Most people who say "Claude isn't that useful" are using it like a search engine: a short question in, a short answer out. The people who get hours back each week use it differently - more like a capable new colleague who needs a good brief.
At BrightBit Digital, Claude is what we work with every day. We help UK businesses roll it out, train their teams, and build it into their systems. The same handful of habits separates the teams that get real value from those that give up.
Here they are, A to Z. No jargon, no code - just what to do and why it works.
A - Ask for the Finished Thing#
Don't ask for "ideas for a proposal" and then write the proposal yourself. Ask for the proposal. Claude can produce finished documents, tables, spreadsheets and even simple interactive pages (called artifacts). Start from something complete and edit it - that's where the time saving is.
B - Brief It Like a New Colleague#
Claude doesn't know your business, your customers or your preferences unless you tell it. A good brief covers four things: who it's for, what you need, why, and what good looks like.
❌ "Write a LinkedIn post about our new service."
✅ "Write a LinkedIn post for finance directors at UK firms of 50-200 staff, announcing our new payroll audit service. Friendly but not salesy, under 150 words, one clear call to book a call."
C - Give It Context, Not Just Instructions#
Attach the documents, paste the email thread, share last year's version. Claude works from what's in front of it. The more relevant material you give it, the less it has to guess - and guessing is where mistakes come from.
D - Define "Done"#
Tell Claude how you'll judge the result: "one page", "a table with these five columns", "every figure checked against the attached spreadsheet". A clear finish line is the single biggest improvement most people can make.
E - Show an Example#
If you have a format you like - a past report, an email you were happy with - share it and say "match this style and structure". One good example beats a paragraph of description.
F - Feed Back, Don't Start Over#
If the first answer is 80% right, don't open a new chat and try again. Tell Claude what to change: "shorter", "less formal", "the second section is wrong because…". It improves quickly with specific feedback.
G - Get Your Governance Sorted First#
Before your team uses Claude for real work, agree some simple rules: which Claude plan you use, what kind of data may go into it, and who's responsible for the settings. Business plans come with stronger privacy protections than personal accounts - make sure work happens on work accounts.
H - Watch for Hallucinations#
Claude can occasionally state something wrong with total confidence - an AI "hallucination". It's rarer than it used to be, but it happens. For anything factual that matters, add: "Mark anything you couldn't confirm." And check key facts yourself before they go to a client.
I - Send Images, Not Descriptions#
A screenshot of an error, a photo of a whiteboard, a chart from a report - attach it. Claude can read images, and it's faster and more accurate than you retyping what's on screen.
J - Keep Your Judgement in Charge#
Claude drafts, analyses and suggests. You decide. Anything that affects a customer, an employee or your money should have a person reviewing it before it goes out. That isn't a lack of trust in the tool - it's good practice with any new team member.
K - Remember Its Knowledge Has a Date#
Claude was trained on information up to a certain point, so it may not know about recent events, new prices or last month's regulations. For anything current, turn on web search or give it the up-to-date documents.
L - Use It on Long Documents#
Claude can read very long documents - contracts, reports, tender packs - in one go. Ask it to summarise, compare two versions, or find contradictions, and ask for quotes with page numbers so you can check each point quickly.
M - Connect It to Your Tools (Carefully)#
Claude can connect to other systems - your email, calendar, documents or CRM - through connectors built on something called MCP (here's a plain-English explainer). That's where it goes from answering questions to doing work. Start with read-only access, and add more only once you trust the setup.
N - Start a New Chat for a New Topic#
Long chats that wander across many topics get muddled. When you switch to a new task, start a new chat. Keep one chat per job, and Claude stays focused.
O - Say What Output You Want#
Format, length and audience change everything. "Bullet points for the board", "a two-line Slack message", "a table I can paste into Excel". If you don't say, Claude has to guess.
P - Use Projects for Repeat Work#
Projects let you store instructions and files once, so every chat inside starts with that context. Set one up for each recurring piece of work - "Client proposals", "Monthly reporting" - and stop pasting the same background every time.
Q - Let It Ask You Questions#
For bigger tasks, add: "Before you start, ask me any questions you need answered." Claude will spot the gaps in your brief, and you'll get a better first draft.
