Home › Study guides › CCAO-F › Domain 3 › Lesson 3.1

CCAO-F · Domain 3 · 12% of the exam · Lesson 3.1 · 19 min read

Choosing the right Claude feature for the job

When to use a plain chat, a Project, research mode or an artifact in Claude, what each wrong choice costs, and how the exam tests the match.

Written against objective 3.1 of the official CCAO-F exam guide (Version 1.0, effective July 2026). An independent resource, not affiliated with Anthropic; the practice questions are written from scratch.

3.1.1 Why one chat box is not one tool

It is Monday at Kestrel & Pine, an outdoor-gear retailer, and Tomasz, the marketing team lead, has five jobs on his list. He wants a quick answer on how long an email subject line should be. The autumn brand posts need drafting again, and every one must follow the brand guide. Leadership wants a competitor landscape report that draws on dozens of sources. Finance has asked for a campaign budget calculator they can try their own numbers in. And before Wednesday's planning call, he needs to know which topics are trending this week.

He does what most people do: he opens one chat and works down the list. The subject-line answer is fine. The brand posts come out in a pleasant voice that isn't quite Kestrel & Pine's, because the brand guide is sitting in last week's chat, not this one. The competitor report reads well but cites nothing, and it describes the market as Claude learned it in training, not as it stands this month. The calculator arrives as a block of text he has to copy somewhere. And the trending topics come from Claude's general knowledge, because it never searched the web.

Claude wasn't the weak link here; the choice of tool was. The Claude apps offer several ways of working, each built for a different shape of job. A chat suits a request you can brief in one go. A Project suits work that repeats with the same rules and files. Research mode suits a question that needs many sources. An artifact suits a result you will keep, edit and share. The fifth job, the trending topics, needs no feature of its own, only a quick web search.

3.1.2 A plain chat is the right default for a one-off

Here is a belief worth dropping early: that a plain chat is the beginner's option, and serious work always needs a Project or research switched on. For a large share of everyday requests, the plain chat is exactly right. There is nothing to set up, nothing to carry over, and the answer arrives in seconds.

What makes a job a good fit for a chat is that it is self-contained. You can put everything Claude needs into one message, and none of it will matter next week. Tomasz's subject-line question passes that test. Look at how the whole brief, down to what to do if there is no single answer, fits in five short lines.

Quick question for a promotional email I'm sending today.

Context: we're an outdoor-gear retailer. The email goes to existing customers, and most of them read email on a phone.

Question: how long should the subject line be so it isn't cut off on a phone screen?

Output: a character range, one sentence on why, and two example subject lines for our autumn rain-jacket sale that fit the range.

If the answer depends on the email app, say so rather than giving one number.

The test takes a few seconds. Could you brief this job fully in one message? Will you need the same rules or files again? If the answers are yes and no, open a chat. A chat can still use the supporting capabilities you will meet below, such as a quick web search, so a one-off that needs one current fact stays a one-off.

3.1.3 Projects: when the same rules apply every time

The brand posts are a different kind of job. Tomasz writes them every week, and every week they need the same things: the brand guide, the approved product names, the phrases the brand never uses. In plain chats he has to paste all of that in each time, and the week he forgets, the posts drift.

A Project is a workspace with its own chats, its own reference files (called project knowledge) and its own instructions. You upload the files once, such as the brand guide and the product list, and write instructions that shape Claude's answers in every chat inside it. Each new chat in the Project starts with those files and rules available to Claude, so your brief shrinks to what is new this week: "three posts for the new trail-running range, one per channel".

Think of someone who hikes every weekend. The map, first-aid kit and rain shell stay packed in the rucksack by the door, and each Saturday they add only the day's food. A Project is that packed rucksack for recurring work: the reusable context is stored once and brought into every chat, and you add only what has changed.

Projects are available on every plan, including free accounts, which can have up to five. On Team and Enterprise plans, a Project can be shared with colleagues, with view or edit access, so everyone on Tomasz's team drafts from the same guide instead of their own copy.

3.1.4 Research mode: when the answer needs many sources

The competitor report fails for a different reason. Each Claude model learns from data that stops at a cutoff date. Without a search, Claude may not know this season's prices or new ranges, and it gives you no citations to check. A web search fixes that for a fact that one or two searches can settle. A landscape of a dozen competitors, their pricing, new ranges and reviews, needs far more.

Research mode (the Research feature in the apps) is built for that job. Claude runs many searches that build on each other, decides what to look into next, and works through open questions until it can write a longer report with citations you can check. It searches the web and, through any connectors you have set up, your own apps such as Gmail, Google Calendar and Google Docs. A connector is a link between Claude and an app you use, and it lets Claude read that app's data within your own access rights.

Research takes minutes rather than seconds, and it is available on the paid plans (Pro, Max, Team and Enterprise). It counts against the same usage limits as any conversation, meaning the amount of Claude use your plan allows, but a research session uses them up faster.

