low-tech arsenal fully stocked. Now — when does the math actually tip in ChatGPT's favor?

PART 3

The Escalation Ladder

ChatGPT earns its cost at one end of a spectrum: open-ended tasks requiring synthesis, language, and judgment with no single correct answer. It loses badly at the other end: tasks with a known, verifiable answer that a faster, free, local tool can retrieve in seconds. The ladder tells you where your task actually sits.

Stay low-tech vs. escalate to ChatGPT

Stay low-tech when

There's a single verifiable correct answer

The answer might have changed recently

You need a source you can cite

A formula or filter handles it

You've done this exact task before

Speed matters more than nuance

Escalate to ChatGPT when

There's no single right answer — only better or worse ones

The task would take 30+ minutes manually

You need language, not just retrieval

You're synthesizing multiple inputs

You want ten options fast

Tone and judgment matter more than facts

Real tasks, real verdicts

LOW

"What's the capital of France?"

Use Google

Google → 0.3 Wh · verified · instant

Single correct answer. ChatGPT also knows this — but so does literally everything with internet access.

LOW

"What happened in the news today?"

Use Google News

Google News → live · free · ChatGPT free can't do this

ChatGPT free has a training cutoff. It does not know today's news. Google News does. This one isn't even close.

LOW

"Summarize this Wikipedia article"

Read the intro paragraph

Wikipedia's first paragraph → 0 Wh · already a summary

Every Wikipedia article begins with a summary. It's the first thing on the page. You scrolled past it to open ChatGPT.

AI

"Explain why this legal clause is concerning and what I should ask my lawyer about it"

Use ChatGPT

ChatGPT → worth the cost

No search engine does this. Requires reading comprehension, legal context, and synthesizing implications. Earns it.

AI

"Write five different subject lines for this email campaign. Tone: urgent but not pushy."

Use ChatGPT

ChatGPT → fast, useful, genuinely good at this

Creative generation with tone constraints is a language model's natural habitat. Five options in ten seconds.

AI

"Here's a 50-page report. What are the three most important things I need to act on?"

Use ChatGPT

ChatGPT Plus (with file upload) → hours saved

Long document synthesis is one of ChatGPT's clearest wins. Upload the PDF, ask the question, verify the answer.

// ChatGPT-specific feature worth knowing

Custom GPTs — the rung between template and AI

ChatGPT Plus includes Custom GPTs — pre-configured versions of ChatGPT built for a specific repeated task. If you find yourself prompting ChatGPT the same way every week, a Custom GPT eliminates that setup time. A Custom GPT for "reformat my meeting notes into action items" is genuinely useful — it's one click instead of a paragraph of instructions every time. This is the one feature ChatGPT has that currently has no equivalent in most other tools.

The five-question tipping point test

1

Is there a single verifiable correct answer?

Google it

2

Could it have changed in the last few months?

Google or news app

3

Could a formula, filter, or sort handle it?

Use your spreadsheet

4

Does it require judgment, synthesis, or language?

ChatGPT earns it

5

Would doing it manually take 30+ minutes?

ChatGPT earns it

Two or more "earns it" answers = justified escalation. Fewer than two = reach for the low-tech tool first. You can always escalate if it's not enough.

You know when to climb. Now: what to actually say when you get there — and what ChatGPT will get wrong regardless.

PART 4

ChatGPT for the Skeptic

The rules for using ChatGPT well are the same as the rules for using any powerful tool: know what it's for, know what it isn't for, and never confuse confidence for accuracy. ChatGPT is uniquely prone to the last one. That's not a reason to avoid it — it's a reason to stay alert.

Prompting like a skeptic — five principles

Skeptics are actually better at this than enthusiasts. You already don't fully trust it, which means you'll be specific, critical, and in control. That's exactly the right posture.

1

Give it context it couldn't possibly have

ChatGPT knows nothing about you, your company, your history, or your situation. The more context you provide, the less it has to guess — and guessing is when it invents things.

Weak

"Write a proposal."

Strong

"Write a two-page proposal for a small marketing agency pitching ongoing social media management to a local restaurant group. Monthly retainer model, emphasize consistency and local knowledge. Formal but warm tone."

