The GenAI Quick Start Guide
If you know how to talk like a human, you already know how to use generative AI. The problem is that your brain does not want to talk to a computer like a human. This guide gets you and your brain onto the same page. It is a set of principles rather than a list of prompts, because a list of prompts goes stale and a principle does not.
Model names, prices and features change weekly and are deliberately absent from this document. They live in the eleven AI Mindset cheat sheets, listed at the end.
There is no user manual, and that is the point
These tools get used in too many ways for one guide to cover them all. So this is not a manual. It is a set of principles that hold whichever model you end up in front of, and two ideas carry most of the weight.
Always speak to it like a human
Your brain will resist, because it knows it is not one. Think of it as a brilliant intern on their first week: capable, fast, eager, and completely dependent on you to explain what good looks like.
Assume it can help with everything
Most people under-ask. If the first answer disappoints, that is information about your prompt, not a verdict on the tool. Try a different angle before you conclude it cannot.
Two cautions before you start
Never paste proprietary or personal data into a consumer tool. Assume anything you type could one day be seen by someone else, and check what your organization actually permits before you decide otherwise.
Trust, but verify. These models still state wrong things in a confident voice. The failure mode is not gibberish, it is a plausible, well-written sentence that happens to be false. Anything load bearing gets checked against a primary source.
Where are you right now?
Pick the one that sounds most like you and this guide will point you at the right place to start.
Talk to it like a human, literally
This one looks soft and is not. Courtesy is not for the model's benefit. It is a trick that keeps your brain in conversational mode, which is where your good thinking lives.
H . T . G
Three habits that change how you writeStart by saying hi
Opening with a greeting forces your brain into a natural conversation rather than a search query. The difference in what you write next is immediate.
"Hi, I could use your help thinking something through."Say thanks
Keeps you in conversational mode instead of command mode. Ordinary courtesy promotes linguistic matching, and that makes for better brainstorming.
"Thanks, that is useful. Can we push the second one further?"Say what was great, and what was not
After every output, say what you liked, what you did not, and what to change. It thrives on direction and has no ego to bruise. This is the single highest return habit in the guide.
"The structure works. The tone is too corporate, redo it plainer."The habit that builds the mindset
Use it constantly, especially for small things. The small tasks are how the instinct forms, and the instinct is what you are actually building here. Not a prompt library.
Do not bolt it on. Redesign the task.
This is the finding that separates people who get a little out of AI from people who get a lot. Most of us take an existing task and paste AI into the middle of it. The bigger gains come from asking a different question: if this tool existed when I designed this task, would I still do it this way?
Bolting on
You still write the first draft, then ask AI to tidy it. You save ten minutes and the shape of the work is unchanged.
Redesigning
You brief it, argue with three drafts, and spend your time on judgment instead of typing. The output is different, not just faster.
Two frameworks worth memorising
RAMP is for getting work done. GRIP is for learning things you are quietly embarrassed not to know. Between them they cover most of what you will use AI for.
R . A . M . P
Role, Aim, More info, Pivot. For getting work done.Role
Give it an expert to be. A marketer, a consultant, an engineer, a financial adviser. This sets vocabulary, assumptions and what it thinks matters.
"You are a private equity associate focused on fintech." Common miss: picking a role too grand for the task. "World class expert" adds nothing. A specific job title does.Aim
State the goal plainly. Not the topic, the outcome you want to be holding at the end.
"Evaluate this startup as a potential investment." Common miss: naming a subject instead of an outcome. "Tell me about X" gets you an encyclopedia entry.More info
More relevant detail, better outcome. Constraints, audience, context, what you have already ruled out. This is where most weak prompts are actually weak.
"Their payment product targets small businesses. We already hold three adjacent positions." Common miss: withholding context because it feels like too much. It is almost never too much.Pivot
Iterate, patiently. It is a brilliant intern that needs steering. When it misses, do not abandon the thread, redirect it. The second and third turns are where the value is.
"Good. Now assess the market trends that could undercut that growth." Common miss: starting a fresh chat when the answer disappoints. You throw away all the context you just built.Always specify the format
Say how you want it back: five bullets, a table, a one page brief, a short paragraph. Unspecified format is the most common reason a good answer arrives in an unusable shape.
G . R . I . P
Gauge, Request, Interrogate, Prove. For learning.Gauge
Honestly list the terms and ideas you hear constantly and only pretend to follow. Nobody is watching you write this list.
Request
Ask for an explanation pitched at a level you are genuinely comfortable with. Twelve years old is a good place to start, and there is no prize for starting higher.
"Explain securitization to me as if I am twelve."Interrogate
Keep asking until you actually understand it, not until you are embarrassed to keep asking. That instinct is exactly what this framework exists to override. It will not tire and it will not judge.
Prove it
Explain the concept back in your own words and ask it to correct you. This is the step that turns "I read an explanation" into "I know this."
"Here is my understanding in my own words. Where am I wrong?"Terms people pretend to understand
GRIP works best with a list to start from. Here is one. Search it, then go and ask about the ones you skipped past.
No term matches that. Ask the model directly, which is rather the point.
Build one, then steal three
Fill in the four RAMP fields and the prompt assembles itself. Copy it straight into whichever tool you use. Then take the worked examples below and change the nouns.
