Your Brand, Inside the AI Chat Window

Someone asks you for the logo.

You know how this goes. Find the right file. Not that one — the reversed version, on the dark background. Check whether it's the approved one. Wonder briefly whether they're allowed to use it for what they're using it for. Zip it. Email it. Get a reply asking whether there's a version without the tagline.

Twenty minutes, for a thing you have already done a hundred times.

Now imagine the other way this goes wrong. Someone skips you entirely and asks Claude to write the press release. It writes something confident and fluent, with a founding year that's off by two years and a description of your services you have never used. Fast, polished, wrong.

Both problems have the same root. Your brand lives in Brandkit. Your AI assistant doesn't know Brandkit exists.

The assistant is working blind

An AI assistant starts every conversation from nothing. It has no idea which of your photos are approved and which are still in review. It doesn't know that one image is licensed for internal use only, or that the photographer has to be credited by name. It has never read your tone of voice guide.

So it does what anyone does with no information: it guesses, confidently.

This is the same gap we wrote about in One Story, Not Ten — except now it's not just ten people telling ten versions of your story. It's ten people and their AI tools, all reconstructing your brand from memory, at speed.

What connecting actually means

There is now a standard way to plug a system like Brandkit into an AI assistant. It's called MCP — Model Context Protocol. You don't need to know anything else about the acronym.

What matters is the shape of it:

It's one connection, not one per tool. You connect Brandkit once. Claude can use it. ChatGPT can use it. Whatever your team adopts next can use it too, without anyone building a custom integration.

The assistant acts as you, not as everyone. It signs in with your account and inherits your permissions exactly. If you can't see a vault, neither can it. Nothing becomes more visible because an AI is asking.

It reads your real library, not a copy. No export, no sync, no snapshot going stale. It reads what's in Brandkit right now.

Once it's connected, you stop describing your brand to the assistant. It goes and looks (in your Brandkit).

What it can actually do

Brandkit gives the assistant specific abilities and tools. You never call these by name — you ask for something, and the assistant picks the right tool. But it's worth knowing what's in the box, because it tells you what's now worth asking for.

Finding things

search_assets — Finds assets by what they show, or by type, vault, tags, file format and orientation, etc.

The important part: you describe what you're after in your own words. It's matched by meaning. "In the wild" finds far more than "birds" would, because it's not hunting for a literal word in an Asset name, description or tags..

"Find landscape photos of the coastline we could use in a press release."

get_asset — Pulls the full record of a single asset once you've singled one out: dimensions, file size, licence, preview image.

"Tell me everything about asset 4821."

Checking before you commit

This is the group people don't expect, and the one that changes the most.

selection_info — Answers what a whole set of assets is licensed for, who has to be credited, and how big the files are — in one pass, before anything leaves the organisation.

This is the question that normally means opening eight tabs and reading eight licence blocks. It also surfaces the things that are easy to miss: an Asset that expired last month, one that isn't released until the campaign launches, a note your brand team wrote about how a particular image may be used.

"Can I send these five images to an outside agency, and who do I have to credit?"

answer_from_knowledge — Answers how-to and brand questions from your own Posts, Guidelines, and FAQ knowledge assets.

This is your Knowledge Library, finally reachable from the place people are already asking questions- their AI Assistant. Instead of a plausible invention, you get an answer based on your recorded knowledge.

"What is our rule about placing the logo over a photograph?"

whoami — Reports which user and account the assistant is working as.

Sounds trivial. It's the thing that explains a surprise: why a search came back empty, or why an action was refused. Usually the answer is your role, not a bug.

Getting things to people

create_transfer — Builds a link to a set of assets that anyone can open without logging in to Brandkit. This is the one for press, agencies and partners. You can set it to expire.

Nothing is emailed automatically. You get the link and decide what to do with it.

"Put these three logos in a link I can email the agency, expiring in a week."

create_share — Builds a link for people who do login to your account. Each recipient still only sees what their own role allows.

"Share the new campaign shots with the marketing team."

Keeping the work

create_collection and add_to_collection — Save a named set of assets in Brandkit to a collection.

The distinction worth internalising: a transfer is a link that goes stale, a collection is a place in your library that people can find again. If the selection outlives a single send, make it a collection.

"Save these as a collection called Spring Campaign 2026."

add_comment — Leaves a note on an asset under your name.

Small tool, disproportionate value. It's how the next person to open that photo finds out it went to the agency for the launch, instead of wondering.

"Note on asset 4821 that it went to the agency for the launch."

What this looks like on a normal Tuesday

A journalist emails asking for imagery for a piece running Friday.

The old version: Login to your Brandkit, search the library, filter by Licence, select the Asset or Assets, open to the Basket, Click Share, choose Transfer or Share, add the email, and expiry date, add a comment and click send.

The connected version is one conversation:

"A journalist needs coastline imagery for a piece running Friday. Find approved landscape shots, tell me what they're licensed for and who to credit, and put them in a link I can send that expires in two weeks."

The assistant searches, checks the licences and credits across the whole set, flags anything that isn't approved, builds the transfer, and hands you the link and the attribution line. Copy the link and text and paste into an email and send.

The work you kept is the judgement. The work you dropped is the fetching.

What it won't do

Worth being straight about, because trust here is the whole thing.

It doesn't invent assets. If nothing matches, it says so. It can't produce a logo that isn't in your library.

It won't quietly hand over unapproved work. Every asset carries its state. Drafts, Archived files, Expired ones and assets not yet Released are visible as exactly that — the assistant is meant to tell you rather than slip them into a send.

It can't exceed your permissions. It is you, with your role, with your vaults.

It asks before it acts. Searching and reading happen freely. Creating a transfer, a share, a collection or a comment writes to your account or sends assets out of it — your AI client will normally ask you to approve each
one. You can see what it's about to do before it does it.

Where to start

You don't need a rollout plan. You need one real task.

1. Connect Brandkit to your assistant once. It's a URL and a login — your Brandkit contact can help sort that out.
2. Take the next asset request that lands in your inbox and do it in the chat instead.
3. Ask a question you'd normally ask a colleague — "what's our tone of voice for social?" — and if the Guidelines in your Brandkit include tone of voice guidance, you'll get a sensible answer,

That third one is the real diagnostic. If the answer comes back thin, the gap isn't in the connection. It's in the Knowledge Library, and it's telling you exactly which Knowledge Asset to write next.

An Asset Library keeps your files straight. A Knowledge Library keeps your story straight. Connecting both to the tools your team already works in is what stops either one from being a place people mean to check, and don't.

Happy Branding :)