Claude projects are self-contained workspaces in Claude that hold their own chat history, knowledge base, instructions and memory, so every conversation inside a project starts from the same context. For marketing teams, the strongest setup in 2026 is one project per brand, client or campaign, with brand and product facts in project knowledge and working rules in project instructions. On Team and Enterprise plans, those projects can be shared with view or edit rights, which turns one well-built project into a shared source of truth for the whole team.

Claude projects changed quickly in the second half of 2026. Anthropic's help article on retrieval augmented generation for projects, updated September 10, 2026, explains how paid plans expand the knowledge capacity of each project. On September 16, Anthropic announced that Claude Cowork and chat are now one Claude, and existing projects carry over into the merged experience. A day later, Anthropic published Projects redesigned: from folder to conversation, a beta in which a coordinator directs parallel threads inside each project. RZLT's roundup of Claude use cases for business teams covers what Claude does well, and this guide covers the project structure underneath that work.

What Claude Projects Are in 2026

Anthropic's overview of projects states that "Projects allow you to create self-contained workspaces with their own chat histories and knowledge bases." Four properties matter for a marketing team.

1. Knowledge loads into every chat. Files and text added to a project become context for each conversation in it. Anthropic's guide to creating and managing projects also notes that "Context is not shared across chats within a project unless the information is added into the project knowledge base," so decisions worth keeping belong in knowledge.

2. Instructions apply project-wide. According to Anthropic, "Claude will use these instructions for all the chats within the project." The project name and description work differently, since "Claude will not have access to these details," so the brief belongs in the instructions field.

3. Capacity scales with RAG on paid plans. Anthropic states that "RAG automatically activates when your project approaches or exceeds the context window limits," expanding capacity "by up to 10x while maintaining quality responses." It is available on Pro, Max, Team and Enterprise, and a visual indicator shows when a project is RAG-enabled.

4. Each project keeps its own memory. Anthropic's guide states that "Each project has its own memory," and turning off "Use account memory" on the project page keeps that memory separate from the rest of the account. For agencies, that separation keeps one client's details out of another client's work.

Projects are available on every plan, and Anthropic notes that "Free users can create a maximum of five projects."

How to Structure Claude Projects by Brand, Client and Campaign

RZLT recommends three layers, each with a clear job.

1. One brand project per brand. This is the long-lived reference for positioning, voice, audience, product facts and approved proof points, and it changes only when the brand changes.

2. One client project per client engagement. Agencies need a hard boundary per client. A client project holds that client's brand material plus the engagement scope, approved claims and stakeholder preferences, with account memory switched off so nothing leaks between accounts.

3. Short-lived campaign projects. A product launch, an event or a quarterly campaign gets its own project with the brief, channel plan, deadlines and draft assets. When it ends, anything worth keeping moves into the brand project.

Name projects with a consistent pattern such as "Client · Campaign · Quarter" so a shared list stays scannable, and give each project one owner who decides what goes into knowledge, which keeps conflicting versions of the same document out.

What Goes Into Project Knowledge and Project Instructions

Anthropic's help center says teams can upload "relevant documents, text, code, or other files to a project's knowledge base." Its own project examples suggest "Style guides, audience personas, SEO best practices" for a content project. The split that works best keeps facts in knowledge and behavior in instructions.

Project knowledge (the facts):

• Brand foundations. The voice and style guide, messaging pillars, positioning statement and a glossary of product names, so every draft uses the same terms.

• Audience and market. Personas, interview or call summaries and competitor notes, which give research and copy real buyer language.

• Approved proof. Case study numbers, customer quotes cleared for use and the claims legal has signed off, so Claude draws on verified material instead of inventing it.

• Reference examples. Two or three of the team's best past assets per format, such as a top-performing email or landing page, which set the quality bar.

Project instructions (the behavior):

• Role and audience. Who Claude is working for and who reads the output.

• Voice rules. The hard rules from the style guide, such as spelling conventions, banned phrases and formatting preferences.

• Output formats. The default structure for recurring deliverables like briefs and social posts.

• Escalation rules. When to ask a clarifying question and which claims always need a source from project knowledge.

Long reference material belongs in knowledge, where RAG can retrieve it.

