Task statement 3.1 · Product and model selection
3.1 · Choosing Product Features
In Wrenfield Housing, the course example, the communications team has four jobs in one week: a quick question about a newsletter heading, the tenant letters it writes every week, evidence on how other housing providers handle damp and mould, and the spring newsletter's layout for the printer. Each job fits a different Claude feature, and the wrong one costs time or lets the house style slip.
share
The tempting habit is to do everything in a plain chat, because it needs no setup. That works for a one-off question. For recurring work, someone must paste the same files in every time, so in the week somebody forgets, the draft drifts from the house style. The opposite habit fails too: if you turn on Research, a mode that runs many searches, for a fact already in your file, you wait for a long report.
KnowledgeFour features and the job each one fits
The exam guide names four product features. A chat is a plain conversation: Claude answers from what you type or attach and from the conversation so far. A Project is a self-contained workspace with its own chats and two settings. Project knowledge is the files you add to it, and project instructions are the guidance that applies only to chats inside it. Research, the apps' name for the guide's "research mode" (1.4), investigates a question with searches that build on each other and returns a cited report. An artifact, a standalone document beside the chat (2.6), holds content you will edit, keep or share. This rules out any answer that treats the four as interchangeable ways to get the same reply.
Research and artifacts both start from a chat: you turn Research on from the + menu, and an artifact opens beside the conversation. So choose by what the job needs beyond one answer:
- Nothing: a plain chat, because a one-off question needs no setup.
- The same instructions and files every time: a Project, so every chat in it starts from them.
- Many outside sources: Research, so Claude searches widely and cites each claim.
- A result to edit, keep or share: an artifact, refined over several turns.
Interactive · one week, four jobs, four features
Choose a feature for each of the week's four jobs, starting with the choice that went wrong.
Pasted every time: each week's letters started in a new chat with the style guide pasted in, so in the week someone forgot it, the drafts drifted from the house style.
Quick question: one answer, read once, so a plain chat is the whole job.
Weekly letters: the same style guide and handbook every week, so they live in a Project and every chat inside it starts with them.
Damp evidence: the answer is spread across many outside sources, so Research runs searches that build on each other and cites them.
Print layout: the newsletter is edited, kept and sent to the printer, so it is an artifact, exported as a PDF.
A Project pays off when work recurs. Without one, every new chat starts empty, so each person must paste the style guide in again. Inside a Project, Claude uses the project knowledge and instructions in every chat, and on a Team plan you can share the Project with colleagues. Projects are on every plan, including Free, which allows five. Setting one up is taught in 5.1. The decision here is only whether the work recurs enough to need one.
Research earns its wait when an answer is spread across many sources. It needs a paid plan, such as Team, and web search, the quick lookup tool, switched on. It also searches connected tools such as Google Workspace. Because it gathers many sources, it can use up your usage limits faster. For a fact you already hold in a file, a plain chat with the file attached is faster.
An artifact suits content with a life after the chat. Claude creates one for substantial, self-contained content, typically over 15 lines, and you can ask for one by name. A Docs artifact exports to Word, PDF or Google Docs. Artifacts start private to you, and on a Team plan sharing stays inside the organisation by default.
KnowledgeOne week, four jobs
Here is the team's week, each job with the feature that fits and the wrong choice that tempts:
"For the boiler item, which heading is clearer: 'Boiler checks, 3–14 March' or 'Your boiler service is due'?"
Wrong choiceResearch: a long report for a one-line answer.
"Tenant comms", with the style guide and tenant handbook as project knowledge, and one instruction: "Follow the style guide in every letter."
Wrong choiceA new chat each week, with both documents pasted in by hand.
"How do other housing providers find and fix damp and mould early? Sources from 2024 on, a link for each claim."
Wrong choiceA chat with web search off: Claude answers from what it already knows, with no links.
"Lay out the spring newsletter as a one-page document: three items, boiler servicing first." Edited, then exported as a PDF for the printer.
Wrong choiceA long chat reply, copied into another program by hand.
SkillWorked example: the letters that drifted from the house style
Wrenfield's communications team writes tenant letters every week, such as repair appointment notices. Each must match the tenant handbook and follow the style guide: "Dear resident", sentences of 20 words or fewer, and the repairs line number at the end. Each officer started a new chat for the week's batch and pasted both documents in:
# before: a new chat each week [style guide pasted] [tenant handbook pasted] Draft this week's letters … → week 6, nobody pasted the style guide: "Dear Sir/Madam … in accordance with your tenancy obligations …", 34-word sentences, no repairs line # after: one Project, shared with the team Project: Tenant comms Knowledge: style guide, tenant handbook Instructions: Follow the style guide in every letter. → every new chat in the Project starts from both documents
The cause was the choice of feature, not the draft. The style guide reached Claude only when a person remembered to paste it, so the house style held only until someone forgot. A Project keeps the documents and the instruction in the workspace, so every chat in it starts from them, including a new joiner's first. The communications lead still checks each letter, as Wrenfield's AI policy requires, but now checks facts, not style.
Memory is not a shared Project. Memory, which lets Claude carry what it learns between your chats (3.4), is managed by each person for themselves. On a Team plan it stays off for each member until they turn it on (checked 2026-10-15). So memory cannot hold a team's style guide.
| Nothing: it is read once | Chat: ask, read the reply and move on. |
| The same instructions and files every time | A Project: put them in once, and every chat starts from them. |
| Evidence from many outside sources | Research: set the scope and dates, then open the cited sources. |
| A result to edit, keep or share | An artifact: work on the document beside the chat, then export it. |
- A file that must shape every chat belongs in a Project, not in each person's habit of pasting it.
- Research is slower and uses up limits faster, so save it for questions no single source answers.
- Research and artifacts both start in a chat; a chat on its own is right for anything read once.