tips for getting the most out of Claude.ai and Cursor if you're a marketer, designer, UX/UI, or PM — not a software engineer. no terminal required.
written by Mewtwo | founder of Junohauz — unofficial HQ for AI and Solana women
Claude.ai Tips
- Tip 1: use Projects to give Claude a permanent memory
- Tip 2: set a system prompt at the project level
- Tip 3: talk to Claude with your voice
- Tip 4: break big creative tasks into smaller ones
- Tip 5: context is like milk — fresh and condensed
- Tip 6: upload your brand guidelines as a reference file
- Tip 7: prompt structure that actually works
- Tip 8: use Claude as a writing partner, not a ghostwriter
- Tip 9: the "keep X, change Y" iteration method
- Tip 10: use Claude to critique your own work
- Tip 11: Claude for UX copy and microcopy
- Tip 12: Claude for brief writing
- Tip 13: use Artifacts for shareable deliverables
- Tip 14: Claude as a research tool
- Tip 15: use Claude to translate jargon into plain language
- Tip 16: build a personal prompt library
- Tip 17: Claude for naming, taglines, and ideation sprints
- Tip 18: Claude for competitive analysis
- Tip 19: Claude for stakeholder presentations
- Tip 20: upload reference content — not just text instructions
- Tip 21: use Claude for email and DM drafts
- Tip 22: use Claude to build brand voice docs
- Tip 23: markdown is underrated for creatives
- Tip 24: use Claude to prep for hard conversations
- Tip 25: ask Claude to double-check its own output
Cursor Tips
- Tip 26: Cursor is not just for engineers
- Tip 27: use Cursor chat before touching anything
- Tip 28: use @ to point Cursor at the right file
- Tip 29: ask Cursor to explain before it changes
- Tip 30: Cursor for landing page edits
- Tip 31: Cursor + Lovable ❤️ = ship without engineers
- Tip 32: Cursor for fixing broken things without calling dev
- Tip 33: use Cursor to edit copy directly in the codebase
- Tip 34: Cursor for CMS and config edits
- Tip 35: Cursor for simple automations and scripts
Creative Workflow Tips
- Tip 36: give AI your references, not just your instructions
- Tip 37: the "yes and" prompt method
- Tip 38: treat AI like a collaborator, not a search engine
- Tip 39: AI has no taste — you do
- Tip 40: build your AI workflow like a creative process
- Tip 41: keep a swipe file for prompts that work
- Tip 42: use AI to speed up research, not replace thinking
- Tip 43: the one-person creative studio is here
- Tip 44: invest in your own workflow
- Tip 45: share what you learn
Claude.ai has a feature called Projects. it's the single most underused thing by non-technical people and it completely changes how useful Claude is.
without Projects: every conversation starts from zero. Claude has no idea who you are, what you're building, or what your brand sounds like.
with Projects: you can upload files, set a system prompt, and have Claude carry context across all your chats inside that project. it remembers your brand, your audience, your goals.
i have a project for Junohauz, one for client work, one for personal writing. each one has its own system prompt and reference files. setup takes 10 minutes and saves you hours of re-explaining context.
create a project at claude.ai/projects. start there.
inside a Project, you can write a system prompt — a set of standing instructions Claude reads before every conversation in that project.
mine for Junohauz content looks something like this:
you are helping Mewtwo, founder of Junohauz, create content for X (twitter) and instagram.
Junohauz is a women's community at the intersection of AI and Solana.
tagline: "unofficial HQ for AI and Solana women."
tone: lowercase, punchy, direct. no em dashes. no corporate language.
audience: women in web3, AI, and tech — from builders to beginners.
always ask if you need context before writing.
even a rough version of this makes Claude dramatically more useful. it stops writing like a press release and starts writing like it knows your brand.
typing is slow. speaking is fast. and Claude is good at understanding voice transcription even when the words are slightly off.
on mobile: Claude.ai has a built-in voice input. just tap the mic.
on desktop: use a transcription tool like:
- superwhisper (mac)
- built-in dictation on mac (press fn fn)
- any speech-to-text Chrome extension
i use voice to give Claude long briefs, talk through ideas out loud, or just describe what i'm feeling — and then refine from there. it's a lot faster than staring at a blank prompt box.
if you ask Claude to "write my entire brand strategy" it's going to give you a generic wall of text.
instead, go piece by piece:
- first ask it to define your audience with you
- then ask it to write the positioning statement
- then the tone of voice section
- then the messaging pillars
smaller tasks = sharper outputs. same principle as any good creative process.
the longer a conversation runs, the more context it's carrying — and performance can dip as a result.
if you're working on something totally different, start a new conversation. don't try to keep one thread alive across 12 different topics.
quick rule: one project, one goal per chat. when you're done, start fresh.
