Is Claude Team enough for a marketing agency? An honest comparison

For some agencies, yes. Claude Team is a good place for an agency's people to work with AI. Shared projects give each client a home for instructions and files, and owners get sign-on controls, analytics and spend limits from a two-seat minimum. It starts to fall short when the agency needs each client's facts kept current without file swaps, a person signing off on every change, freelancers who can use their own AI tools, and a record that stays put when the team changes tools.

Storystak does not include an AI model, and it does not take the place of one. Claude, ChatGPT or whatever your team prefers stays the hub where the work happens.

What Storystak does is help an agency get more out of the AI tools it already pays for, and turn that into money for the agency. It does that with one shared record of every client that every person and every AI tool works from. The payoff shows up as:

• work that comes out the same no matter who made it
• client facts that are right
• a person approving every change
• more work done before anyone hits a usage limit

If you run a handful of clients, all in Claude, with one person who keeps every project tidy, you may not need Storystak at all.

Facts about Claude on this page were checked against Anthropic's help center on September 29, 2026. Claude changes often, so we re-check monthly.

Why does it matter where client facts live?

Almost every agency uses AI now, and very few have made money from it. An April 2026 Forrester survey run with the 4As, covering nearly 200 decision makers at vice president level or above at US marketing agencies, found 61% said AI remains a "cost of business" for their agencies, while just 6% defined AI as a "line of business" where AI capabilities are sold to clients.

Part of the reason is that the gains belong to individuals. Everyone found their own tool, wrote their own prompts and built up their own version of each client. People got faster. The agency did not get anything it could package, price or hand to someone new.

A Claude project per client is a real step toward fixing that. It puts one client's instructions and files where the whole team can reach them. The question is what happens to that setup as clients, people and tools multiply.

What does Claude Team give an agency today?

More than most agencies use. As of this month:

• Seats and limits. Standard seats are $25 per member per month billed monthly, or $20 billed annually, with a two-seat minimum and support for up to 150 seats. Premium seats for power users are $125 monthly or $100 annually. Prices are for US customers and exclude applicable taxes. Usage limits apply per member rather than to the team as a whole, and an organization can enable usage credits so members can keep working after reaching their included limits.
• Shared projects. A project can be private to invited members or public to everyone in your organization, at "can view" or "can edit" permission. Each project has its own instructions and its own knowledge base of uploaded files. On paid plans, when project knowledge approaches the context limit, Claude switches to retrieval instead of loading everything at once, which expands capacity by up to 10 times.
• Skills. Owners can provision a skill to everyone in the organization at once, and members can share a skill with specific colleagues. Giving a skill to only some users requires an Enterprise plan and is done by bundling it into a plugin assigned to a group.
• Connections to your other tools. The Team plan names Google Drive, Gmail, Google Calendar, GitHub, Microsoft 365 and Slack, and only Owners can add custom connectors for a Team organization. On Team and Enterprise plans, connectors are available in private projects.
• Analytics. Owners and Primary Owners can see active members and assigned seats, adoption level, product stickiness, skills including cost per use and number of uses, and connectors including the number of users and counts of read and write actions.

Anthropic is also rolling out a new version of projects, in beta to selected Pro and Max subscribers using Claude Code, with chat, Cowork, Team and Enterprise to follow. Existing projects keep working as they do today. Some of what is below may change when it arrives, and we will update this page when it does.

When is Claude Team enough on its own?

When the agency is small, everyone works in Claude, the client list is short, and one disciplined person keeps every project's files and instructions current. That setup works, and it costs a seat per person. If that describes you, spend the money on something else.

The rest of this page is about what changes once any of those stops being true.

What happens when a client fact changes?

A client moves a launch date, drops an offer or changes a claim legal will not approve. How that change reaches everyone depends on where the fact lives.

A project is excellent at holding recent work. It is harder to trust as the one place a client's approved facts live, because every update is a file operation someone has to remember to do.

With a stak, the correction happens where the mistake was noticed. Someone in the middle of a draft says the offer changed. That becomes a proposed update, and it reaches the whole team once a person approves it.

One honest limit: the stak does not detect on its own that a fact has aged. Canonical stats carry a clock and get flagged after 120 days without confirmation, but an ordinary fact nobody mentions sits there unchanged. What the stak changes is how easy it is to fix something once someone notices.

Who signs off before a change reaches everyone?

In a shared project, anyone with "can edit" can modify the instructions and the knowledge base everyone else relies on, and nothing approves the change before it lands.

In a stak, nothing changes without a person approving it, and every stat carries its source. When a client asks where a number in a deck came from, the answer is attached to the number.

That matters more than it used to. Clients are adding AI terms to contracts, and "how do you know what your AI is working from?" is a question agencies now get asked directly.

How do you bring in a freelancer without buying them a Claude seat?

Agencies lean on freelancers and contractors, and a good freelancer arrives with their own tools. They have built a way of working in their AI tools, and that way of working is part of what you are paying for.

