Product and Model Selection: CCAO-F domain 3 study guide

CCAO-F · Product and Model Selection (12% of the exam)

Product and Model Selection is domain 3 of the Claude Certified Associate Foundations exam, 12% of the scored items. It tests whether you pick the right Claude feature and model for a task, balancing quality, speed and cost, and whether you know when to restart, summarise or keep context in a longer piece of work.

What the official guide covers

The Claude Certified Associate Foundations exam guide (version 1.0, effective July 2026) lists four tasks under domain 3, Product and Model Selection:

What the guide lists	What it means in practice
Select product features (Projects, research mode, chat, artifacts)	Know what each feature does and the kind of task it suits
Differentiate between model types (Haiku, Sonnet, Opus)	Know which tier is fastest and lowest cost and which is most capable
Align model selection with task needs (cost, speed, quality)	Use a lighter model for simple, high-volume work and a stronger one for complex reasoning
Manage context limits and memory (when to restart, summarise or persist)	Know what fills a conversation, what happens in long chats, and where context should live

What each Claude feature is for

Feature	What it does	Use it when
Chat	A single conversation with whatever you paste or attach	One-off questions and short tasks
Projects	A workspace with its own chat history, uploaded knowledge and instructions that apply to every chat inside it	You return to the same documents or the same kind of task again and again
Research	Claude runs several searches that build on each other and returns an answer with citations, usually in minutes. Web search must be on. With Google Workspace connected it can also search Gmail, Calendar and Docs	A question needs many sources, not one quick lookup
Artifacts	Substantial, self-contained output (a document, a diagram, a small tool) shown next to the chat, private by default and publishable to share	You will edit, reuse or share the output
Connectors	Give Claude access to apps such as Google Drive or Gmail, within your own permissions	The information or action lives in another tool
Skills	Folders of instructions and resources that Claude loads only when a task needs them, in any chat or Project. Anthropic provides Skills for Excel, Word, PowerPoint and PDF files; you can upload your own. Code execution must be on	A repeatable procedure should run the same way each time
Code execution and file creation	Claude writes and runs code in a sandbox and can create .xlsx, .docx, .pptx and .pdf files	Figures must be calculated, data cleaned, a chart drawn, or a real file produced
Memory and chat search	Claude keeps topics from past chats and can search earlier conversations	Your role and context should carry over between chats
Incognito chat	A chat outside any Project that is not saved to your history or to memory	A one-off conversation you do not want shaping later chats

Is a Project worth building?

Ask three questions. Does the task recur? Is the background the same each time? Is the output format the same each time? If two or more answers are yes, build the Project; the setup time comes back within a few sessions. A sign you have waited too long: each new chat starts with you pasting in the background paragraph you already pasted last week.

Four layers, each solving a different problem

The features stack. A Project carries the context, a Skill carries the procedure, code execution makes the numbers calculated rather than estimated, and memory carries continuity between chats. Use any combination: a one-off question needs none, a monthly report with figures might use all four. None of them removes variation. Claude can phrase the same report differently each run even with a Skill, so review stays in the workflow.

How the model tiers differ

Anthropic's model guidance describes three tiers in general terms:

Tier	Strength	Typical work
Haiku	Fastest and lowest cost	High-volume, straightforward tasks: short replies, tagging, simple extraction
Sonnet	Balance of speed and capability	Everyday work: drafting, analysis, content creation, reading images and charts
Opus	Most capable	Complex reasoning, long multi-step work, tasks needing careful judgement

In claude.ai you change the model with the model selector next to the send button. You can change it mid-conversation; the change applies from Claude's next response. Two further settings sit alongside the model. Effort controls how thorough Claude is: higher effort gives fuller answers but takes longer and uses more of your allowance. Thinking lets Claude work through the problem before it answers, and you can expand its reasoning to read it. The help centre suggests defaults for routine work and higher effort or thinking for complex tasks such as planning.

Matching the choice to the task

Situation	Choose	Why
200 short, similar replies from a template	A faster, lower-cost model at default settings	Speed and cost matter more than deep reasoning
A strategy memo weighing trade-offs across several reports	A more capable model, higher effort or thinking	The task needs careful reasoning
A market question that needs 15 sources	Research	It searches in several rounds and cites its sources
A question about your own company's past decisions or policies, on a Team or Enterprise plan	Enterprise search	It searches the company tools your organisation connected, shows only what you can already access, and cites them
A weekly task using the same policy files	A Project with the files as knowledge	Upload once, use in every chat
A one-page plan you will edit and share	An artifact	Sits beside the chat, can be revised and published
Classifying 500 agreements against a clear, fixed rubric	A faster, lower-cost model	Structured work at volume; nothing ambiguous to interpret
A board paper drawn from four reports with conflicting signals	The most capable model, in a Project with the reports	Ambiguous inputs and a high-stakes reader
Compound interest or variance figures for a report	Code execution, on any capable model	A calculated figure beats an estimate written as text

How context and memory work

Everything in a conversation shares one context window: your messages, Claude's replies, attached files and any project knowledge that is loaded. When a long chat nears the limit, Claude summarises earlier messages so the conversation can continue (this needs code execution turned on), and the full history stays visible to you. You may see Claude "organizing its thoughts" while this happens. Longer chats also use more of your usage allowance.

