Field Guide · level 5 · Office work, by task

How do you write a report with AI inside Word or Google Docs?

20 min · as of Oct 2026

You will be able to

Draft a report section by section from your own sources inside the document, take AI edits as tracked changes or suggestions, and check every fact and figure.

Part of Check the workTime about 20 minutes, with the try-itChecked 16 Oct 2026

Three things to take away

1
Work in the document, one section at a time, from sources you give it. Brief it, agree an outline, then draft.
2
Make every AI edit visible before you accept it: tracked changes in Word, suggestions in Google Docs, or a saved copy to compare against.
3
Check every fact, number and name against the source yourself, outside the chat. The AI's own review is not the check.

As of Oct 2026 · read 16 Oct 2026

In your assistant

Checked on each vendor's own pages on 16 Oct 2026. Your plan, your licence or your IT team may change what you see, and features arrive in waves.

Same in all three

Each can draft from sources you give it and edit a section you name.
Each vendor says to review the result before you use it.
None of them makes the check for you. Keep a copy, trace every figure, read every change.

Claude

In the chat: ask for the report as a Claude Doc. Docs, Slides and Design work in any conversation, on every plan including Free, with one shareable link. Export to Word, PDF, Google Docs or PowerPoint. For a big job, attach the sources, and on the desktop app choose the folders Claude may use.
In Word: Claude for Word. It works beside your document and shares context with Claude for Excel, PowerPoint and Outlook. Turn on Track Changes in Word, and keep a copy. Anthropic: "Review every edit before accepting it." It does not recommend the add-in for "final client deliverables or counterparty sends without human review", and says "Only use Claude for Word with trusted documents."
In Google Docs: Claude for Google Workspace (public beta, Pro, Max, Team and Enterprise), a sidebar in Docs, Sheets and Slides. Keep its default, "Ask before edits". Anthropic's page says its edits "land as direct edits, not suggestions", so name a version first.
Where Claude stands out: finished documents from the chat on every plan, including Free. When the report has numbers, Claude for Excel answers with clickable cell references, so you can trace each figure to its cell.
Sources: Anthropic: Dashboards and motion (8 Oct 2026); Anthropic: What are artifacts and how do I use them; Anthropic: Word; Anthropic: Excel; Anthropic: Claude now works in Google Docs Sheets and Slides (6 Oct 2026); Anthropic: Use Claude in Google Docs Sheets and Slides

ChatGPT

In the chat: ChatGPT Work. Switch from Chat to Work. Work makes documents, spreadsheets, presentations, reports and Sites. Files already in Box, Dropbox, SharePoint or Google Drive can be added from Library without uploading them again.
In Word: ChatGPT for Word, on every plan including Free (since 17 Sep 2026). It is one of ChatGPT's add-ins for Excel, Word, PowerPoint and Google Sheets. It "cannot reference other files on your computer", so paste the source text into your request. OpenAI's own advice: "Save a copy of important work before making substantial changes." Turn on Track Changes in Word, and keep that copy to compare.
OpenAI's own warning: "ChatGPT can make mistakes and change content you intended to keep. Check important facts, figures, citations, and edits before using or sharing a document."
Where ChatGPT stands out: when the report must follow a company template. OpenAI calls GPT-6 Astra, which runs in Work, its best model for adhering to existing templates for slides and documents (that is OpenAI's own claim).
Sources: OpenAI: ChatGPT work and Codex; OpenAI: ChatGPT for Word; ChatGPT release notes, entries of 10 Sep and 17 Sep 2026 (OpenAI: ChatGPT release notes); OpenAI: Plans; OpenAI: GPT-6 Astra