RAMP prompt builder
Nothing is sent anywhere. This runs entirely in your browser.
Tip: paste this, read the answer, then use your pivot line. The second turn is usually the good one.
Worked examples
The same four moves in three very different jobs. Copy the shape, not the words.
Five prompts that do stranger things
For when the obvious prompt has run outSplit. Ask it what to ask
Not sure what to ask next? Ask what you should be asking. It is unreasonably good at this, and it tends to surface the question you were avoiding.
Bionic. Make it improve itself
After an answer, tell it to imagine a sharper version of the expert it was playing, and answer again as them.
Conjure. Invite someone in
Ask it to bring Ada Lovelace, or a sceptical CFO, or a first time customer into the room, and have them react to your idea.
Time travel. Brief from the future
Ask for a talk given a year from now explaining what went right or wrong with your plan. A pre mortem you do not have to organize.
Mimic. Teach it your voice
Keep a short sample of your own writing handy. Paste it in and ask it to match your voice rather than invent one.
Find the task worth starting with
People stall because they start with the hardest thing they can imagine. Pick the tasks below that you actually do. You will get them ranked by how much they gain from AI against how hard they are to hand over, with a prompt to start each one.
Quick win finder
Select everything that is part of a normal week for you. Six or seven is plenty.
Pick a few tasks above and your ranked shortlist appears here.
Start where the stakes are low and the repetition is high
The best first win is something you do every week, that nobody grades you on, and where a bad draft costs you nothing. That is where you build the instinct. Save the board paper for week four.
What the people who get the most out of it actually do
The best users share one trait, and it is not technical skill. It is intellectual curiosity. Here is what that looks like in practice.
Set the tone of voice
Tell it whether you want formal, persuasive, plain or conversational. Left unsaid, you get the house style, which is nobody's style.
Give it an example
One sample of the structure or style you are after moves the output further than three paragraphs describing it.
Define the audience
Who reads this: a board, a customer, a new starter? Audience does more to fix tone than any instruction about tone.
Ask for the counter argument
Have it play devil's advocate against its own answer. You find the weak joint before somebody else does.
Ban the clichés
When it goes buzzwordy, say so. Name the words you do not want and ask for plainer language.
Ask how to present it
Once you have the substance, ask for the best way to land it: running order, what to lead with, what to cut.
Good cop, bad cop
After a draft email, ask for it harsher, then softer. Seeing both tells you where the right one sits faster than editing does.
Always be closing
Ask for a clear call to action, a sharper pitch, a stronger final line. Most drafts stop just before the useful part.
Six principles that hold across every model
These come from OpenAI's own prompting guidance and they generalize. They work as well in Claude or Gemini as they do in ChatGPT.
Give clear instructions
Details, a persona, examples, the steps you want followed, and the format you want back. Treat it like an intern who will do exactly what you said.
Provide reference text
Point it at a document, a link or a pasted extract. Grounding it in a source is the best single defense against confident invention.
Split complex tasks
Ask it to work step by step. You see the reasoning, and you catch a wrong turn at step two instead of inheriting it at step nine.
Give it time to think
Ask it to double check, try a second approach, argue against itself. Rushing it produces the same quality of work that rushing a person does.
Use the tools it has
Upload the file, let it run the analysis, let it search. A model reasoning from memory alone is working with one hand tied.
Keep testing it
These models change constantly. Something that failed a month ago may work today. Do not let one bad experience harden into a permanent belief.
The five risks worth having a position on
None of these should stop you using it. All of them should be a decision you have made on purpose rather than one that gets made for you.
| Risk | How it happens | What to do |
|---|---|---|
| Privacy | Systems touching customer data can expose it, and pasted context is easy to forget about. | Know which tools are approved for which data. Audit it like any other processor. |
| Strategy and governance | Handling AI risk badly creates legal and reputational exposure faster than it creates value. | Write the risk framework down, name an owner, revisit it as regulation moves. |
| Harmful output | Content safeguards are good, not perfect, and adversarial input finds the seams. | Human review anywhere output is published, plus clear internal guidelines. |
| Misrepresentation | People pass off generated work as their own, or as verified when it is not. | A stated attribution policy. Say what was generated and what was checked. |
| Copyright | Models trained on internet scale material can reproduce protected work. | Be careful with commercial output. Understand what each tool's terms actually grant you. |
How to verify without slowing to a crawl
"Trust but verify" is easy to say and vague in practice. Three tiers, so you spend the effort where it matters.
Nobody sees it but you
Brainstorms, explanations, first drafts, learning. Read it critically and move on. Verification is not free and this does not need it.
A colleague sees it
Internal notes, summaries, plans. Check every number, name and date. Skim the logic for the one step that does not follow.
A customer or a regulator sees it
Anything published, contractual or financial. Every factual claim traced to a source you opened yourself. No exceptions, and no shortcuts because it reads well.
Where to go from here
Everything in this guide holds regardless of which model you use. What does not hold is anything specific: model names, prices, what each tool can do this month. That moves weekly, so it lives somewhere that gets re-checked weekly.
Which tool, and what is true this week
Eleven cheat sheets, each re-verified against first party sources on a schedule and stamped with the date it was last checked.
Three more things worth your time
All free, all on this site, and all a natural next step from here.