Warning signs that a chat needs attention: Claude stops following an instruction it followed earlier, replies only to your last message and ignores earlier decisions, or makes errors that a forgotten early detail would explain. There are three responses. Restart when the chat has drifted or the next task is new. Summarise first when the work continues: ask for decisions, work in progress and open questions, check it, and paste it into the new chat. Persist anything every future chat needs by saving it to project knowledge or memory.

Context can be kept in three places:

Where	What it keeps	Note
The current chat	Everything said so far, summarised when long	Older detail can be compressed
A Project	Uploaded knowledge and instructions for every chat in it, plus its own separate memory	On paid plans, large project knowledge switches to retrieval so more fits
Memory	Topics about you and your work, carried into future chats	View, edit or delete in Settings > Memory; pause or reset it; switch it off for one chat before the first message

When this happens	Do this
The chat is long and Claude refers to decisions you replaced	Ask for a summary of current decisions, check it, and start a new chat with it
You keep re-uploading the same files	Move them into a Project
The task is finished and the next one is unrelated	Start a new chat
Context should carry across many chats on one piece of work	Use a Project; it keeps its own memory, separate from your other work
Two clients' work must never mix	One Project per client; each Project's memory stays separate

A handoff prompt for restarting a long chat

We are going to continue this work in a new chat. Write a handoff note with:
1. The goal and who the output is for
2. Decisions we have made, newest first, and any we reversed
3. The current version of the draft, in full
4. Open questions and what I still need to supply
Do not include options we rejected unless I need to know why.

Check the note before pasting it into the new chat; any error in it carries forward.

Rules that decide exam answers

Match the model to the task, not to prestige. Simple, high-volume work goes to a faster, lower-cost tier; complex reasoning goes to a more capable one. "Always use the best model" is the tempting wrong answer.
Research is for many sources. A question needing broad, cited investigation calls for Research; a simple lookup does not.
Repeated work belongs in a Project. Re-uploading the same files to new chats is the wrong option when a Project would hold them.
Artifacts are for output you will reuse. Short answers stay inline; documents and tools you will edit or share become artifacts.
Calculate, do not estimate. Figures that will be used or reported go through code execution, whichever model you chose.
Restart with a checked summary. When a long chat drifts, carry forward a summary you have checked, not the whole history and not nothing.

Where it appears in the exam

Domain 3 carries 12% of the exam, around 7 of the 60 items. Questions describe a task with a constraint (volume, deadline, depth of reasoning, repeated files) and ask which feature or model fits, or what to do when long work loses context.

Two sample questions

These are original Timo practice questions. They are not official exam questions.

Question 1

A team lead in a Team plan workspace uploads the same five leave and expense policies into a new chat every week to answer staff questions. Three colleagues do the same. Which setup fits best?

A. A shared Project holding the five policies as knowledge, with answering instructions.B. The same weekly uploads, but with the most capable model chosen for every answer.C. Research turned on so that Claude searches the web for leave and expense rules.D. One long chat with all five policies pasted in once and kept open for the whole year.

Answer: A. A Project keeps the policies and instructions in one place for every chat, and it can be shared with colleagues. B keeps the repeated uploads, C searches the web instead of the company's own policies, and D builds a chat that grows long enough to be summarised.

Question 2

A policy analyst on a Pro plan must brief her director on how eight countries regulate electric scooters. The director wants every claim linked to a source she can open. Which approach fits best?

A. A normal chat with web search off, using the most capable model and higher effort.B. A Project holding last year's internal briefing, with instructions to update it.C. Research, with web search on, then open and check a sample of its citations.D. One quick web search per country in eight chats, then paste the answers together.

Answer: C. The question needs many current sources and checkable citations, which is what Research returns. A relies on training data that may be out of date, B starts from an old document with no new sources, and D gives up the planned, cross-checked search that Research runs.

Build exercise

Write down five tasks from your week. For each, choose a feature and a model tier from the tables above and note why.
Run one task on a faster model and on a more capable one at higher effort. Compare quality and time taken.
Ask the same broad question in a normal chat with web search and then with Research. Count the sources each cites and open three of them.

Practise this topic

Claude Certified Associate practice exam: free, 20 questions, no sign-up
CCAO-F study guide: all topics
Previous topic: Output Evaluation and Validation
Next topic: Workflow Integration and Solution Design

Sources

Claude Certified Associate Foundations Exam Guide, version 1.0, effective July 2026 (Anthropic), domain 3: Product and Model Selection
Anthropic documentation: Choosing the right model
Claude Help Center: Change the model, effort, and thinking settings
Claude Help Center: What are Skills?
Claude Help Center: Use Claude's chat search and memory to build on previous context