Gemini

In Google Docs (work accounts): at a company on Google Workspace, Gemini is inside Docs, Sheets and Slides: draft, summarise, analyse a sheet, make slides. Look for "Ask Gemini" at the top right. What you get depends on your company's Workspace edition. Google's page says Gemini "will show suggestions directly in your document", which you accept or reject. Name a version first anyway.
In the Gemini app (personal accounts): Google's plans page says you can upload up to 1,500 pages of files; check what your own plan allows. With Google AI Pro or Ultra, a task (click Task below the text box) can create Word, PDF and Google Docs files.
Where Gemini stands out: built into Docs at Google companies, with nothing to install. For a report that must use only your own sources, Gemini Notebook is built for answers drawn from the files you add.
Sources: Google: Gmail Help; Google: Docs Editors Help; Google: Subscriptions; Google: Gemini Help; blog.google, NotebookLM is now Gemini Notebook (16 Jul 2026)

Claude, ChatGPT and Gemini · as of Oct 2026

Which one for a report?

The job	Claude	ChatGPT	Google Gemini	My pick
Writing the report	Clear business writing (Anthropic's claim for Opus 5.5); Claude Docs on every plan	Fast drafts in Chat; ChatGPT for Word even on Free	Writing inside Docs at companies on Google Workspace	Claude
A finished document	Word, PDF or Google Docs from one conversation; Claude for Word	Reports from Work	Gemini in Docs for Workspace users	Claude
The figures in it	Claude for Excel: cell references, formula-safe edits, overwrite warnings	Excel and Google Sheets add-ins; the Data plugin	Gemini in Sheets for Workspace users	Claude

My recommendation: Claude first, ChatGPT second, Gemini third.

For reports, Claude gives the most finished result for the least effort: a full document from one conversation on every plan, an add-in for Word and a sidebar for Google Docs, and cell references when the figures come from a spreadsheet. ChatGPT is a strong second, especially when the report must follow a strict company template. Gemini wins if your company writes in Google Docs every day, because it is already there.

Use the tool your company approves for company documents. That rule comes before mine.

On your own work · nothing private

Try it (10 minutes)

Use one section of a report you really have to write, and one source you are allowed to use: a public report from a source you trust, a page from your organisation's own website, or a file your workplace's rules let you put into this tool. Check your workplace's rules before pasting a confidential document into a chatbot. Never paste personal data, passwords or anything a client shared in confidence.

Save a copy (1 minute). Save a copy of the document, or name a version.
Make edits visible (1 minute). In Word, turn on Track Changes. In Google Docs, switch to Suggesting; with Claude for Google Workspace, keep "Ask before edits" on. In every tool, keep your copy to compare.
Brief and draft (3 minutes). Ask for that one section only: who it is for, how long, and "use only the attached source, and say which line of it each figure comes from".
Check every figure (4 minutes). For each number, name and date in the draft, find it in the source yourself. Sort them into three: found and right, found but wrong, not in the source.
Decide (1 minute). Accept or reject each change one at a time. Write down how many figures needed fixing.

That number is the most useful thing you will learn today: it tells you how much checking this kind of section needs next time.

Three questions · pick one answer each

Check yourself

1An AI add-in has drafted a section of your report. What do the vendors' own pages tell you to do next (as of Oct 2026)?
The answer: C OpenAI's page for ChatGPT for Word says to check important facts, figures, citations and edits, and Anthropic says to review every edit before accepting it. Asking it "are you sure?" isn't checking.
2Before an AI edits your report, which step works with every tool?
The answer: B Not every add-in writes its edits as tracked changes or suggestions (Anthropic says Claude's edits in Google Docs land as direct edits), so a saved copy or a named version is the one check that works everywhere.
3Your AI draft says "costs fell 8 percent" (an example figure). You attached the source. What is the check?
The answer: A With the real text in front of it, the likely answer is usually the true one. Usually, not always: still check, outside the chat.

0 of 3 answered

Read the whole lessonThe idea in full, the worked examples, common mistakes and what to check

The idea

Reports are one of the most common things people make with AI at work. In Anthropic's Economic Index report of 26 Jun 2026, documents and reports were the most common thing work conversations with Claude produced: 20 percent of them. That is one company's study of its own assistant, using chat and Cowork conversations on its consumer plans sampled between 10 Apr and 10 Jun 2026, sorted into types by Claude itself rather than read by people.