2

Never ask it for current facts without verifying

ChatGPT will state statistics, citations, and recent events with complete conviction — and be completely wrong. On the free tier, it doesn't know what happened after its training cutoff. On Plus with browsing, it can look things up but still misreads sources. Verify everything factual before you use it.

Weak

"What's the current minimum wage in California?" → use ChatGPT's answer in a document.

Strong

Ask ChatGPT to help structure the argument. Look up the current minimum wage yourself at the official source. Combine both.

3

Ask for options, not answers

ChatGPT is a draft machine. The less you need it to be definitive, the more useful it becomes. Ask for five options instead of one recommendation. Ask for arguments on both sides instead of a verdict. You pick the winner.

Weak

"Should I hire a contractor or an employee?"

Strong

"Give me five reasons to hire a contractor and five reasons to hire an employee for a part-time bookkeeping role at a 10-person company."

4

Push back directly — it responds well to criticism

ChatGPT is not fragile. Telling it the output is wrong, too formal, too long, or missed the point will produce a genuinely different result. Most people accept the first draft. Don't.

Weak

Accept the output. Paste it. Send it. Realize it sounds like a chatbot wrote it.

Strong

"That's too stiff and too long. Rewrite it at half the length and make it sound like an actual person wrote it, not a corporate newsletter."

5

Always edit before you send

ChatGPT output sounds like ChatGPT. Every reader with a brain can tell. The fix is two minutes of editing — cut the filler phrases, add one specific detail that only you would know, and adjust the tone to sound like you. That's the difference between "AI-generated" and "actually useful."

Weak

"In conclusion, I hope this email finds you well and I look forward to our continued collaboration going forward."

Strong

Delete those sentences. Add one line that only you could write. Send that.

Ten tasks ChatGPT is genuinely good at

01

Drafting emails you've been dreading

"Write a response to a client who's angry about a missed deadline. Firm, honest, not groveling. We're offering a 10% discount on the next invoice."

Tone under pressure is hard. ChatGPT is good at threading that needle. Edit the result — but the hardest part is starting.

02

Summarizing long documents (Plus only)

"Here's a 40-page report [upload]. What are the three most important findings and what should I do about each one?"

Requires file upload — ChatGPT Plus only. Genuinely saves hours. Verify the summary against the original before acting on it.

03

Explaining something complex in plain English

"Explain what a floating rate mortgage means for someone who's never owned property before. Use an analogy. No jargon."

Infinitely patient, never condescending. Better than most explainers you'd find searching. Don't use it for current rates or legal advice.

04

Brainstorming options when stuck

"Give me 10 team-building activities for a remote team of 12 that aren't trivia nights or virtual happy hours. Budget under $200."

Fast, varied, and lower-stakes than asking a colleague for ideas they'll expect you to implement. Use two, ignore eight.

05

Editing and tightening your own drafts

"Here's a proposal I wrote. Cut it by 30%, fix the structure, and make the opening sentence actually interesting."

Editing your own writing is hard because you can't see it fresh. ChatGPT can. Verify that it didn't drop important information.

06

Generating job descriptions and interview questions

"Write a job description for a part-time office manager at a 15-person law firm. Include 6 interview questions that reveal attention to detail."

Structured writing with clear parameters. Saves the blank-page problem. Review carefully for anything that might have legal implications.

07

Translating jargon into plain language

"Here's a quote from an IT vendor for network infrastructure upgrades. Explain what each line item actually means and flag anything unusual."

Bridging specialist language and plain understanding consistently. Verify technical accuracy with the vendor — ChatGPT may miss current pricing.

08

Preparing talking points for hard conversations

"I need to tell a long-term employee their role is being restructured. They'll likely feel blindsided. Help me think through what to say and how to handle the likely reactions."

Rehearsal partner. Helps you anticipate objections and find language before you're in the room. The actual conversation is still yours.

09

Voice mode for thinking out loud (Plus only)

Speaking your situation out loud: "I have a decision to make about two vendors — talk me through the tradeoffs as I explain them to you."

