Level, beginner

Getting useful
with Claude

Your first proper introduction, written for people who have used Claude once or twice and want to get genuinely useful with it in the next hour.

Most people's first Claude session is a mix of curiosity, a slightly vague question, and a slightly disappointing answer. This page is the stuff we wish someone had told us before our first session. It will take you about ten minutes to read. By the end, you'll have two or three concrete things to try this week and a much better sense of what Claude is actually good at.

The five-minute primer

Five things to know before you even open Claude. None of them are complicated, and most of them are common sense once someone says them out loud.

1.

Just talk to it like a person. There is no correct syntax, no secret commands, and no magic words. Full sentences, half sentences, and "can you do this for me" all work. If you can brief a junior colleague on a task, you can prompt Claude.

2.

How to get in. Your Abingdon lead will have added you to the team account. Log in at claude.ai with your work email and you will land in a chat window. Each conversation is a fresh memory by default, so for anything substantial, consider using a Project where you can upload reference docs once and have them follow you around.

3.

It will sometimes be wrong with confidence. Claude is very good but not infallible. If the output is load-bearing, a client email, a piece of SQL you're about to run, a number you're about to quote, check it. Treat it like a clever colleague's first draft, not a finished answer from an oracle.

4.

Think before you paste. Useful rule: don't put anything into Claude you wouldn't put into a client email without thinking about it. Our team account is covered by proper enterprise terms, so your chats are not training public models, but client-identifying data and anything sensitive still deserves a moment's pause.

5.

The quality of what comes out depends almost entirely on what you put in. This is the thesis of the whole page. We'll come back to it in a minute.

What "wrong with confidence" actually looks like

A deliberately silly example to show the failure mode. Claude rarely produces anything this bad now, but the shape of the mistake is worth recognising.

An output that sounds plausible

The 1976 Apollo moon landing was the first to collect rock samples from the lunar highlands, and the mission commander Frank Borman famously described the experience as "a small step for man, a giant leap for mankind".

Two things wrong: the Apollo programme ended in 1972, so the 1976 mission never existed. The "one small step" line is from Neil Armstrong on Apollo 11, not Frank Borman. Both errors are delivered with total confidence. This is the failure mode to watch for. It is almost never this obvious in real work, which is exactly the point.

The one big idea: context is everything

Almost every bad Claude experience I see comes from the same place. Someone types a one-line prompt, gets a vague one-size-fits-all answer, and concludes the tool is a bit rubbish. It isn't. It just had nothing to work with.

Claude is extraordinarily good at giving you exactly what you ask for. The trick is that "exactly what you ask for" is wildly different from "what you actually wanted". Closing that gap is almost entirely about context: what you're doing, who it's for, what good looks like, and where the landmines are.

Same task, two prompts

Click between the two tabs. The prompt changes. More importantly, so does the output.

Youwrite me a weekly status update for my team
What Claude sends back

Hi team,

Here is this week's update. We've been making good progress on our Q2 initiatives and the team has continued to deliver against our commitments. Exciting times ahead as we work through the pipeline.

Let me know if you have any questions. Thanks for all your hard work.

YouI need a weekly status update for my team of 8 engineers. We are midway through a quarterly migration project, moving our reporting system to a new vendor. Here is last week's update as a voice reference: [PASTED EXAMPLE] Here is what happened this week, in bullets: - 3 of 12 report templates migrated (Q2 forecast, SLA tracker, cost variance) - auth layer blocker, resolved Thursday - Jo out sick Wed and Thu, back Friday - decision needed on batch vs per-report migration approach Draft this week's update matching the voice of the reference. Under 200 words. Flag the decision at the top, not buried. Do not invent progress I have not mentioned.
What Claude sends back
Reporting migration, week 6
Decision needed on Monday

Batch migration vs per-report. My recommendation is batch, reasons in #eng-migration. Want to settle it at standup.

Shipped this week
  • 3 templates migrated: Q2 forecast, SLA tracker, cost variance. 9 to go.
  • Auth blocker resolved Thursday, docs updated in the project wiki.
People

Jo was out Wed-Thu, back Friday and picked up the cost variance template.

Next week

Templates 4 to 6, plus a spike on the batch approach before the Monday decision.

The difference between a good and a bad Claude output is almost always the input.

The four elements of a good prompt

Every prompt worth typing has four things. You can include them in any order, in any voice, with as little ceremony as you like. But include all four and you are already in the top decile of prompters.

Anatomy of a strong prompt

The four elements, shown on a single everyday example.

YouI need to write a first draft of a new starter welcome email for my team. 1 Here are three previous welcome emails I was happy with, pasted below. 2 Please write an equivalent email for our new starter, Jamie, who joins on Monday as a product analyst. 3 If anything about Jamie or the role is unclear to you, ask me before guessing. 4
1
Context. Who, what, for whom. Without this, Claude is guessing at half the answer.
2
Example. A concrete "this is what good looks like". Worth a thousand words of description.
3
The ask. What you actually want, phrased in the shape you want it back.
4
Escape hatch. Permission to flag uncertainty rather than guess. This is where most trust is won or lost.