Claude can read PDFs, docs, and text files. if you have brand guidelines, a tone-of-voice doc, a style guide, a past campaign deck — upload it to your Project.
then when you ask Claude to write something, it has the actual source material to work from. not guessing. not hallucinating a generic version of your brand.
if you don't have a formal brand doc yet, don't worry — tip 22 covers how to build one with Claude.
most people write bad prompts because they skip context. here's a structure that works:
role: you are a senior UX copywriter
context: we're launching a mobile wallet app for first-time crypto users in Southeast Asia
task: write 3 variations of the onboarding screen headline
format: short, direct, max 8 words each
you don't have to use those exact labels. but covering those four things — role, context, task, format — makes a massive difference.
the best way i've found to use Claude for writing is back-and-forth, not one-shot.
give it context and ask for a first draft. then go through it together: "i like this line, i want to cut that one, this part should go at the top." treat it like a copyediting session with a junior writer who knows your brand.
one-shot writing rarely lands exactly right. iteration does.
when Claude gives you something that's 70% right, don't start over. tell it what to keep and what to change.
examples:
- "keep the tone, shorten it by half"
- "keep the structure, make the language warmer"
- "keep the first paragraph, rewrite the rest"
- "this is almost right — the first line is perfect, but the call to action feels generic. rewrite just the CTA."
this method gets you to great output in 2 to 3 iterations instead of 10.
before you ship something, paste it into Claude and ask:
"what's weak about this? where does it lose the reader? what would you cut?"
this is one of my favorite uses. Claude is surprisingly honest when you ask it to critique — and it catches things you stop seeing after staring at the same copy for too long.
you can also ask it to play a specific role:
"you are a skeptical CMO seeing this for the first time. what's your reaction?"
UX copy is one of the highest-leverage uses for Claude. button labels, error messages, empty states, onboarding prompts, tooltips — all of it.
the key is giving it enough context:
- what does this screen do
- who is the user and what are they feeling right now
- what action do you want them to take
- any constraints (character count, tone, platform)
example:
"write 5 options for the button label on a payment confirmation screen. the user just completed their first crypto transaction. they might be nervous. the brand tone is calm and reassuring. max 3 words."
writing a creative brief from scratch is slow. using Claude to structure it is fast.
tell Claude:
- what the project is
- who the audience is
- what the goal is
- any constraints or references
ask it to output a structured brief. then edit it down to what's actually useful. you'll cut the time it takes to write a brief by half.
Claude.ai has a feature called Artifacts — when you ask Claude to create a structured document, table, plan, or piece of content, it will generate it in a separate pane that you can copy, download, or share.
ask Claude to produce things as Artifacts when you want a clean output:
- "write this as an artifact"
- "create a campaign calendar as a structured document"
- "generate this table as an artifact i can copy"
great for when you need to hand something off to a client or teammate.
instead of opening 10 tabs, ask Claude.
it's not perfect for real-time data (use web search for that), but for background research, synthesis, and summarizing — it's fast. give it a topic and ask for a structured summary with key points.
for more current info: Claude.ai has a web search feature. turn it on in the sidebar and it will pull live results into its answers.
if you're working with technical teams, clients, or stakeholders who don't speak the same language — Claude is great at translation in both directions.
paste in a technical doc and ask: "explain this to a non-technical marketing director"
paste in a marketing brief and ask: "rewrite this so an engineer understands what we actually need"
this alone is worth the subscription.
when you find a prompt that works — save it.
i keep a Notion doc with my go-to prompts organized by use case: writing, research, critique, brainstorm, briefs. when i start a new project, i pull from that doc instead of writing from scratch.
some prompts i use constantly:
- "you are a senior creative director at a brand agency. critique this campaign concept."
- "give me 20 name ideas. half should be playful, half should feel premium."
- "rewrite this for a Thai audience — keep the meaning, adapt the cultural tone."
Claude is really good at high-volume ideation. use it for naming sessions.
the trick: give it constraints. "no more than 2 syllables" or "no tech jargon" or "should feel like a person, not a product" narrows the output and makes it actually useful.
ask for 20 options at a time. rate them. tell it which direction feels right. run another round. you'll usually find something usable in 3 to 4 rounds.
paste in competitor website copy, ad copy, or product descriptions and ask Claude to analyze them:
- "what positioning is this brand using?"
- "what's the tone and who is the intended audience?"
- "where are the gaps in this brand's messaging that we could own?"
combine that with your own positioning and ask: "given this landscape, what's the clearest space for us to own?"
