Claude.ai offers several distinct surfaces for getting work done, and picking the wrong one costs time, accuracy, or both. A plain chat reply is the default: a self-contained answer read once in the conversation itself. A Claude Project adds persistent context — custom instructions and a knowledge base of reference material — available across many separate conversations, not just the current thread. Research mode has Claude actively search and synthesize across multiple external sources over several steps, producing a sourced answer grounded in current information rather than only what it already knows. An Artifact is a separate panel for long-form, self-contained, or iteratively-edited content that benefits from living outside the back-and-forth chat flow. And plain structured output, like a table rendered directly in a reply, suits data meant to be scanned and compared rather than read as prose.
These aren't just five ways to dress up the same answer — each solves a different problem, and the exam tests recognizing which problem a given scenario actually has.
A Decision Framework
Work through these questions roughly in order:
- Does this need to be available across many future conversations, not just this one? A style guide, a client's background, a standing set of product facts — anything you'd otherwise have to re-paste every time you start a new chat — belongs in a Project's custom instructions or knowledge base.
- Does answering require finding and citing information from multiple current external sources, rather than what Claude already knows or what's already in front of it? That's research mode.
- Is the output itself long-form, self-contained, and something the user will keep revising or reuse outside the conversation — a report, an essay, a working prototype? That's an Artifact.
- Is the output fundamentally comparative or tabular — rows and columns meant to be scanned side by side rather than read as prose? Ask for structured output like a table, which can live inline in chat or inside an Artifact depending on length.
- None of the above? A plain chat reply is the right, lightweight default — not every substantial answer needs a heavier feature wrapped around it.
Artifacts, Specifically: Long-Form and Iterative Content
Two signals point toward an Artifact rather than a plain reply: the content is long-form and self-contained (a report, an essay, a case study meant to be read or shared on its own), or it's something the user plans to keep iterating on rather than reading once. A 400-word case study meant for stakeholder review and several rounds of wording revisions is a textbook case: it's long-form, self-contained, and meant for iterative editing. For interactive content, like a small playable prototype, the Artifact panel keeps the running version visible and playable while accepting live edits — useful for playtesting, where you want to tweak one rule and immediately see the update without losing the running version or scrolling back through chat history.
A short, conversational answer meant to be read once and not revised doesn't need an Artifact. Length alone isn't the trigger — self-containment and the intent to iterate are.
Layering Features Together
These features aren't mutually exclusive. A conversation inside a Project might use research mode to gather current external information, then place the synthesized findings into an Artifact for the user to keep refining. The point of the framework above isn't to pick exactly one feature forever — it's to make a deliberate choice for each part of the task, instead of defaulting to plain chat for everything (losing persistence and sourcing you actually needed) or defaulting to an Artifact for everything (adding overhead to answers that were fine read once).
Key Concept
Match the feature to the actual need: Project for context that must persist across future conversations, research mode for multi-source current information with citations, an Artifact for long-form or iteratively-edited self-contained content, structured output for comparative/tabular data, and plain chat as the lightweight default otherwise. These layer together rather than being an either/or choice.
Common Exam Distractor
Watch for research mode offered as the answer when the question doesn't actually require live, multi-source external sourcing — it adds real latency for no benefit if Claude's existing knowledge already covers it. Watch equally for a Project offered for genuinely one-off information with no future reuse, and for "put it in an Artifact" offered for a short answer with no self-containment or iteration need.