How to use Claude for work. The plan comes first, because it decides where the data goes; then one project for each area, with the company's context already inside.
Most guides on Claude for work start from Cowork and skip the choice that decides the contract: the plan. Here is the set-up sheet, with the company context file to fill in.
Using Claude in a business takes three decisions, made in order: the plan, one project per work area with its context file, and the data that stays out. Cowork and connectors come later.
The plan decides the contract. Anthropic puts Free, Pro, Max and Ultra under its consumer terms, where training on chats depends on a setting, and Team, Enterprise and the API under its commercial terms, where your content does not train the models. Team starts at two people, at 20 dollars per seat per month billed annually (18 euros in Italy), excluding tax.
The project is where Claude learns the business: instructions, price list, product sheets and glossary, uploaded once and present in every chat.
The template downloads below, in Word and PDF, with no email required.
Written by: Mattia Esposito, AI and marketing consultant, founder of Itria AI. From Puglia, based in Turin since 2023, he works remotely in Italy and abroad. LinkedIn profile.
The Claude for business set-up sheet, to download
A sheet to fill in before you hand Claude to your colleagues, in eight parts, from the plan to the data that never gets uploaded. Each part has an example from a made-up company, a twelve-person maker of aluminium profiles, and an empty column for yours. Part 4 is the company context file, the document Claude reads in every chat of the project.

| File | Contents | Link |
|---|---|---|
| Wordto fill in, in Italian | The eight parts with the worked example and empty rows for your company. Part 4 is copied into a file of its own and uploaded to the project. | |
| PDFto read or print | The same sheet, to keep beside you while you set up the projects. |
| Part | What to write | Where it comes from |
|---|---|---|
| 1. The plan | How many people, which plan, who the Owner is, and the training setting, checked. | Anthropic's plan page and Privacy Center |
| 2. One project per area | Name, aim in one line, who can view and who can edit. | Sharing on Team and Enterprise |
| 3. The instructions | Role, language and tone, what it always does, what it never does, how it hands work back. | Project instructions (Help Center) |
| 4. The context file | Who you are, what you sell, typical customers, trade vocabulary, frequent questions, things never to say. | The project knowledge base |
| 5. Knowledge files | Every file with its version and who keeps it up to date. | An old price list gives old answers |
| 6. Connectors | Which app, whether read-only, who has signed in. | Inherited permissions (Help Center) |
| 7. Cowork | Who uses it, for which tasks, how actions get approved. | Cowork on Team and Enterprise |
| 8. What never gets uploaded | Special categories of data, credentials, confidential documents. | GDPR and contracts with customers |
On Itria's AI transparency page I wrote down, supplier by supplier, what each one processes and under which contract. The Anthropic row covers Claude used through the API and reads: “Data Processing Agreement with Anthropic; data excluded from training.” Ask anyone who brings an AI system into your business the same thing: under which contract the data travels, and whether it ends up in training.
That row can be written because the API sits under Anthropic's commercial terms, which incorporate the data processing agreement. Free, Pro, Max and Ultra subscriptions sit under the consumer contract. For a business uploading price lists and customer emails, the difference matters more than any feature.
On 5 October, in the post about Itria's first seven months, my first piece of advice was “Start from a problem.” (the full post, in Italian). Claude works the same way: a project starts from a task the business repeats every week, and the context file holds what that task needs. A project opened “to try it out” stays empty.
One contract per supplier, in public. For Anthropic: a data processing agreement, with data excluded from training.
From one page to five in a day. On 25 September 2026, for Claude's web search, itria.io went from one page to five after Brave was asked to recrawl seven.
From 2 to 4 answers out of 8. The number of times itria.io turned up among the results Claude read, morning and evening of the same day. Citations in the answer stayed at zero.
Claude for business: plans and pricing
For a business there are two plans to look at: Pro for a single person, Team from two people upwards. On Anthropic's official plan page, read on 13 November 2026, Team costs 20 dollars per seat per month billed annually and 25 billed monthly, for groups of 2 to 150 people; viewed from Italy, 18 and 21.04 euros. Prices exclude tax.