ChatGPT's voice mode is genuinely useful for people who think better by talking. No equivalent in most other tools right now.

10

First drafts of anything long

"Draft the executive summary and introduction for a business plan for a mobile pet grooming service targeting urban pet owners. Emphasize convenience and premium quality."

Blank page problem solved in 90 seconds. The rest is editing, adding specifics, and making it yours. That split saves real time.

Where ChatGPT specifically falls short

Critical

Confident hallucination — the defining problem

ChatGPT invents citations, fabricates statistics, and states false information with complete fluency. It doesn't say "I'm not sure." It says "According to a 2022 study by Harvard..." — and that study may not exist. This is worst for factual research, legal or medical questions, and anything with specific numbers.

Critical

Free tier knowledge cutoff

GPT-3.5 and even GPT-4o without browsing have training cutoffs. They do not know what happened recently. Using them for current events, prices, laws, or any time-sensitive information is actively dangerous.

Weak

Sounding like itself

ChatGPT output has a distinctive voice: polished, organized, slightly corporate, and completely without personality. Anyone who reads a lot of it can spot it instantly. Editing is required to make it sound human.

Weak

Complex multi-step math and logic

For anything beyond basic arithmetic, ChatGPT can make errors that look correct. For any calculation you'll rely on, use a calculator, a spreadsheet, or code — not a language model.

Weak

Niche technical and specialized domains

In highly specialized fields — advanced medicine, novel legal interpretations, specific engineering problems — ChatGPT will produce plausible-sounding nonsense. Experts should treat every output as a first-year student's attempt that needs full review.

The skeptic's rules of engagement

Always do this

→

Read every output before using it

→

Verify any facts before acting on them

→

Edit it to sound like you

→

Push back when it misses the point

→

Own the final result

Never do this

—

Paste AI output without reading it

—

Treat statistics it gave you as verified

—

Use it when Google is faster and more accurate

—

Let it make decisions you should be making

—

Stop thinking because it sounds confident

Mindset locked. Now let's put it into actual workflows you use every week.

PART 5

The Office Anti-Suite

Real workflows, start to finish. Coral steps are the escalation points where ChatGPT genuinely earns its keep. Everything else stays in the tool that was already on your computer. The goal is the minimum viable AI — not maximum AI.

Common office workflows — escalation points marked

Writing a client-facing proposal

From blank page to sent document

Weekly task

1

Draft the proposal structure and opening

ChatGPT — "Draft a proposal outline for [service] to [client type]. One page, formal but warm."

~3 Wh

2

Fill in your specific pricing, timeline, and deliverables

Your knowledge — no AI can do this for you

0 Wh

3

Format in Google Docs or Word

Your saved template — always faster than starting fresh

~0.005 Wh

4

Tighten the language — cut filler, sharpen the opening

ChatGPT — "Edit for concision. Cut anything that sounds like a template."

~2 Wh

5

Export to PDF and send

File → Export → PDF. This has existed since 2003.

~0.01 Wh

Without ChatGPT

90

Minutes average

Staring at blank page. Rewriting the opening five times. Second-guessing every sentence.

With this workflow

22

Minutes average

AI handles blank page and final polish. You handle the substance. That's the split that saves time.

Responding to a negative review or complaint

From angry customer to resolved situation

Unpredictable

1

Read carefully. Understand what they actually want.

Your judgment — this step cannot be outsourced

0 Wh

2

Draft a response — empathetic, not groveling, actionable

ChatGPT — "Customer is upset about [X]. We can offer [Y]. Write a response: empathetic but not defensive, focuses on resolution."

~3 Wh

3

Edit so it sounds like you, not a chatbot

Two minutes of editing. Non-negotiable.

0 Wh

4

Send and follow up if needed

Your email or the review platform directly

~0.01 Wh

Preparing for a job interview (as the interviewer)

From job description to first candidate conversation

Occasional

1

Generate the interview question set

ChatGPT — "Give me 8 interview questions for a [role] that reveal [key quality]. Include one scenario-based question."

