The cross-app brief: from email to spreadsheet to slides in one session

What changed in May 2026

On 7 May 2026, Anthropic launched Claude as a native add-in inside Excel, Word, and PowerPoint, with Outlook in public beta. The cross-app context feature is the part that changes how knowledge work actually flows: Claude keeps the same conversation thread as you switch between Office apps. What you told it in Outlook, it still knows when you open Excel. What it found in the spreadsheet, it carries into PowerPoint without being asked.

This pattern is drawn from Anthropic's official guide to working across Microsoft 365 apps, published alongside the May 2026 general availability launch.

The immediate implication: you stop copying and pasting between apps to keep the AI informed. You write one brief at the start of the session and let context carry the rest.

The problem this solves

Most knowledge workers produce deliverables that cross at least two or three file types. An analyst receives a brief by email, builds or updates a model in Excel, writes a summary in Word, and drops the key numbers into a PowerPoint for an exec review. Each handoff between tools has historically meant starting over with the AI: pasting in context, re-explaining who the audience is, reminding it what the data showed.

With cross-app context, the session state persists. Claude does not forget what it read in the email when you switch to the spreadsheet. This makes the brief — the single block of context you give at the start — worth investing in properly.

The prompt

Open this prompt in Claude at the start of your session, before you open any file. Fill in the bracketed fields. Everything after this point — Excel analysis, Word prose, PowerPoint slides — inherits this context automatically.

Context for this session
=======================
Deliverable: [what we are producing — e.g. Q2 trading-update deck and summary note]
Audience: [role, seniority, what they care about most — e.g. CFO and two board members who want variance to plan, not raw numbers]
Source data: [file or sheet name, what it contains — e.g. "Q2-actuals.xlsx", sheet "P&L", columns Revenue / Cost / EBITDA by business unit]
Key message: [the 1–2 things the audience must leave knowing — e.g. EBITDA beat plan by £1.2m; Services growth offset Products decline]
What to avoid: [sensitivities or out-of-scope items — e.g. do not mention the restructuring programme; headcount figures are embargoed]
Tone: [brief description — e.g. direct, board-level, no jargon]

Session plan
============
Step 1 — Excel: Analyse the P&L sheet. Identify the top 3 variances to plan, flag any line that moved more than 10% month-on-month.
Step 2 — Word: Draft a 400-word summary note. Lead with the key message. Use numbered findings. One recommendation at the end.
Step 3 — PowerPoint: Build a 5-slide deck. Slide 1: title and key message. Slides 2–4: one finding per slide with supporting number. Slide 5: recommendation and next step.
Step 4 — Outlook: Draft a covering email to [name/role] with the key message in the first sentence and a link placeholder for the deck.

Do not begin Step 1 until I open the Excel file and confirm you can see it.

Why each piece earns its keep

Audience and key message up front. Claude calibrates vocabulary, detail level, and emphasis for the rest of the session. An exec summary for a CFO differs from a working note for a finance business partner; stating it once means every output is calibrated without repeated instructions.

What to avoid. Embargoed figures, sensitive topics, and out-of-scope items are easy to miss when Claude is pattern-matching across a spreadsheet. Naming them explicitly at the start prevents them appearing in a slide or subject line.

The session plan. Listing the four steps tells Claude the arc of the whole session. When you reach Step 3 (PowerPoint), it already knows the findings from Step 1 and the framing from Step 2; it is not starting from scratch.

The confirmation gate. The final line — "Do not begin Step 1 until I open the Excel file and confirm you can see it" — prevents Claude from starting work on file contents it has not actually seen. It is a small approval surface that catches mismatched assumptions before they propagate into three deliverables.

What the cross-app context does in practice

Once the brief is in place, the typical session runs like this:

You switch to Excel. Claude can see the sheet. You confirm. It runs the analysis, names the top variances, and flags anomalies. You do not have to re-paste the audience context or the key message — it already knows.

You switch to Word. Claude opens the summary note with the correct tone, the right level of detail, and the findings it just surfaced. You review and edit. Tracked changes are native.

You switch to PowerPoint. Claude builds the deck. The slide structure matches the findings from Excel and the prose from Word, not a generic template. Speaker notes are included because you asked for them in the brief.

You switch back to Outlook. The covering email opens with the key message from the brief in the first sentence. The audience name is correct. The deck is referenced with a placeholder link.

The whole cycle — analysis, write-up, slides, email — runs in a single session thread, driven by one brief written once.

Failure modes

An underspecified key message. If you write "summarise the results" instead of naming what the audience must leave knowing, Claude will surface everything equally. You will get a comprehensive summary and a lot of editing. Spend the extra ninety seconds on the key message field.

A brief that grows mid-session. Do not add instructions mid-session that contradict the opening brief — Claude will attempt to reconcile them and the output may become inconsistent. If the brief needs to change, state the change explicitly: "Revised: the key message is now X. Please apply this to all remaining steps."

Assuming cross-app context is unlimited. Context has a practical ceiling. Very long spreadsheets or deeply nested PowerPoint files may require Claude to work section by section. If it signals uncertainty about what it can see, treat that as a confirmation gate: clarify before proceeding.

Skipping the review step. Cross-app context does not remove the need to check each output before moving to the next. A wrong figure in Step 1 propagates into Step 2 and Step 3. The brief does not make the outputs infallible; it makes them faster to produce and easier to review because the context is consistent.