| Plan | Price on the plan page (USD, excluding tax) | Training on your content |
|---|---|---|
| Free | 0 dollars. Usage limits every 5 hours and every week, up to 5 projects, no Cowork. | The user's choice |
| Pro | 17 dollars a month on the annual plan, 20 billed monthly (15 and 18 euros in Italy). Unlimited projects, Claude Code, Cowork. | The user's choice |
| Max | 100 dollars a month (90 euros in Italy), five times the usage of Pro. | The user's choice |
| Ultra | 200 dollars a month (180 euros in Italy), ten times Pro's weekly usage. | The user's choice |
| Team, Standard seat | 20 dollars per person per month billed annually, 25 monthly (18 and 21.04 euros in Italy). 2 to 150 people, shared projects, Cowork, single sign-on (SSO). | Off by default |
| Team, Premium seat | 100 dollars a month billed annually, 125 monthly (90 and 105.23 euros in Italy): five times the usage of a Standard seat. | Off by default |
| Enterprise, self-serve | 20 dollars per seat per month billed annually, shown in dollars in Italy too, 20-seat minimum, plus usage at API rates. Audit logs, Compliance API. | Off by default |
Viewed from Italy, the page shows prices in euros and warns that tax is not included: “Prices shown don’t include applicable tax”. Its footer adds that in the European Union the services are provided by Anthropic Ireland Limited. Prices change: check them on the official plans page on the day you choose.
The API, meaning Claude used by a program or an automation, is billed separately, by usage. For working in chat the plans above are enough; what a custom-built system costs is covered in the guide on how much an AI system costs a small business.
Claude for small businesses: Pro or Team
If only one person uses Claude, Pro is enough; from the second person, Team makes more sense, as the first plan with the commercial contract and shared projects. Anthropic's help page puts it plainly: “Team plans require a minimum of two members.” Below two people, a Team plan cannot be opened.
With Team, projects are shared with colleagues, with permission to view or to edit. An administrator, the Owner, decides which connectors are available and whether Cowork is on. And chats stay out of training by contract, with no settings to check person by person.
With Pro, projects stay personal. Under the consumer terms announced on 28 August 2025, with the choice made compulsory by 8 October, Anthropic trains new models on Free, Pro and Max chats if the setting is on, and keeps the data for five years rather than thirty days. On a personal account used for work, the first check is in the privacy settings.
How to use Claude projects for company context
A project is a workspace with its own chats, its own instructions and a knowledge base: Claude reads the files uploaded there in every conversation in the project. The Help Center describes it like this: “Projects allow you to create self-contained workspaces with their own chat histories and knowledge bases.” They are available on every plan, with up to five on the free one.
For a small business, one project per work area works well: sales, accounts, marketing. The instructions hold the role, the language, the tone and what to do when a piece of data is missing. The knowledge base holds the company context file, the public price list, the product sheets and the brand voice document, which has its own guide on brand tone of voice in ChatGPT and Claude.
On paid plans, when the files outgrow what Claude can read at once, the project switches by itself to a retrieval system, up to ten times the normal capacity. The practical limit becomes housekeeping: a 2024 price list left next to the 2026 one produces wrong prices. So the sheet asks for the version and the owner of each file.
Projects are changing. In November 2026 Anthropic flags a new version in testing, where a project becomes a single conversation that Claude splits into parallel threads; for now it reaches some Pro, Max and Ultra users. Screens may change; instructions and context files come with you.
Claude Cowork: how it works, and whether you pay for it
Cowork is Claude carrying out a multi-step task for you: you describe the result, Claude makes a plan, runs it and hands back documents, Excel spreadsheets with formulas, presentations or research. You pay for it: Anthropic's Italian help page says Cowork is available only on paid Claude plans (Pro, Max, Ultra, Team, Enterprise).
Sessions run on Anthropic's servers, in an isolated environment created for that session, and keep going with the computer closed. Working on files from your computer needs the desktop app open and connected, and Claude sees only the folders you connect; those files are processed on Anthropic's servers.
On Team and Enterprise, Cowork is on by default, and the Owner switches it off in the settings. Standing approval for actions that write into connected apps is off: each person approves them task by task. Local sessions save their history on the person's computer, outside Anthropic's retention rules, and the administrator cannot delete it.
The risk depends on what Claude can read and what it can do. Anthropic's Cowork safety page separates tools that read from tools that act: “Write tools inherently carry more risk as they can result in undesired actions.” An email or a web page can hide instructions written to manipulate the assistant, known as prompt injection. Hence, in the sheet, limited folders and actions approved by hand.
Connectors and MCP: Claude inside Drive, Slack and other apps
A connector lets Claude read data from an app such as Google Drive or Slack and take actions there, with the permissions of the person using it. The Help Center is clear: “Claude inherits each person’s permissions from the connected service.” If a colleague cannot see a folder, the connector cannot see it for them.
