Phase 3 · Claude · Level 1 · Foundations
Claude's limits, and how to work around them
By the end, you'll be able to…
- Spot when Claude might be out of date, and know when web search helps and when it doesn't
- Guard against confidently wrong answers with the quote-and-check habit
- Carry context between chats deliberately, using a Project where it earns its place
Why it matters
Knowing where a tool falls short is what turns you from a nervous beginner into a confident, careful user. This lesson is honest about three things Claude can't do, and, more usefully, gives you a workaround prompt for each. None of these are reasons to avoid Claude; they're reasons to use it well.
Why this lesson matters
Claude is useful, but it isn't magic, and it lets people down in predictable ways. The good news is that the three big limits below each have a simple, practical workaround. Learn them and you'll trust the right things and check the rest, which is the whole difference between a beginner who gets burned and one who quietly gets a lot done.
Limit 1: it may not know recent events
Claude learns from information up to a point in time, its knowledge cutoff, the date after which it hasn't been taught anything new. Ask about something from last week and it may not know, or worse, it may fill the gap with a confident guess. Different Claude models have different cutoffs, and none of them is "today".
The fix built into Claude is web search: it can look things up and give you links. Where it's available it fills the recency gap and shows you sources to open. But web search is a starting point, not a guarantee; pages can be wrong too, so you still open the links and confirm.
Before you answer: is this based on your training data, which may be out of date, or on a live web search? If you searched, give me the source links so I can check them. My question is: [your time-sensitive question].
Why this works: Instead of assuming, you make Claude state whether it's working from possibly-stale training or from a live search, which tells you exactly how much to trust the answer and whether to check the links.
Workarounds:
- For anything recent or time-sensitive, don't assume it's current. Ask directly, as above.
- Use web search for current facts, then open the links it gives and confirm.
- For breaking news, a normal search engine is often the better first stop.
Limit 2: it can be confidently wrong
This is the big one. Claude sometimes states things that are plainly incorrect, in the same calm, fluent voice it uses for everything else. It can invent a quote, a figure, or a source that looks completely real. This is hallucination: the tool producing plausible-sounding text that simply isn't true. It happens less when Claude works from a document you've given it, but even then it can misread or overstate.
The habit that protects you is to make Claude show its evidence, and then check it. When it's working from a document, ask for the exact line. When it's making a factual claim you'll act on, verify it independently. And know that asking "are you sure?" is not a real check; Claude can cheerfully be wrong a second time.
You said the policy requires two office days a week. Quote the exact line that says so. If there's no line that states it directly, tell me plainly that you inferred it.
Why this works: Demanding the source line converts a smooth assertion into something checkable. If the quote isn't actually in the document, you've caught a slip in seconds instead of repeating it in a meeting.
Workarounds:
- For anything that matters (health, money, legal, or facts you'll act on) check it against a trustworthy source.
- Working from a document, ask Claude to "quote the exact line", then confirm the quote yourself.
- Don't rely on "are you sure?" It isn't a real check. Verify externally instead.
- Beware loaded questions. Asking "why is this clause illegal?" can lead Claude to justify a claim that isn't true. Ask neutrally: "Is this clause legal, and how would I check?"
Limit 3: each chat starts fresh
Claude remembers what you've said within a single conversation, but a brand-new chat is a blank slate. It won't recall your last chat, your preferences, or the document you shared yesterday, unless you give them again. This surprises people who expect it to "remember", but it's also why chats stay clean and private to their own thread.
For one-off tasks, just keep a topic in one chat and return to that chat to continue. For work you'll come back to repeatedly (a project running over weeks, where you keep pasting the same background) the answer is a Project: a workspace that holds related chats and reference files together, so the same material is available in every chat inside it without re-pasting. Projects are available on the free plan too, with paid plans allowing more of them; the official guide to Projects has the current detail.
This is a new conversation, so here's the background you need before my question: [paste the key facts, decisions, or document]. Now, given that, help me with the following: [your task].
Why this works: When you can't return to the original chat, this hands Claude the background explicitly at the top, so a fresh conversation isn't working blind. It's the manual version of what a Project does automatically.
Workarounds:
- Keep one topic in one chat, and reopen that chat to continue it.
- For recurring work, set up a Project so the background carries over automatically.
- Starting fresh anyway? Paste the key context back in; Claude isn't psychic, and it can't see your inbox, files or calendar unless you hand them over.
Try it now
Common mistakes
- Assuming it knows today's news. Check date-sensitivity, use web search or a search engine, and open the links.
- Trusting a confident answer on something that matters. Confidence is not accuracy. Verify the important things independently.
- Relying on "are you sure?" as a check. It isn't one; Claude can be wrong again just as confidently.
- Expecting it to remember a previous chat. It doesn't. Continue the same chat, re-supply context, or use a Project.
- Over-trusting Claude precisely because it writes so well. The fluency that makes it a great drafting partner is also what makes a wrong answer sound right. The polish is in the language, never a promise about the facts, so the better the writing, the more deliberately you should check anything load-bearing.
Keeping current
These limits are fundamental to how the tool works, so they'll persist, but the details soften over time: knowledge cutoffs move forward, web search and memory features change, Projects gain abilities. For what's true today, see the Claude release notes and the help centre. Accurate as of 13 July 2026.