Sharing Claude Projects on Team and Enterprise Plans

Anthropic's article on project visibility and sharing states that these features "are available to users on Team or Enterprise plans." Four details shape team workflows.

• Two visibility settings. A public project means "Everyone in your organization can view and use the project," while a private project is limited to invited members. Brand projects suit public visibility, and client projects usually stay private to the account team.

• Two permission levels. With "Can view," members "can see project contents, knowledge, and instructions, and chat within the project, but cannot edit it." With "Can edit," they can change instructions and knowledge and update member settings. RZLT recommends giving edit rights to project owners and view rights to everyone else.

• Private chats inside shared projects. Anthropic states that "Chats within a project are not shared by default," so the project and its knowledge are shared while each person's conversations stay their own.

• Admin control. Projects can be shared with named members, a pasted list of emails or the whole organization, and Anthropic's projects overview adds that organization owners can turn off sharing, which also turns off public projects.

On price, Anthropic's pricing page lists "Project sharing and collaboration" on the Team plan, where Standard seats cost $25 per seat a month billed monthly or $20 billed annually.

Claude Projects vs Cowork vs Skills

Projects, Cowork and Skills each do a different job.

• Projects hold context. Anthropic's help article on skills puts it plainly: "Projects provide static background knowledge that's always loaded when you start chats within them." A project answers the question "what does Claude need to know about this brand or client?"

• Skills hold procedures. Anthropic describes skills as "folders of instructions, scripts, and resources that Claude loads dynamically," and notes that they "only load when relevant." A skill answers "how should Claude do this task?" and works across every project. Organization owners can provision skills so they appear in every member's list. RZLT's list of Claude Skills for developers shows how SKILL.md files package repeatable work.

• Cowork carries out the work. Anthropic's guide to projects in Claude Cowork describes projects that "group related tasks into dedicated workspaces with their own files, context, instructions, and memory." A Cowork project can reference a local folder, a linked chat project or a URL, and can run recurring tasks specific to that project. Creating one from an existing Claude project transfers its files and instructions.

Since the September 2026 merger, the lines have moved closer together. Anthropic's article on the merged Claude experience states that "Projects work in every conversation," and Cowork tasks, projects, connectors, skills and files carry over as accounts move. The rollout started with Pro and Max, and Anthropic's announcement says Team and Free plans will follow soon. RZLT's CEO Luka Ciganek described the same layering at the first Claude Code meetup in Sofia: custom instructions first, then projects, then skills and connectors, then Cowork for automation that actually runs.

The New Version of Claude Projects (Beta)

Anthropic is rebuilding projects for longer, agentic work. In the redesign, "Projects have threads that do the work and a coordinator that directs them," every thread adds to and draws from a shared memory, and a library collects the files added and the artifacts Claude produces. Threads run in the cloud, so the work continues after the laptop is closed.

Anthropic's projects overview says the new version is "available in beta to select Pro and Max subscribers who use Claude Code," with chat, Cowork, Team and Enterprise to follow. Some existing Pro and Max projects are being moved to it, and Anthropic states that "Nothing is lost or deleted in the move," since instructions, files, full chat history and scheduled tasks come along. Running several threads at once uses the plan's usage faster. A well-organized knowledge base carries straight into the new version.

How Marketing Teams Should Set Up Claude Projects

Claude projects give a marketing team one home for brand context. RZLT recommends this order for 2026:

• Build the brand project first. Load the style guide, personas, product facts and approved proof, then test it on three real tasks before sharing it.

• Split clients and campaigns early. Create one private project per client with account memory off, and a short-lived project per campaign that is archived when the campaign ends.

• Share with clear roles. On Team or Enterprise, give edit rights to one owner per project and view rights to everyone else.

• Move procedures into skills. When the same instructions get pasted into several projects, turn them into a skill so every project uses one version.

For teams still choosing a model, RZLT's comparison of Claude vs ChatGPT for B2B content, research and strategy sets out where each one leads. For teams ready to scale output, RZLT's breakdown of its content production stack for shipping 60 pieces in 6 weeks shows how skill files, Claude and n8n turn that context into a full production pipeline.