~2 Wh

2

Review and cut questions that don't fit your context

Your judgment — you know what your team actually needs

0 Wh

3

Prepare your explanation of the role, team, and culture

ChatGPT — "Help me explain [role] to a candidate in 2 minutes. Emphasize [values]. Anticipate their likely questions."

~2 Wh

4

Conduct the interview

In person or video — this part is entirely yours

0 Wh

// ChatGPT-specific: the Custom GPT shortcut

Build it once, use it every week

If you find yourself giving ChatGPT the same long setup instructions every time — your role, your company, your tone — build a Custom GPT (ChatGPT Plus only). Set those instructions once, permanently. Then every session starts already knowing the context. A Custom GPT for "draft my weekly status report" or "respond to customer service emails in our brand voice" eliminates two minutes of setup on every single use.

When your current tool hits its limit — escalation map

What you started with

When to escalate to ChatGPT

Word / Docs draft

→

When you've rewritten the opening three times and still hate it

Spreadsheet with data

→

When you need to explain what the numbers mean in plain English

Email thread to reply to

→

When the tone is delicate and the relationship matters

Brainstorm list

→

When you're stuck and need ten more options in thirty seconds

Template you use weekly

→

When this instance is different enough that the template doesn't fit

Google search results

→

When you have the facts but need help understanding what they mean for your situation

Workflows mapped. Now the part most people skip: what you're actually handing over when you use this thing.

PART 6

AI Hygiene for Normal People

ChatGPT is the most widely used AI product in the world. It is also the one with the most questions swirling around data, privacy, and what OpenAI actually does with your conversations. Most of the fear is overblown. Some of it is completely legitimate. Here's what's actually true.

What you're sharing when you use ChatGPT

Generally fine to share

General business descriptions
Anonymized scenarios
Document drafts without PII
Industry research questions
Creative writing and brainstorming
Public information you're analyzing

Share with care

Client names — anonymize if possible
Internal financial figures
Employee performance details
Legal matters under active dispute
Proprietary business strategy
Anything under NDA

Keep this offline

Passwords or credentials
Social security numbers
Full credit card or bank details
Medical records with identifying info
Classified or regulated information
Other people's private messages

Five ChatGPT privacy myths — corrected

Myth

"ChatGPT is reading my conversations and selling them to advertisers."

OpenAI does not sell user data to advertisers and does not run an advertising business. Conversations on the free tier may be reviewed by staff and used to improve the model — which you can opt out of in settings. That's different from selling data, but it's still worth knowing.

Myth

"The paid tier means my data is completely private."

ChatGPT Plus gives you the option to opt out of having your conversations used for training — but it's opt-out, not automatic. You need to go to Settings → Data Controls → turn off "Improve the model for everyone." On the free tier, there's no opt-out for training data.

Myth

"If I use ChatGPT for work, OpenAI can see my company's secrets."

Using the consumer product (chat.openai.com) for sensitive business data means those conversations are processed by OpenAI's servers and subject to their standard data terms. ChatGPT Enterprise has a different agreement: conversations aren't used for training and data handling is contractually controlled. If your company has sensitive information to discuss, Enterprise is the appropriate tier — or don't use it at all.

Myth

"ChatGPT is always listening through my microphone."

The ChatGPT app is not ambient. It only listens when you actively tap the microphone button. It does not passively capture audio. Voice mode requires explicit activation every time.

Myth

"Deleting a conversation makes it gone forever."

Deleting a conversation removes it from your visible history. OpenAI may retain the data for a period per their privacy policy. For truly sensitive material, the safest practice is simply not to share it in the first place.

Settings worth knowing about

Training data opt-out

Settings → Data Controls → Improve the model

Turn this off to prevent your conversations from being used to train future models. Available on Plus — not available on the free tier.

Recommended: turn off if using for business

Chat history & training

Settings → Data Controls → Chat history

Turning off chat history also disables training on your conversations. Conversations aren't saved. Useful for sensitive one-off sessions.

Recommended: use for sensitive conversations

Memory settings

Settings → Personalization → Memory

ChatGPT can build a memory of what you've told it across sessions. You can view, edit, or delete any stored memory. Review it periodically.