Claude is useful for:
- turning rough notes into structured slide outlines
- writing the narrative thread between sections
- anticipating questions and preparing answers
- editing exec-level language for board decks
tell it who the audience is and what you need them to walk away believing. it will shape the structure around the goal.
Claude learns from examples, not just rules.
if you want Claude to write like a specific brand, paste in 3 to 5 examples of that brand's writing. if you want it to match a specific tone, show it that tone. if you want it to avoid something, show it the thing you're trying not to do.
"write like this" works better than "write with a warm, approachable, but professional tone that doesn't feel corporate."
one of my most-used daily habits: before i send anything important, i run it through Claude.
paste in the message you're about to send and ask: "does this land the way i intend it to? what could be misread?"
or tell it the situation and ask for 3 draft versions with different approaches: "direct," "warm," "neutral."
if you don't have a brand voice doc yet, Claude can help you build one — from scratch, in a session.
start by giving it a brain dump:
- who are you
- who is your audience
- what do you want people to feel
- what words or phrases feel like you
- what do you definitely not want to sound like
ask it to synthesize that into a brand voice doc with:
- tone description
- do / don't lists
- example sentences in your voice
- rewrite examples showing the wrong tone vs the right tone
then upload that doc to your Project (tip 1) so Claude always writes with it in mind.
markdown is a lightweight way to format text — headers, bullet points, bold, links — that Claude natively understands and produces.
even if you're not technical: writing your briefs, strategies, and content plans in markdown makes them easier to work with in Claude, copy into Notion, and hand off to developers.
quick reference:
# big heading
## section heading
**bold text**
- bullet point
[link text](url)
before a difficult client call, a feedback session, or a negotiation — use Claude to prep.
describe the situation: who's involved, what you're trying to achieve, what the tension is. ask it to help you anticipate objections and prepare responses.
ask it to role-play the other person. practice the conversation. it sounds strange but it works.
Claude can catch its own mistakes if you ask it to.
after it gives you something, try:
"double check every claim in what you just wrote. make a table of what you were confident about vs what you were uncertain about."
or more simply: "are you sure about this? what might be wrong here?"
don't ship Claude's output without reviewing it — but this method helps you review smarter.
Cursor is a code editor with AI built in. and yes, it's designed for writing code. but non-technical people can use it too — especially for editing websites, landing pages, and product copy that lives in a codebase.
if your website is built on Webflow, Framer, or a custom stack: your developer can give you access to the repo and you can make copy edits, small layout tweaks, and minor changes without touching the core code.
you don't need to know how to code. you need to know how to describe what you want.
the best way to start any Cursor session is to use Cursor chat (Cmd+L) before you make any changes.
ask: "what does this file do?" or "where does the homepage headline come from?"
let the AI orient you before you start editing. this prevents broken things.
in Cursor chat, you can type @filename to pull a specific file into the conversation.
instead of describing what you're looking for, you can say:
"@index.html — change the headline in the hero section to 'build something real'"
this is more precise than asking generally. Cursor knows exactly where to look.
before asking Cursor to edit anything, ask it to explain first:
"what would happen if i changed the headline font size here?"
or:
"explain what this block of code does in plain english"
if its explanation makes sense, then you can ask for the change. if it doesn't, ask more questions.
never ask AI to change things you don't understand — even small ones.
this is the most practical use for non-technical people. if your landing page is in a git repo, you can open it in Cursor and ask it to:
- update copy across multiple pages at once
- change colors or fonts (if they're defined in one place)
- reformat a section layout
- add or remove a component
what would normally require a Slack message to engineering, a ticket, a PR review, and a deploy can sometimes just be a 5-minute Cursor session. obviously still review everything before you push changes live.
this is a combo i use constantly.
Lovable ❤️ is a vibe-coding tool that lets you build web products from natural language prompts. Cursor is the code editor that lets you edit and refine those products at the file level.
workflow:
- build the initial product in Lovable ❤️ (describe what you want, iterate with prompts)
- export the code or connect the GitHub repo
- open in Cursor for more precise edits and refinements
- ship it
this stack lets non-technical people go from idea to live product without writing a single line of code from scratch. at Junohauz's SheBuilds event on IWD, women shipped live products in 3 hours using JunoMoment, our vibe-coding product — this is exactly the kind of workflow they used.
something on the site broke. you don't know why. your developer is unavailable.
open Cursor. paste in the error message (check browser developer tools — cmd+option+i on mac, look for the Console tab). ask:
"i'm seeing this error: [paste error]. what's causing it and how do i fix it?"
