- In short
- Choosing among Research Mode, Chat web search, and Artifacts turns on the task pattern. A quick fact check or single current data point is a web-search-in-Chat task. A question that requires synthesis across many sources before the answer can be trusted is a Research Mode task. When the output itself is the deliverable the recipient will open and read, it belongs in an Artifact. Reaching for a heavier tool than the task needs wastes time without improving quality.
Three surfaces, three task shapes
Once you know the four entry points exist, the harder skill is telling three of them apart in the moment: a quick lookup, a deep investigation, and a deliverable. The CCAO-F exam tests this at the analyse level because the surfaces overlap in appearance -- all three can involve external information or produced text -- but differ sharply in the effort they carry. Choosing the heavier tool for a light task is a common, quiet mistake.
The distinction runs on task shape. How many sources must be reconciled before you can trust the answer, and is the output something to be handed over or just read once? Answer those and the surface is determined. This knowledge point pairs with the worth-building test, since a recurring version of any of these may also warrant a Project.
- Research vs web search vs Artifacts
- A selection rule among three surfaces: a quick fact check or single current data point is a web-search-in-Chat task; a question requiring synthesis across many sources before it can be trusted is a Research Mode task; and output that is itself the deliverable a recipient will open and read belongs in an Artifact. Using a heavier tool than the task requires costs time without improving quality.
Web search in Chat versus Research Mode
The line between web search in Chat and Research Mode is about depth, not about whether the internet is involved. Both can reach current information. What separates them is how much synthesis the trustworthy answer requires.
A quick fact check -- a single current figure, one recent event, a single-source lookup -- is a web-search-in-Chat task. Chat can fetch the current information and answer directly, and that is enough. Research Mode, by contrast, is for questions where no single source suffices and the answer cannot be trusted until Claude has gathered and reconciled many sources. A competitive landscape, a multi-angle investigation, a synthesis across a dozen reports: those justify the heavier, multi-step machinery of Research. The mistake to avoid is invoking Research for a single-fact, single-source question that Chat web search already answers in one step.
Availability differs between the two, and it is worth knowing before you plan a workflow around either. Web search in Chat is offered across all Claude plans as a per-chat toggle in the input, though on Team and Enterprise plans an Owner has to enable it for the workspace before members can switch it on. Research, the deeper multi-source mode, is a paid-plan capability rather than a universal one. So the lighter tool is also the more broadly available one, which reinforces defaulting to Chat web search for anything a single lookup can settle.
When the output is the deliverable
Artifacts sits on a different axis from the other two. Web search and Research are about how the answer is found; Artifacts is about what happens to the output. The question to ask is whether the output itself is the deliverable -- the thing the recipient will open, read, and possibly share as a standalone document.
If yes, it belongs in an Artifact, rendered cleanly for review rather than left as inline chat text a reader must scroll to reconstruct. A polished report, a structured brief, a formatted summary destined for someone's inbox: all Artifact material. The trap here is producing exactly such a deliverable as inline chat text when it will be reviewed as a standalone document. Note that these axes combine -- a Research investigation whose output is a formatted report is a Research task whose result lands in an Artifact.
What the CCAO-F exam trips candidates on
Two over- and under-reaching errors are tested. The first is reaching for Research Mode for a single-fact, single-source question that Chat web search already answers. The scenario will describe a genuinely simple lookup; the credited answer keeps it in Chat with web search, because Research adds steps and time without improving a one-source answer.
The second is producing a polished report as inline chat text instead of an Artifact when it will be reviewed as a standalone document. The tell is that the output is meant to be opened and read by someone as a finished piece. The credited answer places it in an Artifact. Both traps reward matching the tool's weight to the task's actual shape rather than defaulting up or down.
Worked example
A marketing lead has three tasks: confirm one competitor's launch date announced yesterday, assemble a synthesised view of every competitor's moves over the last quarter for a planning session, and turn that synthesis into a formatted brief the leadership team will read. Which surface fits each?
Each task has a different shape, so each takes a different surface. Confirming one competitor's launch date is a single current fact from essentially one source. That is a web-search-in-Chat task: Chat fetches the current information and answers in a step. Reaching for Research Mode here would be the classic over-reach, adding multi-step machinery to a one-source question.
Assembling a synthesised view of every competitor's moves over a quarter is the opposite shape. No single source gives a trustworthy answer; the value comes from gathering and reconciling many sources. That is exactly what Research Mode is for, and web search in Chat would under-serve it.
Turning the synthesis into a formatted brief that leadership will read is an output-axis decision. The brief is the deliverable someone will open and review as a standalone document, so it belongs in an Artifact, not as inline chat text. Notice the last two combine: the investigation runs in Research, and its result is rendered as an Artifact. Matching each task to the lightest surface that fully serves it is the whole skill.
Common misreadings to avoid
Misconception
Any question that touches current information should use Research Mode.
What's actually true
Misconception
A finished report is fine as inline chat text as long as the content is good.
What's actually true
How this shows up on the exam
Questions describe an information task or a deliverable and ask which surface fits. Ask two questions: how many sources must be reconciled for the answer to be trusted, and is the output something handed over? One source means Chat web search; many means Research; a handed-over deliverable means Artifacts. Watch for the distractor that escalates a simple lookup to Research.
This knowledge point builds on the four entry points and feeds the recurring-workflow scenario and combined entry-point and model matching, where surface choice is decided alongside model tier for the same task.
An analyst needs to verify a single interest-rate figure published this morning, then later compile a synthesised market outlook from many reports that leadership will read as a formatted document. Which pairing is correct?
People also ask
When should I use Research Mode instead of web search in Chat?
When does a task belong in an Artifact?
Is Research Mode overkill for a single fact check?
Watch and learn
Official Anthropic Academy lessons first, then hand-picked walkthroughs. Videos load only when you press play.
No videos curated for this concept yet
We are still curating the best official and community videos for this topic.
Official prep for this domain
Anthropic's own free prep module for this part of the syllabus, on the official prep course. Free with an Anthropic Academy sign-in.
References & primary sources
Master this concept with Archie
Practice it inside an adaptive study session. Archie, your Socratic AI tutor, tracks your mastery with Bayesian Knowledge Tracing and schedules the perfect next review.