The assistant now sits inside the document, and it can also hand you the whole document from the chat. As of Oct 2026:

In Word: Claude and ChatGPT each have an add-in that works beside your document: Claude for Word, and ChatGPT for Word (on every ChatGPT plan, including Free).
In Google Docs: Gemini is built in for companies on Google Workspace. Claude has a sidebar in public beta, Claude for Google Workspace.
From the chat: Claude can make a finished Claude Doc in any conversation and export it to Word, PDF or Google Docs. ChatGPT Work makes documents and reports.

They no longer just hand you text to paste: they draft sections, rewrite paragraphs and restructure the file you are working in. So the question to ask is no longer only "is this answer right?" It is also "what did it change in my document?"

Treat it like a fast new writer on your team who has never seen your work before. You decide what the report is for, you hand over the sources, you read every change, and your name goes on it.

An illustration, not a real case: a team lead has to write the quarterly service report. She asks the assistant for the whole report in one go. It comes back fluent and well laid out, with a customer satisfaction figure in the summary that appears in none of her source files. Section by section, from the source, with each figure checked, that figure would have been caught before it reached her director.

The six steps

1. Brief it. Say who will read the report, what decision it supports, how long it should be, the tone, and what to leave out. OpenAI's own steps for ChatGPT Work say the same: add your files, then "describe the outcome you want", with any rules and how you will judge the result (as of Oct 2026).

2. Outline first. Ask for headings with one line under each, and fix the outline yourself before any drafting. OpenAI's help page for ChatGPT for Word says the same for larger changes: ask it to suggest an outline or explain its proposed edits before it changes the document (as of Oct 2026).

3. Draft section by section, from sources you attach. Point it at the file, or paste in the text. Paste in the contract, the policy or the report when you ask about it: with the real text in front of it, the likely answer is usually the true one. Usually, not always: still check. And check your workplace's rules before pasting a confidential document into a chatbot.

4. Make every edit visible. You want to see every change it made, accept the good ones one at a time and reject the rest. Before any AI edits the file, turn on Track Changes in Word, or switch Google Docs to Suggesting. Not every add-in writes its edits that way, so also save a copy or name a version: that is the one step that works with every tool, because you can always compare. The tool steps below say what each vendor's page describes.

5. Check every fact, number and name. Open the source yourself and find each figure. Quotes and links can be invented too. Asking it "are you sure?" isn't checking. Check outside the chat.

6. Keep version history. Save or name a version before the AI edits, so you can always see the report as it was and go back.

One caution for documents that come from outside your team: Anthropic warns that files such as downloaded templates, counterparty files or files shared by email "can contain hidden instructions" that manipulate its Word add-in (as of Oct 2026). Treat what an AI tool reads from outside as untrusted, in any tool.

A report you write every month: use a project

If you write the same report every week or month, set it up once. Make a project in Claude or ChatGPT (in Gemini, a notebook), add last month's report, the template and the sources you always use, and write a short instruction: who reads it, the length, the house style. Each month, open the project and brief only what is new. A project keeps its own chats, files and instructions; in ChatGPT, project instructions override your general ones.

Which model for a report. Start with the default. For a careful report, a messy spreadsheet or a long document, move up: Opus 5.5 in Claude, GPT-6 at High in ChatGPT, or Pro with thinking on High in Gemini. Check your own menu, because what you can pick depends on your plan.