R - Ask It to Review Its Own Work#
After a draft, ask: "Review this as a sceptical client would. What's weak or unclear?" Claude is often a sharper critic of a draft than an author. Then ask it to fix what it found.
S - Keep Sensitive Data Out Unless You're Set Up for It#
Don't paste customer lists, staff records, passwords or anything confidential into Claude unless your company has agreed that's allowed on your plan. When in doubt, remove names and identifying details first - Claude rarely needs them to do the job.
T - Teach It Your Tone#
Share a few examples of your writing and ask Claude to match them. You can also save a writing style so it sounds like you every time - no more editing out phrases you'd never use.
U - Use the Right Model for the Job#
Claude comes in different models. Larger ones are best for complex analysis and hard problems; faster ones are fine for quick rewrites and simple questions. If a quick task feels slow, try a faster model. If an answer feels shallow, try a more capable one.
V - Verify the Numbers#
For calculations, ask Claude to work them out properly using its analysis tools rather than in its head, and to show the figures it used. Then sanity-check the result against what you'd expect.
W - Turn Repeat Tasks Into Workflows#
If you do the same task every week, write the perfect prompt once and save it - in a project, a shared document, or as a reusable skill. Your whole team then gets the same quality result, not just the person who's best at prompting.
X - Ask It to eXplain Its Choices#
When Claude recommends something, ask: "Why did you choose this? Give me three reasons." You'll catch weak reasoning early, and you'll learn something about the problem along the way.
Y - Don't Let It Just Say Yes#
AI assistants can be too agreeable. If you ask "is this a good plan?", you'll often hear yes. Instead ask: "What are the three strongest arguments against this plan?" You'll get a far more useful answer.
Z - Zero In on One Workflow First#
Don't try to "roll out AI" across the whole business at once. Pick one task that eats time every week - meeting notes, first drafts of proposals, answering common customer questions - get it working well, measure the time saved, then move to the next.
Your One-Page Checklist#
- Brief it like a new colleague: who, what, why, what good looks like
- Give context and an example
- Define "done"
- Ask for the finished thing, then give specific feedback
- Mark what it couldn't confirm; check key facts and numbers
- One chat per task; projects for repeat work
- Keep sensitive data out unless your setup allows it
- A person reviews anything that affects customers, staff or money
- Start with one workflow and measure it
If You're a Developer#
The same habits apply to Claude Code, Anthropic's tool for software work - with a few extras:
- Write a
CLAUDE.mdfile in your project: a standing brief with your conventions, commands and rules. It's the "Projects" idea for code. - Give it a finish line it can check - "the tests pass" beats "fix the bug".
- Keep permission prompts on for anything destructive, and review changes before they're merged.
Frequently Asked Questions#
What is the best way to write prompts for Claude?#
Brief it like a capable new colleague: say who the work is for, what you need, why, and what a good result looks like. Give it relevant documents and an example of the format you want. Clear, specific context matters far more than special wording or tricks.
Is Claude safe to use for business?#
It can be, if you use a business plan and agree rules on what data goes in. Business plans come with stronger privacy protections than personal accounts. Keep sensitive personal data out unless your company has confirmed its setup allows it, and have a person review anything important before it's sent.
Why does Claude sometimes give wrong answers?#
Like all AI assistants, Claude can occasionally state something incorrect with confidence, and its knowledge stops at a training date. Ask it to mark anything it couldn't confirm, use web search for recent information, and check important facts and figures yourself.
How do I get Claude to write in my style?#
Share a few examples of your writing and ask it to match them, or save a writing style so it's applied automatically. Specific feedback on the first draft - "less formal", "shorter sentences" - also helps it learn what you like within a conversation.
Where should a business start with Claude?#
Pick one time-consuming task that happens every week, such as meeting notes or first drafts of proposals. Get that working well, measure the time saved, and share the prompt with the team before moving on to the next task.
These habits get individuals great results. Getting a whole business there - the right plan, safe settings, connections to your own systems, and a team that knows how to use it - is what we do. BrightBit Digital specialises in Claude for UK businesses: our AI consultancy plans and runs the rollout, and our MCP integration service connects Claude to the tools your team already uses.
AI engineer at BrightBit Digital