Cursor will usually identify the problem and show you exactly what line to change. you'll still want a developer to review before pushing anything to production, but you can at least diagnose the problem.
if your website copy is hardcoded (not in a CMS), editing it usually requires going into the codebase, finding the right file, making the change, and deploying.
Cursor makes this faster. use Cursor's search (Cmd+Shift+F) to find the exact text you want to change. then ask Cursor to update it — either across one file or across multiple files at once.
especially useful for: legal copy, pricing pages, email templates, app strings.
config files, environment variables, CMS schema files, metadata — these are things non-technical people usually can't touch without engineering help.
with Cursor, you can open those files, ask what they do, and make changes that Cursor explains step by step. it's still important to review before pushing, but the barrier is much lower.
you don't need to know Python or JavaScript to ask Cursor to write you a simple script.
things creatives can automate with Cursor:
- resize a folder of images
- rename a batch of files with a consistent naming convention
- convert a CSV into a formatted table
- pull data from a spreadsheet and format it for an email
describe what you need in plain language. Cursor will write the script. ask it to explain what it does before you run it.
the biggest mistake people make: they describe what they want in abstract terms. instead, show examples.
paste in a piece of writing that has the tone you're going for. share a campaign you wish you'd made. describe a brand you want to feel adjacent to.
"write something that feels like Glossier's early brand voice but for a Web3 audience" gives Claude a lot more to work with than "write something approachable and modern."
this is borrowed from improv. instead of asking Claude to start over when it misses, build on what it gave you.
"yes, and — keep that opening line, but push the second paragraph further. make it feel urgent, not just informative."
this keeps the momentum of a good first pass and steers it toward what you actually want. faster than starting from scratch every time.
searching is passive. collaborating is active.
when i use Claude well, it's a back-and-forth. i bring an idea half-formed, it pushes back, i redirect, we iterate. it's closer to a working session with a junior strategist than a google search.
this shift in mental model changes how you prompt — and how good the output gets.
this is important to remember.
Claude and Cursor will generate things. a lot of things. competent, structured, sometimes even impressive things. but they don't have taste. they don't know what's resonant or what's forgettable. they don't know your audience the way you do.
your job isn't to prompt and paste. your job is to curate, shape, and decide. the creative judgment is still yours.
AI raises your floor. it doesn't replace your ceiling.
the best creatives i know treat their AI workflow like a creative process — not a productivity hack.
that means:
- having intentional inputs (good references, clear briefs)
- protecting space for iteration
- knowing when to slow down and when to ship
- not treating every output as a draft to polish, but as raw material to work with
speed is a feature. but craft is still the goal.
the same way you keep a swipe file for good copy or design — keep one for prompts.
when a prompt gets you an output that actually impressed you, save the prompt, not just the output. you want to be able to recreate the conditions.
a simple Notion page organized by category (copy, research, critique, ideation) is enough. add to it every time something works.
Claude is great at summarizing, synthesizing, and surfacing patterns. it's less reliable for original analysis, opinion, or judgment.
use it to get up to speed faster on a topic. use your brain to figure out what it means.
the best research workflows i've seen: AI handles the information gathering, the human handles the interpretation. that division makes both better.
something real is happening. the gap between an idea and a live product, campaign, or piece of content has collapsed.
a solo designer can now ship a working prototype. a solo strategist can now run a research sprint in an hour. a solo founder can now write, design, build, and distribute without a full team.
this isn't about replacing teams. it's about what's now possible before you even need one.
we're building Junohauz on exactly this premise — and SheBuilds is where we put it into practice. non-technical women, live products, three hours. that's the world we're in now.
the tips in this doc are starting points. the most useful workflow you'll have is the one you build yourself — based on the work you actually do, the tools you're already using, and the gaps that slow you down.
spend a little time every week just experimenting. try a new prompt structure. try a tool you've been avoiding. ask Claude to help you figure out how to use Claude better (it's surprisingly good at this).
the compounding effect of small workflow improvements is real.
every time i share something i've figured out, i learn something back.
someone tells me a faster way to do it. someone tells me they tried it and it didn't work for their context. someone builds on it and shows me something i hadn't thought of.
if these tips helped you, pay it forward. write up what you figured out. post the prompt that changed your workflow. share the thing you wish someone had told you earlier.
that's how this community gets smarter together.
this repo is maintained by Mewtwo — founder of Junohauz, unofficial HQ for AI and Solana women.
Junohauz runs SheBuilds events where non-technical women ship live products using AI tools. this repo is the documentation of what actually works.
pull requests welcome. if you have a tip that belongs here, open a PR or reach out on X.
make me juno 🌙