Hold onto those four. The habits in the next section are all ways of making one or more of them easier.

Six habits worth building

If you only take one section from this page, let it be this one. Six small habits. None of them require any technical knowledge.

1. Ask Claude to ask you questions

The single highest-leverage line you can add to any prompt. At the end of any non-trivial request, type "feel free to ask me as many follow-up questions as you like before starting". Claude will genuinely do it, and the questions are usually the ones you hadn't thought to answer.

Planning a team offsite

A non-technical example. Same prompt, but ending with permission to ask questions.

YouHelp me plan the quarterly offsite for my 15-person team. Before you suggest anything, feel free to ask me as many follow-up questions as you like.
Claude's follow-ups
1. What's the actual goal? Team bonding, strategic planning, both, or something else I haven't thought of?
2. How many days, and is travel involved? Are people flying in, or is everyone local already?
3. What's the budget ballpark? Even a rough number will change what I suggest significantly.
4. Any non-negotiables on the logistics side? Accessibility needs, dietary requirements, "no work sessions past 6pm", that sort of thing.
5. What worked, or didn't, at the last offsite? If there's something people specifically liked or hated, I want to know before I suggest anything similar.
6. Who owns what after we decide? Someone needs to book the venue, someone needs to run the content, someone needs to chase the catering. Is that all you, or can you delegate?

Notice what happened: you didn't have to know the right questions up front. Claude surfaced them for you. That's the move.

2. Show it before and after

If you have an example of the input you're starting from and an example of the output you want, paste both. This is the single fastest way to get Claude to match a format without explaining the format in words, which you will always do badly.

Cleaning up a messy vendor list

Weak prompt on the left tab, strong on the right. Notice what changes in the output.

Youclean up this list for me
What Claude sends back MegaCorp Inc - 45000 Tiny Widgets Co - 3200 (March) Acme Holdings - 12500 Q2 Beta Services - 8700 Feb 15

Slightly tidier, but you still have four different date formats, currencies are gone, and Claude had to guess at what "clean" meant. You rewrite the whole thing by hand anyway.

YouHere is a raw vendor list, pasted below: MegaCorp Inc., 45000, 2024 Tiny Widgets Co - £3200 - March 2024 Acme Holdings Ltd £12,500.00 (Q2 2024) Beta Services - 8700 - 2024-02-15 Here is the format our finance team needs it in: vendor, amount_gbp, date_iso Convert the first into the second. Show me the first three rows so I can sanity-check before you do the rest. If a row is missing a field, flag it rather than inventing a value.
What Claude sends back

Here are the first three rows. Check these and let me know if I should continue with the rest.

vendor, amount_gbp, date_iso MegaCorp Inc., 45000, 2024-??-?? Tiny Widgets Co, 3200, 2024-03-?? Acme Holdings Ltd, 12500, 2024-??-??
Two things to flag: two rows are missing month or day. I've used "??" placeholders rather than inventing a date. Also, MegaCorp's amount had no currency symbol. I've assumed GBP to match the target format. Confirm and I'll do the rest.

The "show me the first three rows" line is a quiet superpower. It means you catch mistakes on three rows, not four hundred.

3. Ask Claude before you do the task

The most useful habit in this whole list, and the hardest to build. Before you start anything that will take more than about ten minutes, ask Claude if it can do it for you, or help you do it faster. Not after you're halfway through and frustrated. Before.

Renaming 300 files

A rough reconstruction. The point isn't the exact words, it's the shape of the conversation.

YouI need to rename about 300 image files from "IMG_xxxx.jpg" to "ClientProject_2024-03_xxxx.jpg". I was about to do it by hand in Finder but figured I'd ask first.
ClaudeGood call, that's exactly the kind of thing that will take you 40 minutes by hand and 30 seconds with the right approach. I can either walk you through a one-line terminal command, or write a small script with a dry-run preview so you can see what it would do before it actually does anything. Which sounds more like your speed?
YouThe script. I don't want to accidentally rename a thousand files.
ClaudeSensible. Quick question before I write it: the "xxxx" at the end, are those already in the original filenames and we want to preserve them, or do you want me to renumber them sequentially?
YouAlready in there, preserve them.
ClaudeGot it. Here's the script with a --dry-run flag that prints what would happen without actually doing anything. Try it with --dry-run first, spot-check three or four of the outputs, then run it properly.

Forty minutes of manual work became a two-minute conversation and a script. This is the whole game.

4. Use Claude to write your prompts

This feels silly and it is, in a good way. Type a rubbish version of what you want and ask Claude to rewrite it into a proper prompt before running it. It sounds circular, but Claude is much better at writing prompts than most of us are at typing them cold.

