Home › Study guides › CCAO-F › Domain 2 › Lesson 2.6

CCAO-F · Domain 2 · 21% of the exam · Lesson 2.6 · 19 min read

Curating information and choosing the output format

How to sort a long chat into what the reader needs, then choose an inline answer, an artifact, structured data or a file by what happens next.

Written against objective 2.6 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.

2.6.1 One planning chat, four different outputs

Your professional association's annual conference is six weeks away, and most of the planning has happened in one long chat with Claude. It holds venue quotes (one of them revised), speaker details, a Saturday option the committee has since dropped, catering and audiovisual quotes, and timing constraints. Then four requests arrive on one afternoon. You need to know which venue is cheaper per attendee. The programme committee wants a run-of-show, the minute-by-minute running order for the day, to edit before Thursday. Finance wants the vendor quotes in the budget spreadsheet. The volunteer lead wants a checklist volunteers can tick off on the day.

The tempting move is one request: "Write up everything we have on the conference." You'll likely get a long, tidy summary that serves none of the four. The venue answer is buried in paragraph five, finance can't paste prose into columns, and nobody can tick a paragraph. Worse, it may still mention the dropped Saturday and the first venue quote, because nothing told Claude they were out of date.

Two decisions got skipped. The first is what goes IN: sorting the material, keeping what is current and cutting what this reader doesn't need. The second is what SHAPE it takes. An inline answer is an ordinary reply in the chat. An artifact is a self-contained piece of work that opens beside the conversation for you to edit and share. Structured data is laid out so a spreadsheet or other tool can read it. And some work has to leave Claude as a file.

2.6.2 Curate before you choose a format

It is tempting to pick the format first. Ask straight away for "the run-of-show as a document", and every keynote time the chat ever mentioned can end up in it. A format only arranges what it is given: unsorted material becomes an unsorted document with good headings.

Curating is the editor's work that comes before any writing. It has four moves:

1. GROUP related material under a few headings: venue, speakers, schedule, suppliers, open questions.
2. PRIORITISE what this reader needs: put it first, and decide what they don't need at all.
3. CUT repeats and out-of-date versions, so each fact appears once, in its latest form.
4. STRUCTURE what remains with headings, and with tables where items share the same details.

Think of handing your work to a colleague who covers for you while you're on holiday. You wouldn't forward your inbox; you'd give them the decisions, the open questions and who to call, each once. Curating does that for the reader, and for Claude too. In a long chat, the old quote and the new one both sit in front of Claude, and it can't be sure which one you treat as current unless you say.

You can have Claude do the sorting, then check the outline before anything gets written. Look at the third and fourth lines: they handle what changed during the chat, which is where long conversations go wrong.

Before we write anything, sort what we have discussed in this chat about the conference.
Group it under these headings: Venue, Speakers, Schedule, Suppliers, Open questions.
Where a figure or decision changed during the chat, keep only the latest version and note what it replaced.
Mark each item Confirmed or Pending. List anything we dropped at the end, one line each, so I can confirm it is out.
Each fact appears once, under one heading.
Show me a short outline, not a finished document. I will check it before we choose formats.

2.6.3 Inline answers and artifacts

"Which venue is cheaper per attendee?" is a question you'll read once, act on and move past. It needs an inline answer: a reply in the conversation itself, short enough to take in at a glance. Turning it into a document adds a step and nothing else. A good inline answer still has an order, though: the answer first, then the reason, then any assumption it rests on.

Question: At 320 attendees, which venue is cheaper per attendee?
Answer: The Grange Hotel, at about $62 per attendee, against about $84 for Riverside Hall.
Why: Riverside charges $14,800 for the day plus $38 a head for catering. The Grange charges $62 a head, all in.
Assumes: 320 attendees and Riverside's revised quote of 12 May.
Watch: the Grange has a 250-person minimum, so re-check if registrations drop below that.

The run-of-show is the opposite case. The committee will edit it, you'll revise it after Thursday, and people will work from it on the day. As a chat reply it scrolls away, and each revision arrives as a new copy further down. That is the job of an artifact: something Claude makes that opens in its own window beside the conversation, where you can edit it, come back to it and share it. Claude makes one by itself when the content is substantial, likely to be edited or reused, understandable without the chat, and worth coming back to. When it matters, don't leave that to chance: "Create the run-of-show as an artifact."

The volunteer checklist is an artifact of a different kind: a small interactive tool that people use rather than a document they read. Artifacts range from documents and diagrams to dashboards, single-page websites and tools like this one. On paid plans, an artifact can also store data between sessions, so the checklist remembers what was ticked. Artifacts need the code execution and file creation setting switched on for your account or organisation. Each one starts private to you, and your plan decides who you can share it with.

2.6.4 Structured data and files: output that leaves the chat

Finance asked for the vendor quotes "for the budget sheet". Ask for a comparison without saying where it's going, and Claude may write three good paragraphs. The caterer is "a little over $9,000 depending on numbers", the audiovisual firm "quoted a flat fee", the photographer "is still confirming". Pasted into a spreadsheet, that is one cell of text. Someone now retypes every figure, and retyping is where errors creep in.

When software reads the output, or you'll compare it with next month's version, you need structured data: one row per item, the same fields in every row, values in the same units. Think of a letter versus a form. A letter tells its story in whatever order suits the writer. A form puts each answer in the same box every time, so a hundred can be compared at a glance. Structured data is the form.