Recommended: review quarterly

Shared links

Any conversation → Share

Sharing a conversation link makes it publicly accessible to anyone with the URL. Be careful sharing conversations that contain sensitive context.

Think before sharing

The AI hygiene checklist — ChatGPT edition

Before you type

Would I be uncomfortable if a colleague saw this full conversation?

Can I anonymize the names and specifics without losing what I need?

Is this covered by an NDA, employment agreement, or confidentiality clause?

Before you use the output

Have I verified any statistics, citations, or named sources it gave me?

Does this actually sound like me, or does it sound like ChatGPT wrote it?

Have I read the full output before sending or publishing it?

Am I comfortable owning this as my own work?

Periodically

Check what ChatGPT has stored in Memory — delete anything outdated or sensitive

Clear old conversations that contained sensitive details

Ask yourself: am I using this because it helps, or out of habit?

OpenAI is a for-profit company with a complex structure and commercial interests. Their privacy practices are legitimate but differ meaningfully between the free tier, Plus, and Enterprise. Knowing which tier you're on and what it means for your data isn't paranoia — it's just being informed. The same common sense you'd apply to any cloud service applies here.

That's the full guide. One page left: the reference card you'll actually keep.

// Appendix

The Quick Reference Card

Everything in this guide, on one page. Keep it. Share it. Argue with it.

Always use a simple tool for

—Calculations and unit conversions
—Current facts, prices, and hours
—Weather, timers, reminders
—Sorting and filtering structured data
—Generating invoices and QR codes
—Anything with a single correct answer
—Tasks you do the same way every time

Escalate to ChatGPT when

→There is no single correct answer
→You need to produce or improve language
→The task would take 30+ minutes manually
→You need to synthesize multiple inputs
→Tone and judgment matter
→You're stuck and need a first draft
→You need 10 options fast

The five-question tipping point test

Stay low-tech

1Is there a verifiable correct answer? → Google it

2Can a formula or filter handle it? → Use your spreadsheet

3Have you done this exact task before? → Use a template

ChatGPT earns it

4Does it require judgment, synthesis, or language?

5Would doing it manually take 30+ minutes?

—Two "earns it" answers = justified escalation

Be specificInclude your role, audience, tone, and constraints

Say what you don't wantNo bullet points, no formal tone, cut the fluff

Ask for a draftNot a final answer — a starting point you'll edit

Push back"Too stiff." "Missed the point." "Try again."

Verify factsChatGPT hallucinates confidently — check anything you'll act on

Own the outputEdit it to sound like you before using it

Ten tasks ChatGPT is genuinely good at

01Drafting difficult emails
02Summarizing long documents
03Explaining complex topics simply
04Preparing for hard conversations
05Editing and tightening your drafts
06Building decision frameworks
07Generating options when stuck
08Translating technical language
09Writing job descriptions and questions
10Creating first drafts of anything long

Where ChatGPT falls short

—Confident hallucination — sounds right, isn't
—Current facts on free tier (older cutoff)
—Specific legal, medical, or financial advice
—Highly specialized technical domains
—Your genuine personal voice
—Making decisions for you

Never share with any AI

—Passwords or credentials
—SSNs, banking details, medical records
—Anything under NDA

Free tier (GPT-3.5 / limited GPT-4o)

—Older knowledge cutoff

—Slower, less capable model

—Conversations may train future models by default

—No image generation, no voice mode

—Fine for drafting and brainstorming

ChatGPT Plus ($20/mo — GPT-4o)

→More current knowledge, better reasoning

→Image generation via DALL-E

→Voice mode — genuinely useful

→Custom GPTs for repeat tasks

→Data not used for training by default

// The hallucination warning — read this

ChatGPT will state incorrect information confidently and without hesitation
It is more prone to fabricating citations, statistics, and names than it is to admitting uncertainty
Any fact you plan to act on — verify it independently, every time, no exceptions
This is not a flaw that will be fixed — it is a fundamental property of how language models work

The goal was never to make you an AI user. It was to make you a better judge of tools.

ONTOMATON · NEWBRAND.US