Research has two close relatives that people mix it up with, and Tomasz's own trade offers a picture of all three. Checking tomorrow's forecast takes one look. Planning a three-day route means cross-checking maps, trail reports and hut bookings. Choosing the best line on a map already in your hand needs no new information at all, only careful thought. Those are web search, research mode and extended thinking, where Claude takes longer to reason before it answers.

Memorise the middle column: the names and switches in the apps change, but the jobs don't. Web search, for instance, is a switch in some versions of the app, while in the newest one Claude searches by itself when it helps. Either way, a searched answer carries citations. If a current fact comes back without one, Claude answered from training. Citations make checking easy, but they don't do the checking for you.

3.1.5 Artifacts: when the result is something to keep

Finance's budget calculator runs into yet another problem. As an ordinary answer, it is text in the middle of a conversation. Tomasz copies it into an email, finance asks for a change, and soon there are two versions and nobody is sure which one is current.

An artifact is a substantial piece of work that opens in its own window beside the conversation: a document, a dashboard, a diagram or a small interactive tool. You can change it by asking Claude, come back to it later, and share it. Claude makes one on its own when the content is substantial and self-contained, likely to be edited or reused outside the chat, and something you will refer back to. You can also ask for one directly. Artifacts are available on every plan, and each one starts private to you until you share it.

For Tomasz, the calculator becomes an artifact with an input for each channel's spend and a running total against the season's budget. Finance opens it, tries their own numbers and asks for one more column, and there is still one current version. If finance would rather have a file, Claude can create one too, such as a spreadsheet with working formulas or a slide deck, to download and keep refining. That capability, code execution and file creation, is also on every plan, and artifacts rely on it.

3.1.6 Match the job, not the habit

Which feature is best? None, in general, and that is the heart of this objective. The tempting mistake is to standardise: always switch on research "to be thorough", put everything in one Project, or never leave the plain chat. Each habit fits some jobs and costs you on the rest, as the last column shows.

Real work mixes these. Tomasz's campaign has stages: research mode for the competitor landscape, the Project for the weekly drafting, an artifact for the calculator. The features also combine: you can ask for an artifact inside a Project, or pair research with extended thinking when a big question needs careful planning as well as wide searching. A good plan varies the feature by stage instead of forcing one mode on the whole workflow.

The apps are changing too: chat and Claude Cowork, built for longer multi-step tasks, are merging into one conversation where Claude can decide which tool to use, starting with Pro and Max plans. Knowing what each capability is for lets you ask for the right one, notice when Claude reached for the wrong one, and judge what comes back.

3.1.7 The exam traps

Every trap here is a mismatch between a job and a feature. It runs in two directions: too little feature for the job, or far too much.

• ✗ Running recurring work in fresh plain chats and pasting the rules in each time. ✓ Put the rules and files in a Project, so every chat starts with them.
• ✗ Switching on research mode for a single fact. ✓ Use web search. Research spends minutes and extra usage on what one lookup can do.
• ✗ Asking a plain chat, with no search, for a current report built from many sources. ✓ Use research mode. Without a search, Claude answers from training data that stops at a cutoff date, and gives you nothing to check.
• ✗ Leaving a deliverable as chat text that gets copied around. ✓ Ask for an artifact, so there is one version to edit and share.
• ✗ Standardising on one feature for a whole workflow. ✓ Vary the feature by stage: research to gather, a Project to repeat, an artifact to deliver.

3.1.8 Put it together: pick the feature before you start

You now have the four features, the job each one fits and the supporting capabilities around them. Trying each one on your own work turns the decision table into a habit, and taking a Project's key ingredient away shows you what it was doing all along.

The rest of Domain 3 turns from features to models, the second half of the same choice. It covers how Haiku, Sonnet and Opus differ (3.2), how to match a model to cost, speed and quality (3.3), and when to restart, summarise or persist context (3.4). In Domain 5, configuring a Project's instructions and knowledge (5.1) and managing connectors (5.2) turn today's quick setup into something a team can rely on.

Key takeaways

• ✓ Match the feature to the shape of the job, not to habit or to the most powerful option available.
• ✓ A plain chat is the right default for a self-contained request you can brief in full in one message.
• ✓ A Project keeps files and instructions together so every chat in it starts with the same context; use it for recurring work, and share it on Team and Enterprise plans.
• ✓ Research mode runs many searches and returns a cited report in minutes on paid plans; one current fact needs only web search, and hard reasoning on what you have needs extended thinking.
• ✓ An artifact is a substantial deliverable that opens beside the chat, where you can edit it, return to it and share it.
• ✓ Vary the feature by stage of a workflow, because the most elaborate option wastes time and usage on simple jobs.

Check your understanding

4 questions written for this lesson, then one from the CCAO-F question bank on the same topic. Every answer option is explained, including the ones you did not pick. Nothing is stored.

Sources

• When should I use web search, extended thinking, and research?
• Use research on Claude
• What are projects?
• What are artifacts and how do I use them?
• Claude Cowork and chat are one Claude
• How up-to-date is Claude's training data?
• The CCAO-F exam guide, summarised