Common mistakes

Asking for the whole report at once. You get fluent pages you cannot check. Outline first, then one section at a time.
No source attached. Without the real text, it writes a likely version of your figures.
Accept all, unread. Accepting every change at once is quick. Do it only after you have read every change.
No copy, no version. Then you cannot see what it changed, or go back.
"Are you sure?" as the check. Open the source instead.
Losing the hedges. "About", "in one survey" and "estimated" quietly disappear in a rewrite. Put them back.
Pasting what should not go in. A confidential file in a tool your workplace has not approved, or personal data in any tool.
Trusting an outside document. A template or a file from another organisation can carry hidden instructions. Use AI editing on documents you trust.
Starting from scratch every month. A project with last month's report, the template and the sources saves the setup.
Sending it as final. Anthropic does not recommend its add-in for final client deliverables without human review. In any tool, a person reviews it first.

What to check before you trust it

Every figure is traced to a line in the source, and you opened that source yourself.
Every name, date, quote and link is real and right.
Nothing you meant to keep was changed: compare with your saved copy or version.
The hedges survived the rewrite.
It answers the brief: the reader, the decision, the length.
If it says it checked something, ask what was checked.
You would put your name to it. "The AI did it" is never the answer to "who checked this?"

Check it. One section, one source, every figure found at its source before you accept the change.

Keep it safe. Check your workplace's rules before you paste or attach a document, and use AI editing only on documents you trust.

Checked against · read 16 Oct 2026Every claim, its source and the date read

As of: product steps as they were on the vendors' own pages on 16 Oct 2026.

Every tool line above was checked on the vendor's own page on 16 Oct 2026.

Neosaipien's platform reference (checked on the vendors' own pages, 16 Oct 2026): model names, plans, the document features of all three, the comparison and the recommendation
Anthropic, Build live dashboards and animate explainers with Claude (8 Oct 2026; Docs, Slides and Design on every plan)
Anthropic, What are artifacts and how do I use them (export to Word, PDF, Google Docs or PowerPoint)
Anthropic, Use Claude for Word
Anthropic, Use Claude for Excel (cell references; shared context with Word, PowerPoint and Outlook)
Anthropic, Claude now works with Google Docs, Sheets, and Slides (6 Oct 2026; public beta, Ask before edits)
Anthropic, Use Claude in Google Docs, Sheets, and Slides (edits land as direct edits, not suggestions)
Anthropic, Economic Index report: Cadences (26 Jun 2026; chat and Cowork conversations on consumer plans, sampled 10 Apr to 10 Jun 2026)
OpenAI, ChatGPT Work and Codex (what Work makes; describe the outcome you want)
OpenAI, ChatGPT for Word
OpenAI, ChatGPT release notes (10 Sep 2026, files from Library; 17 Sep 2026, ChatGPT for Word on all plans)
OpenAI, ChatGPT plans (add-ins for Excel, Word, PowerPoint and Google Sheets)
OpenAI, GPT-6 Astra (3 Sep 2026; best model for adhering to existing templates, OpenAI's claim)
OpenAI, Projects in ChatGPT (project instructions override general ones)
Anthropic, What are projects?
Google, Get started with Google Workspace with Gemini (Gemini in Docs, Sheets and Slides; Ask Gemini)
Google, Write and edit with Gemini in Docs (suggestions directly in your document)
Google, Google AI plans (up to 1,500 pages of file uploads)
Google, Manage your tasks and workflows in Gemini Apps (files a task can create)
Google, NotebookLM is now Gemini Notebook (16 Jul 2026)

The illustration in this lesson is made up. Neosaipien is not affiliated with Anthropic, OpenAI, Google or Microsoft. Product names belong to their owners.

Where next

Next lesson · level 6 · 20 min

What rules does your team need for AI?

Agree a one-page set of team AI rules, and know which 2026 laws it touches (not legal advice).

Open the next lesson

The kit · free download

The AI Fact-Check Kit, a five-step checking routine anyone can run on AI work: Sort, List, Trace, Hedge, Log, with review cards for a research report, numbers and a deck.

For the whole team: Workshop in a Box: Check What AI Writes, a 60-minute session a team lead can run with their own team.

Bring this to your team

The Team workshop. Hands-on training built around your team's real work. I can run it live.

Book a 20-minute call

Also on this path

Back to level 5, Office work, by task