On Claude Team, a project is shared with people from your organization. The roles are User, Admin, Owner and Primary Owner, with no guest role for projects, and an invitation has to use one of your organization's allowed email domains and occupies a seat as soon as it is sent. To give a freelancer a client's files and instructions, you put them in a paid seat, and they work inside your Claude setup instead of their own. When the job ends, someone has to remember to take the seat back.

With Storystak, a freelancer gets a seat on your stak and connects from the AI tool they already use. They read the client's approved record directly, they keep their own setup, and they start with what your best person on the account knows. Stak seats are for the agency's own people, staff, freelancers and contractors alike. The client's own team does not get stak access.

One limit worth stating plainly. Each client's context is walled off from every other client's, and that is enforced in code, so one client's facts cannot turn up in another client's work. But roles gate pages rather than clients, so anyone you give a sign-in can reach every client on the roster. If a freelancer should only ever see one account, that is not something the stak can enforce today.

The same record covers the harder version of this problem. When someone on staff resigns or goes on leave, the next person picks up from the record instead of from a handoff call. The client record belongs to the agency, not to whoever last worked the account.

What is your team asking that nobody has answered?

When Claude cannot find something in a project's files, it usually answers anyway, from general knowledge or whatever else is in the conversation. The answer reads fine, and nothing records that the files did not have it.

Claude Team's analytics tell an owner how many people are active and which skills and connectors get used. They do not report what the team asked about a client and could not get answered.

A stak does. When someone asks their AI a question the stak has no answer for, the miss is recorded, and repeated misses turn into a list of what is missing for each client. Some gaps the agency fills itself. Others go to the client through Dot, Storystak's interviewer, which asks the client directly and files each answer as an update for a person to approve.

The owner sees what people are looking for without reading anyone's chats. That covers stak activity only. Even if a project is public, chats within it stay private and inaccessible to other members of your organization unless someone manually shares them.

Does one client's information stay out of another client's work?

In Claude Team, separation depends on people working in the right project, uploading the right file to it, and not pasting one client's material into another's chat. Each project keeps its own knowledge base, which helps. But the walls are only as good as each person's habits on a busy day.

In Storystak, every read names one client, so one client's content cannot bleed into another client's output, however busy the day. That is separation between clients' content rather than a restriction on people: a sign-in reaches the whole roster.

What happens when you switch AI tools, or add one?

The best AI tool for a given job changes every few months, and teams follow it. Instructions and files built inside Claude projects stay in Claude. If the agency moves, or adds another tool for some of the work, all of that has to be rebuilt.

The stak lives outside any one AI tool. It connects over MCP, the open standard for connecting AI tools to outside sources, so any client that speaks MCP can read the same record. Claude and ChatGPT both work today, and Claude is the one we test end to end. Pick the best AI for the job and switch when something better comes along. Your clients come with you.

How do Claude Team and Storystak work together?

Claude, ChatGPT or whatever your team prefers is where your team works. Storystak is what they all work from.

On a Team plan, an Owner adds the Storystak connector to the organization once, and each person then connects to and enables it individually. From then on, any chat or private project can draw on the client's stak. Projects stay useful for what they do well, which is grouping the work and keeping recent chats together. The stak holds the approved record every project reads from.

Side by side

What does each cost?

Claude Team is priced per person, so its cost follows headcount, including any freelancer who needs a project. You keep paying for your AI tool with Storystak too, because Storystak does not replace it.

Storystak is $100 a month per active Client Stak, billed on the first, month to month with no minimum and no setup fee. The rate is the same for every client. Adding staff, freelancers or new hires to a stak does not change the price, and a client stak that is not active is not billed. A client added mid-month is prorated for the rest of that month. The agency's own stak, the Surface, Dot and training are included.

For the technical reader

Storystak serves each client stak over MCP, the open standard AI tools use to connect to outside sources. The AI reads from the stak and can propose changes, and every change is a human-approved update with its source recorded.

Claude is the tool we test end to end. ChatGPT connects too. In ChatGPT the connection currently needs a Plus or Pro plan, runs in a browser, and is set up through Developer mode.

On token use, we measured one real client build. Uploading the client's raw folder cost about 50,000 tokens a chat to deliver about 9,500 tokens of readable text, and some of that text was broken: a signed agreement that was scans, a branding guide that scrambled under multi-column extraction, an empty metrics spreadsheet. The stak delivered about 7,000 tokens per chat while drawing on 6 times more client knowledge, which is roughly 86% fewer tokens per chat. That comparison was against a raw folder, not a configured Claude project. A project that switches to retrieval instead of loading every file would narrow the gap.

See it on one of your clients. We will build a stak from what you already have and connect it to the AI your team uses. Book a call. For the wider comparison, including Notion and Google Drive, read Claude Projects, Notion, custom GPTs or Google Drive.