Getting more
out of Claude
For people who already chat with Claude regularly and want to stop getting the same middling outputs. Four ideas that take you from "useful" to "I honestly don't know how I did this job a year ago".
By now you've figured out that Claude is good at a lot of things. You've probably also figured out that it's good at completely different things on different days, and you haven't fully worked out why. This page is the bridge. The four things below are what separate the people who use Claude like a search engine from the people who use it like a collaborator.
1. Put your reference material in a Project
If you're still copy-pasting the same context into every chat, stop. Projects are Claude's answer to this. A Project is a workspace where you upload reference documents once and every conversation inside the Project already knows about them. Team style guides, onboarding docs, report templates, pricing sheets, brand voice, anything you keep having to explain.
Upload context to a Project once, and every chat inside it already knows.
The test for whether something belongs in a Project is simple: if you've pasted it into a chat more than twice, it belongs in a Project. Start small. Upload three or four documents. Set a one-sentence system prompt at the top ("you're helping me draft outputs in the voice of these reference docs"). Have one real conversation inside it. You will feel the difference within about four minutes.
If you've pasted the same thing into Claude more than twice, it belongs in a Project.
2. Use your connectors
Claude can connect directly to Slack, Google Drive, GitHub, your calendar, and a growing list of other tools. Connectors mean Claude can read your real data, not just whatever you paste. "Summarise the #incidents channel this week" stops being a copy-paste job. "Find the proposal Jo shared last Tuesday and draft an executive summary" stops being a ten-minute dig.
Pick a connector, see what you could ask
Four of the most common ones. Click between the tabs to see a realistic prompt for each.
Kills the "where was that link again" problem for good.
Turns a folder of documents into a finished thing you can send.
Useful whether you write code or just need to know what is changing.
Ten seconds on Sunday evening, Monday becomes tractable.
Your Abingdon lead can switch connectors on for your team. The best ones are usually the ones you didn't know you wanted. Spend twenty minutes clicking through the list and imagine one real use case for each. Three or four will be obvious. One will change how you work.
3. Before-and-after as your default habit
Beginners learn to add context. Intermediates learn that the best kind of context is an example. If you have a raw input and an example of the output you want, paste both. Every time. Even when it feels slower.
Turning raw meeting notes into a decision log
The format of the output matters as much as the content. Showing the format is faster than describing it.
The team discussed several items in today's meeting. They talked about the upcoming launch, reviewed progress on ongoing initiatives, and touched on some open risks. A few action items were agreed. Overall the meeting was productive and the team is aligned on next steps.
It's technically a summary. It is also useless: no decisions, no owners, no dates, no way for someone who missed the meeting to know what actually happened.
Here are the first two. Confirm these look right and I will do the remaining six.
date decision decided_by rationale action owner deadline 2026-04-11 Push launch by 1 week Whole team Q1 sign-off still pending Update launch comms Priya Mon 14 Apr 2026-04-11 Move weekly sync to Thursday Jo Clashes with exec review Wed Send new calendar invite Dan Tue 12 AprThe first few times, this habit feels like overhead. By the fifth time it's reflex, and it has quietly shaved hours off your week. Pair it with the line "show me the first three rows before you do the rest" and you've also eliminated the class of bug where Claude confidently gets four hundred rows wrong and you notice after you pasted it into the client report.
4. Know when Claude is wrong, and know how to recover
The single biggest skill that separates intermediate from beginner is the ability to spot when Claude is confidently wrong. It isn't rare. It isn't common either. But when it happens, a beginner accepts the output and moves on. An intermediate notices.
Three reliable tells:
- The answer is suspiciously clean. Real data has edges. If Claude has produced a table where every row fits neatly, go check a handful of rows against the source.
- The answer references something you didn't give it. If Claude cites a fact, a number, or a source you know you didn't paste, treat it as a guess until you verify it.
- The answer is vague in the one place that matters. If everything else is specific and one paragraph is hand-wavy, that's usually where Claude didn't actually know the answer.
A plausible-sounding output with a few problems
You asked Claude for a short summary of a competitor from public information. Here is what it gave you. Read it carefully before you scroll to the callout.
Northwind Shipping, founded in 2011, is a mid-market logistics company specialising in cold-chain transport across the UK and Northern Europe. They reported 34% year-on-year revenue growth in their most recent public filings, driven largely by their new Rotterdam hub which opened in early 2023.
Their main competitive edge is a proprietary route-optimisation engine that reduces delivery time by an average of 18% compared to industry benchmarks. They employ around 450 people and their CEO, Marcus Hanley, has a background in supply-chain consulting at McKinsey.
(1) The 34% year-on-year revenue growth. Claude did not cite a source, and you did not give it one. Treat as unverified until you pull the actual filing.
(2) The 18% delivery-time improvement. This is exactly the kind of specific-sounding number that gets quoted in decks. Confirm it came from Northwind themselves, not from Claude's general sense of what sounds plausible for the category.
(3) Marcus Hanley's background. Possibly true, but the "ex-McKinsey" detail is the kind of incidental fact Claude is most likely to confabulate. If you're going to mention it to your team, verify it on LinkedIn first.
When you spot something like this, the fix is almost never "argue with Claude". The fix is to add the missing context and run the prompt again. Or, better, ask Claude what it would need to know to do the thing properly, and provide exactly that. "What sources would you need me to give you to make this brief defensible?" is the single most useful follow-up question in the whole handbook.
When the output is wrong, the fix is usually more context, not a different prompt.
Six prompts to steal
These are sharper than the Beginner prompts. Copy, replace the bits in square brackets, run.
Try this week
One concrete action from each of the four sections above. Pick one.
If this page has left you thinking "I should be building things with Claude, not just chatting with it", the Advanced guide is for you. That's where Claude Code, the API, and proper agentic workflows live.
If instead you want to go deeper on your own work, your role page will have examples specific to what you do. Either is a good next step.