Beyond the official directory you can add connectors to any service that speaks MCP, the Model Context Protocol: the open standard for connecting AI applications to external systems, which names Claude and ChatGPT among the compatible applications. How to build an assistant that uses these links is in the guide on how to build an AI agent.
On Team and Enterprise the Owner enables a connector, then each person signs in with their own account. Avoid a custom connector with a shared credential: it sees everything that credential sees, no matter who is using Claude.
Where your data goes when you use Claude at work
On a commercial plan your content stays out of training; claude.ai data, however, is stored in the United States. Anthropic's Privacy Center says so for commercial products: traffic may be processed in the United States, Europe, Asia and Australia, and “Note that data is stored in the US.” Data residency in Europe is available through AWS Bedrock and Google Cloud Vertex, with regional endpoints.
| Where the business is based | Who signs the commercial contract | Where the data is kept |
|---|---|---|
| Italy | Anthropic Ireland, Limited; data processing agreement incorporated in the terms, arbitration in Dublin. | claude.ai: United States. In the EU: via Bedrock or Vertex |
| Other EU and EEA countries | Anthropic Ireland, Limited, on the same terms. | As for Italy |
| United Kingdom | Anthropic Ireland, Limited: the commercial terms group the United Kingdom with the European Economic Area and Switzerland. | claude.ai: United States |
| United States | Anthropic, PBC. | United States |
Anthropic's commercial terms, in force since 17 June 2025, put the ban in one line: “Anthropic may not train models on Customer Content from Services.” The data processing agreement is incorporated by reference, so there is nothing separate to sign. This is the contract that Article 28 of the GDPR requires when a supplier processes personal data on a company's behalf, with sufficient guarantees on technical and organisational measures.
Content from connectors, such as Google Drive files, stays out of training even on consumer plans, unless it is copied into the chat. The feedback button, on the other hand, keeps the whole conversation, and on commercial plans Owners can switch it off. What the supplier's US base means for your data is covered in the guides on ChatGPT, GDPR and the CLOUD Act and on AI without the CLOUD Act.
What can I do with Claude, and how to get the most out of it
The tasks that suit Claude are the ones with plenty of text and clear rules: reading long documents, preparing first drafts, sorting data into a spreadsheet, researching public sources. It is also the most common use in Italian businesses: according to Istat, the Italian statistics office, among companies with at least 10 employees that use AI, 70.8% use technologies that extract information from text documents.
| Task | Where it happens in Claude | What a person checks |
|---|---|---|
| Reading a long contract or tender specification | A chat in the project, with the document uploaded and the questions written out first. | Every quotation against the original text |
| Answering a request for a quote | The sales project, with price list and product sheets in the knowledge base. | Prices, lead times, discounts: always |
| The monthly summary in Excel | Cowork, with the folder of files connected. | Formulas and totals |
| A presentation for a customer | Cowork, starting from the project documents. | Figures and names, before sending |
| Research on competitors or regulations | A chat with web search switched on. | That the sources exist and say what is claimed |
| Copy for the website and social media | The marketing project, with the voice document. | Tone and promises |
The right-hand column comes from the way models go wrong: they write plausible falsehoods with the same confidence as true statements, the problem described in the glossary entry on AI hallucinations. A human check before work leaves the business has a name, human in the loop, and in the sheet it closes the last part.
On web search I have a measurement of my own. On 25 September 2026, asked the same questions, Claude went from finding one page of my site to five among its results, after I had asked Brave to recrawl seven: Anthropic lists Brave Search among the suppliers of its web search (the Brave and Claude lab note). The sources Claude finds for you depend partly on that index; to appear there, see the guide on how to show up in Claude.
Istat also measures the obstacle. AI use among Italian companies with at least 10 employees rose from 8.2% in 2024 to 16.4% in 2025, and among SMEs from 7.7% to 15.7%; of the firms that considered an investment and then dropped it, almost 60% cite a lack of skills.
Training the people who use these tools, even just a chat assistant, is covered in the guide on AI literacy under the EU AI Act; what happens when everyone uses their own account, in the one on shadow AI.
Claude vs ChatGPT for work
In how they organise work the two are alike: both have projects with instructions and files, team plans with a commercial contract, and connectors to business apps. For a small business, the useful differences lie in the plans, the limits and where the data is kept.
The task-by-task comparison, with official prices side by side, is in the guide to ChatGPT, Claude or Gemini for work. For ChatGPT there is how to use ChatGPT for work and the guide to ChatGPT projects and custom GPTs. The context file from the sheet works in both.
The other guides on automation and marketing for small businesses are gathered in the index of guides for small businesses; the custom systems Itria builds are on the services page.