The everyday shapes are a table in the chat that you copy into a spreadsheet, and CSV (comma-separated values), plain text that spreadsheet apps open as columns. JSON, a text format that labels every value, is for software tools that expect it. Whichever shape you pick, the request does the work. Anthropic's prompting guidance says to be specific about the output format you want; for data, that means naming the columns yourself. Notice the last two lines: they say what to do with gaps and keep commentary out of the paste.

Turn the catering, audiovisual and photography quotes in this chat into a table for our budget spreadsheet.
One row per vendor we are still considering, using each vendor's latest quote.
Columns, in this order: Vendor, Service, Price basis (flat or per head), Price (number only, no currency sign), What is included, Quote date (YYYY-MM-DD), Status (Confirmed or Pending), Source (the email or document it came from).
If a value is missing, leave the cell empty rather than estimating it.
Give me the table only, with no text above or below it, so I can paste it straight into the sheet.

The same thinking covers anything tracked over time. If you send the committee registration numbers every Monday, ask for the same columns each week, and each new row lines up under the last one.

Sometimes the deliverable must leave Claude as a file: an Excel workbook for finance, a PDF programme for the printer. That's file creation. With the same setting on, Claude creates Word, Excel, PowerPoint and PDF files on every plan, and you can download them or save them to Google Drive. The difference from an artifact is where the work lives next. An artifact stays in Claude, where you keep editing and sharing it; file creation hands you a file to open in other apps.

2.6.5 Choose the format by what happens next

Given a need, which shape fits? The answer comes from purpose, not from what looks most impressive, and four questions settle it:

1. Who uses it? You alone, a committee, volunteers, a finance spreadsheet.
2. What happens next? Read and decide; edit and approve; paste into a tracker; use on the day.
3. Will it be edited or reused? Once-and-done can stay inline. Anything revised, reused or shared needs to stand on its own.
4. Does a person or a machine read it? People read documents; spreadsheets and tools read rows and columns.

Here is where the answers point.

Memorise the middle column. Feature names change; the purposes behind them don't.

One deliverable can also combine formats. The committee needs to edit the run-of-show, but the chair wants to see which timings are confirmed and which still wait on a speaker. Ask for one artifact: the editable schedule, followed by a short table of item, source, date and status. People edit the narrative; reviewers check the facts.

2.6.6 Make reusable outputs carry their evidence

Next spring, someone on the committee opens this year's run-of-show to plan the next conference. It looks finished: tidy headings, exact times, a named keynote speaker. Nothing tells them the keynote time was never confirmed, the catering figure was a quote from 12 May, or the room plan changed a week before the event. A reader who never saw the chat can't judge any of it, and the polish makes it look more settled than it was.

So anything others will reuse should carry its evidence, the way a carton of milk carries its date. Three details do the job:

• Source: where each important fact came from, such as the venue contract or a speaker's email.
• Date or version: when it was true, such as "quote of 12 May" or "version 4, 3 June".
• Status: whether it is confirmed, pending or dropped, or a draft rather than approved.

Ask for them as part of the format. The run-of-show gets a top line with its version, date and status ("Draft for committee review") and the evidence table at the end. The vendor table already has the columns. The inline venue answer needs only its "Assumes" line, because it is read once.

2.6.7 The exam traps

Every trap here picks a format for the wrong reason: it looks impressive, it's the default, or nobody asked what happens next.

• ✗ Asking Claude to "write up everything" from a long chat. ✓ Curate first: group, keep the latest version of each fact, cut repeats and dropped options.
• ✗ Turning a quick, routine question into a document. ✓ Answer it inline, with the reason and any assumption.
• ✗ Leaving a deliverable you'll edit and share as a long chat reply. ✓ Ask for an artifact, so it stands on its own and is edited and shared in one place.
• ✗ Writing figures that feed a spreadsheet as paragraphs. ✓ Ask for structured data with named columns, one row per item and consistent units.
• ✗ Treating a polished document as ready to reuse. ✓ Show the source, date or version, and status of each key item, so a later reader can judge it.
• ✗ Fixing a format problem with a more capable model or a more professional look. ✓ Say who uses the output and what happens next. The problem was purpose, not polish.

2.6.8 Put it together: turn one chat into the right outputs

You now have the whole sequence. Curate first, so only current, relevant facts go forward. Then let purpose pick the shape: inline, artifact, structured data or file, with source, date and status on anything others will reuse. To make it stick, watch it work, then take the curation away and see what the format does without it.

Formats come back in the next domain from another angle: selecting product features (3.1) weighs artifacts alongside Projects, research mode and chat as tools for a job, not as shapes for an output. And when you want the same shape every time, such as one layout for every vendor table, Project instructions (Domain 5) can carry that request so you stop retyping it.

Key takeaways

• ✓ Curate before you choose a format: group the material, put what the reader needs first, cut repeats and out-of-date versions, and decide what to leave out.
• ✓ An inline answer suits a quick, one-off question you read once in the chat; give the answer, the reason and any assumption.
• ✓ An artifact suits substantial, self-contained content you will edit, reuse or share, from a document to an interactive checklist.
• ✓ Structured data suits output a spreadsheet or tool will read, or that you track over time: one row per item, the same named fields, consistent units.
• ✓ File creation produces Word, Excel, PowerPoint or PDF files when the deliverable must leave Claude as a file.
• ✓ Choose by purpose: who uses it, what happens next, whether it will be edited or reused, and whether a person or a machine reads it.
• ✓ Reusable outputs carry their evidence: the source, date or version, and status of each key item.

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

• What are artifacts and how do I use them?
• Share artifacts
• Creating with artifacts (Claude 101)
• Create and edit files with Claude
• Prompting best practices
• What are projects?
• The CCAO-F exam guide, summarised