Three prompts, rewritten

Click between the examples. These are realistic rewrites, not templated fill-in-the-blanks.

"write me an email"
↓
Claude's rewriteDraft an email to Sarah Chen, our head of comms, asking for approval to reuse last month's customer testimonial in the Q2 marketing campaign. The testimonial is pasted below. Tone: friendly and professional. We have a good working relationship so not too formal. Keep it under 80 words. If comms approval normally goes through a form rather than an email reply, ask me about that before I send anything.
What changed: Names the recipient and the relationship, states the specific ask, attaches the thing being asked about, caps the length, and flags a process uncertainty rather than assuming.
"summarise this document"
↓
Claude's rewriteAttached: a 15-page vendor proposal. I need a one-page summary for my manager, who is evaluating us against four other vendors. She cares most about: pricing structure, implementation timeline, and references from similar-sized customers. She is going to read it in 90 seconds before the meeting. Lead with the strongest point, flag the weakest, and include one line on where we sit relative to the alternatives. Do not invent anything, stick to what is in the proposal. If something she cares about isn't covered in the document, say so explicitly.
What changed: Picks an audience, names what they care about, sets a reading-time budget, and gives Claude permission to say "not in the source" instead of making something up.
"help me with this analysis"
↓
Claude's rewriteAttached: a CSV of last quarter's customer support ticket volume by category. I'm trying to work out why the "billing" category spiked in March. Specifically, I want to know two things: (1) is it concentrated in one customer segment or spread across the base, and (2) does it correlate with any of our launches or policy changes in the same month (list attached). Walk me through what you would look at first, then actually look at it. If the data is not enough to answer either question, tell me what else you would need before going further.
What changed: Frames the actual question, lists the two sub-questions, asks for thinking first and then doing, and leaves an exit ramp if the data is insufficient.

5. Share what works with the team

The fastest learning happens sideways, not top down. If you stumble on a prompt or pattern that saved you an hour, post it in the team channel. Messy is fine. "Look what Claude just did for me" is the whole format.

RM
Rachel M. #ai-wins 10:42 this morning

Look what Claude just did for me. Pasted my rough notes from the Monday standup and asked it to pull out every action item with an owner and a date. It caught the two I'd forgotten to assign. 40 minutes saved plus I actually feel on top of my follow-ups for once.

Actions from your Monday standup: 1. Jo to review the draft SLA by Fri 22 Mar 2. Priya to confirm vendor pricing by Thu 21 Mar 3. [NEEDS OWNER] follow up with legal on the data-sharing clause 4. [NEEDS OWNER] book the Q2 review room for 5 Apr ... plus 8 more with clear owners.
12 clapping hands 5 heart 3 mind blown

Rachel didn't write a training doc. She posted a screenshot. Within a day, four other people on her team had tried the same pattern. That's how the good stuff travels.

6. Use it outside work too

This feels optional and isn't. The fastest way to build real intuition is to use Claude for things that don't matter. Plan a holiday. Plan meals. Draft the awkward email to your landlord. A month of low-stakes use will do more for your work prompts than any training session.

Four non-work examples

Click between them. None of these will make you better at your job directly. All of them will make you better at prompting.

YouPlan me a Saturday in Bath. I like old buildings, good coffee, and not too much walking. Budget around £80 for the day. Finish with somewhere decent for an early dinner.

Comes back with three stops, a specific coffee stop, and a 6pm dinner recommendation. You haven't booked a trip faster in your life.

YouMy sourdough starter is bubbling but not really rising. It's been active for five days. I'm feeding it a 1:1:1 ratio at room temperature. What am I doing wrong?

It asks about hydration, flour type, and room temperature. You realise the answer is probably "you're not doing anything wrong, wait another day". Sometimes that's the most useful output.

YouI have a fiddle leaf fig, a snake plant, and a pothos. I travel for work every other week. Help me work out a watering schedule I can actually stick to.

A schedule for each plant plus a "what to do before and after you travel" note. More thoughtful than whatever you would have written down yourself.

YouHelp me write a polite email to my landlord about a leaky tap I've reported twice already. I don't want to sound aggressive but I want it fixed this week.

A short email that's firm without being rude, with a specific ask and a deadline. You send it. The tap gets fixed on Wednesday.

The test isn't whether Claude is useful at any one of these. It's whether, a month from now, you reach for it without thinking. That reflex is the whole point of this section.

Six prompts to steal

Copy any of these, replace the bits in square brackets with your real inputs, and run them.

Try this week

Pick one. Try it within the next seven days. Come back here and tick it off.

If you take one thing from this page, let it be the bit about context. Everything else is a variation on the same theme: the more you tell Claude about what you're doing and why, the better the thing that comes back.

When you are ready to level up, head to the Intermediate guide, or jump into your role page for examples closer to your actual work.