Questions and answers
How much does Claude cost for a business?
On Anthropic's official plan page, read on 13 November 2026, the Team plan costs 20 dollars per person per month billed annually and 25 billed monthly, with a minimum of two people; a Premium seat costs 100 or 125. Viewed from Italy, the page shows euros: 18 and 21.04 per seat. Self-serve Enterprise is 20 dollars per seat per month, with a 20-seat minimum, plus usage at API rates.
Prices exclude tax, as the page says, and they change. Check them on the official page.
Is Claude Cowork free?
No. Anthropic says Cowork is only available on paid plans: Pro, Max, Ultra, Team and Enterprise. On desktop it runs on macOS, Windows and Linux (in beta); on web and mobile it comes with Pro, Max, Ultra and Team, and with Enterprise when an Owner switches it on.
Does Claude use my company's data to train its models?
It depends on the plan. On Free, Pro, Max and Ultra, under the consumer terms, chats are used for training if the setting is on. On Team, Enterprise and the API, under the commercial terms, Anthropic does not use inputs and outputs for training, unless you give explicit feedback or consent.
Can I share a Claude project with colleagues?
Only on the Team and Enterprise plans. A project can be shared with individuals or with the whole organisation, with permission to view or to edit, and Owners can switch sharing off. On individual plans a project stays personal.
How can a business make money with Claude?
By taking time off the tasks that come round every week: summaries, first drafts, long documents to read, spreadsheets to prepare. The gain is measured by timing the same task before and after, review time included. Anything going out to a customer still passes through a person.
Notes on sources
- Anthropic, Plans & pricing, read on 13 November 2026 with a plain request (prices in US dollars) and in a browser from Italy (prices in euros, Enterprise in dollars), excluding tax (“Prices shown don’t include applicable tax”); the table on training by plan; the EU supplier. Prices change.
- Anthropic, Claude Help Center: What is the Team plan? (updated 9 November 2026; source of the quotation on the two-member minimum) and What is the Pro plan? (30 October 2026: local currency and tax by country).
- Anthropic, Claude Help Center, What are projects?, updated 30 October 2026: knowledge base, instructions, five-project limit on the free plan, automatic retrieval on paid plans, sharing on Team and Enterprise, new version in testing. Source of the quotation.
- Anthropic, Claude Help Center: Get started with Claude Cowork (30 October 2026, Italian edition, paraphrased here), Use Claude Cowork on Team and Enterprise plans (12 November 2026) and Use Claude Cowork safely (30 October 2026, quotation on write tools). Read on 13 November 2026.
- Anthropic, Claude Help Center, Use connectors to extend Claude’s capabilities, updated 6 November 2026 (quotation on permissions); Model Context Protocol, What is the Model Context Protocol?, read on 13 November 2026.
- Anthropic, Updates to Consumer Terms and Privacy Policy, 28 August 2025: training with the setting on for Free, Pro and Max, retention of five years or thirty days, commercial services excluded.
- Anthropic, Privacy Center: commercial products (18 August 2026), consumer products (16 March 2026) and server locations (15 June 2026, quotation); Regional compliance, undated, read on 13 November 2026.
- Anthropic, Commercial Terms of Service, in force since 17 June 2025, version read on 13 November 2026: contracting entity for the EEA, Switzerland and the UK, data processing agreement incorporated, quotation on training.
- Regulation (EU) 2016/679, Article 28, Italian text read on 13 November 2026 from the EU Publications Office service.
- Istat, Imprese e ICT, anno 2025 (Businesses and ICT, 2025), 15 December 2025, companies with at least 10 employees in industry and services: AI use, technologies used, obstacles. This is the latest edition.
- The contract with Anthropic is described on Itria's public AI transparency page; the 25 September 2026 measurement is in the Brave and Claude lab note; the line about the problem comes from Mattia Esposito's public LinkedIn post of 5 October 2026, linked in the text and translated here. The company in the sheet is made up.
A Claude project takes an afternoon to set up. Connecting it to the company's software, with a person approving before anything goes out, is the work that follows. The system first, then the growth.
Mattia Esposito, who wrote this guide, builds tailored AI systems for small businesses with Itria: automations connected to the software the business already uses, and a Dashboard showing what has happened and what is waiting for a decision. Nothing goes to a customer without a person’s yes. On a test bench on 4 September 2026, with 30 documents including orders, delivery notes and invoices, the system extracted 234 fields out of 237 correctly, and the unreadable document deliberately placed in the batch was set aside without a single invented field.