How to onboard a client brand into AI: an agency checklist

Leave onboarding with a usable task record

Receiving a brand folder does not tell an agent which logo is current, what the client means by “warm”, or whether a product claim is supported. Onboarding should resolve the questions that affect the first deliverable and check that the agent uses those answers.

The useful result is a record another person can continue from:

Client and brand reference:
Current sources and instructions for conflicts:
Reusable files and permitted use:
First task, audience and required copy:
Guidance and files supplied to the agent:
Output and checks performed:
Corrections made:
Unanswered questions and who can answer:
Any approval required by the client agreement:
How future brand changes will be reported:

The steps below fill in this record. Start with the task the client actually needs; you can extend the setup when the account adds other kinds of work.

Check how the material may be used

Read the client's existing agreement and account instructions before uploading its material. Record which tools will handle it, what access the team needs, and any restrictions that affect the work. Embargoed product details may need different treatment from a public logo.

Tool and plan choices matter. Anthropic's commercial-data policy, for example, says inputs and outputs from products such as Claude for Work and its API are not used for training by default. It also describes exceptions when users provide feedback or opt in. Check the terms and settings of the actual account your team will use, including sharing and memory settings; a product name alone does not establish how that account handles data.

Keep this part of intake proportionate to the client:

Those contacts may all be the same person. What matters is that the team knows who can answer the open questions.

Collect sources for the work you are about to make

Ask for the brand the client uses now. The oldest document may be the most complete, while the website may contain a newer palette and an outdated tagline. Ask which sources should take precedence instead of letting the agent infer that from file names.

Use this checklist to request what the upcoming work needs. Missing print colours need not stop an email draft; missing evidence for the email's central claim does affect that task.

Keep evidence with the claim. In the US, the FTC's advertising-substantiation policy says advertisers and agencies need a reasonable basis for objective claims before disseminating them. A model's fluent draft does not supply that basis. If the task needs an unsupported claim, leave it unresolved and ask the appropriate client contact rather than filling it in from a model's guess. Use any legal or compliance review the account already requires.

Turn conflicts into instructions

Read the sources together before creating the brand. Look for conflicting identities, adjectives without examples, missing task facts and files that will not work in the intended format. A screenshot of a logo may show which mark the client means, but the designer may still need its SVG.

Write the questions down once. When the client answers, preserve both the answer and its scope: does it change future brand work or only this campaign?

For illustration, Acme is a fictional company whose Atlas Pro product is a design app. These are proposed rules for that fictional client:

If the client has no answer yet, record the gap and the work it affects. Some gaps can wait; one that changes the first deliverable needs an answer or a narrower task. The Markdown brand-guidelines template can help turn resolved guidance into checkable instructions.

Our example for Eager, a juice brand shows why a task idea needs to stay distinct from a brand fact. We asked Claude to develop a hypothetical blackcurrant-flavour campaign using brand guidance and files supplied by Rainbrand. The saved task guidance identified the seven-carton source range and said the proposed flavour, size and packaging direction came from the task, not verified product specifications.

Ask the same practical question of your sources: which one establishes this fact? Keep a launch idea in the campaign request when no source establishes it as an existing product or lasting brand rule.

Create and inspect the client's brand

Once the sources are usable, make the maintained context accessible to the agents doing the account's work. Keep delivery requests and campaign decisions in your existing tools, with a reference to the client's brand. Pasting the guide into every assistant creates copies someone has to reconcile after each change.

Rainbrand maintains shared brand guidance and assets that connected agents and products can use. Add a brand from the client's sources and instructions. For files on your computer, use the documented upload workflow before asking the agent to use them.

Confirm the account first. The MCP quickstart explains that new brands go into the personal account by default; list the teams and name the intended one when shared access is appropriate. Choose an account or team whose access matches the client's arrangement. Adding a brand to an agency team is useful for shared work, but it should not expose a restricted client's files to people who may not use them.

This request uses the fictional Acme sources described above. Substitute your client's actual sources and resolved instructions:

Confirm the connected Rainbrand account and list my teams. Add Acme to the team specified in our intake record, using the attached current brand guide, resolved rules, logo and font files, and approved screenshots of the Atlas Pro design app. Follow the resolved rules where sources conflict. Keep the launch campaign's offer and dates in the task record. Wait until the brand is ready, then show its name, Brand ID, logos, colours, fonts and a summary of its guidance.

Leave retired files out of the working inputs. Keep their history elsewhere if the account needs it, rather than asking every future agent to ignore them.

Inspect the returned identity and guidance against the intake record. If a logo, colour or instruction is wrong, prepare a Brand Edit. Rainbrand proposes the change and may request more information. Inspect the proposal, then refine, apply or discard it. The active brand changes when the edit is applied.

Add the reusable images, footage and audio you want agents to find to the Brand Library, the brand's asset collection. Experiments and temporary outputs stay outside that collection unless you add them. Agents can then use the brand through MCP, while internal workflows can call the API.

Check the first upcoming task

Choose a task specific enough to reveal whether the setup works: the launch email due next, the first social graphic, or a section of the client's landing page. Include a relevant difficult detail, such as a renamed product or a claim with restricted wording. Expand testing to other formats as the account needs them.

First, request a Brief. It selects the brand guidance and assets relevant to that task. Read what the agent will receive before creating the output. Check that it includes the correct product, useful files and the instructions the task depends on.

Use Acme's ready brand from the intake record. Request a Brief for a 1600 × 900 Atlas Pro launch graphic aimed at product designers. Required headline: Meet Atlas Pro. Required call to action: Explore Atlas Pro. Use the supplied interface screenshot and the claim wording in the task record. Show the Brief's guidance and selected assets before making the graphic. If a required fact or file is missing, identify it before continuing.

Use the output format and requirements of your actual first deliverable.

Make the work with the tools you intend to use for delivery, then check it against the source record. Rainbrand Review can check work against the brand and return findings. For the graphic above, submit the final composition with the expected copy and relevant constraints. For a page or slide, capture the finished layout as an image for that request; review each layout you need checked rather than treating one capture as evidence for a whole deck.

Read the findings and evidence where provided. The agent's creation or editing tools make the corrections; Review does not edit the output automatically. Check product facts against their sources too, and follow whatever client approval the actual deliverable requires.

When the result is wrong, identify where to fix it:

Keep the Brief, output and correction together. After a repair, repeat the affected check so the record shows whether the fix reached the work. This procedure is checked against the documentation; we did not run a new client onboarding for this article.

Hand over what worked and what remains open

Complete the task record from the opening. Link the current sources, exact files, Brief and output. Record the brand version used where available, the checks performed, corrections made and any unresolved questions. If your agreement calls for client approval of the guidance or deliverable, keep that decision alongside the relevant record.

Agree where later changes will be reported so repeated feedback can reach the maintained brand. When an enduring rule changes, prepare and apply a brand edit, then request a new Brief for new work. Identify active jobs that need the change too; previously downloaded files and earlier Briefs do not update themselves.

Have the next person retrieve the intended brand and continue the task from this record. Any question they cannot answer tells you what the handoff still needs. The same record becomes the starting point for the client's next deliverable; managing many client brands covers keeping it useful across the agency's roster.