Chatting with the Brain & uploading documents

The fastest way to use the Brain is to just talk to it. Open Brain in the sidebar and the chat box sits right at the top of the page, under your knowledge graph.

Asking questions

Type a question in plain English - "which channel had the best ROAS last month?", "what should we focus on?", "why did conversions drop?" - and the Brain answers using your connected data and everything it's learned. Because answers are grounded in your account, they're specific to your business, not generic.

Not sure where to start? Before your first message the chat offers four questions you can click straight away - What's working best right now?, Where am I wasting spend?, Why did performance change last month? and What should we do next? They're ordinary questions, not canned demos: each one is answered from your real numbers.

Asking about the platform itself

The chat also answers questions about the platform - how to connect a tool, what a card means, what a feature does and how to switch it on. Ask the way you would ask a colleague: "how do I connect Meta?", "can it read my Shopify orders?", "what does the halo card mean?", "how do I use this from Claude Desktop?". The answer comes from this Help Centre and knows your own workspace - what is connected, which plan you are on - so it tells you the step that applies to you rather than a generic one, and links the page here if you want the full read.

A few things it will not do, on purpose: it does not pitch features you did not ask about, it does not keep reminding you of something it has already mentioned in the same conversation, and it does not walk you through a tool that is still marked Coming soon. If it is not sure how a part of the platform works, it says so rather than guessing - the Help Centre search above, or I have a problem at the bottom of the dashboard, are the next stop.

Sharing updates and context

The chat is also how you tell the Brain things - an upcoming campaign, a change of plan, notes from a call. There's nothing to tick first: the Brain reads every message you send and decides for itself what's worth keeping, so a question stays a question and an update gets recorded.

Whatever it did is listed under the reply, point by point:

• Noted in your brain - recorded straight away.
• Flagged to confirm - the Brain understood the point but the wording was tentative, or it clashes with something it already believes, so it's holding it for a check. If you're sure, hit Save it anyway on that line and it goes in - no need to retype it.
• Nothing in that needed recording - it judged the message to hold nothing durable. If you disagree, Remember it anyway stores your message word for word.
• A review of earlier records - you told it something replaces what it knew. See the next section.

For a very long update - full meeting notes, a multi-page briefing - attach it as a file rather than pasting everything into one message, or send it in a few pieces. A pasted message is mined for its distinct points and the chat will tell you if it may not have caught them all; an attached file is read end to end.

Telling the Brain something is out of date

Say it plainly - "the new brand manual replaces the old branding", "disregard the previous pricing, this deck is current". The Brain does not take your word as an instruction to delete things. It checks its earlier records, and under its reply you get a short list of the ones that now look out of date, already ticked, for you to confirm.

• Nothing is retired until you confirm. The reply will say it is checking, never that it has already cleared anything.
• Records it was less sure about, and anything marked as a key fact, are listed unticked - tick them yourself if they should go too.
• A few records are kept up to date somewhere else - a competitor, a goal, a Brand Identity field. Those are listed with where to change them, because retiring them here would only last until the next sync.
• Retired records stop being used but are never deleted. Undo this review puts every one of them back.
• If nothing earlier looks out of date, it says so and changes nothing. If it could not check just then, it says that instead - say it again in a little while.

Anything that arrived with the replacement itself - the points from the new manual you just uploaded - is never on the list.

Uploading documents

You can attach a document to teach the Brain something it can't get from your platforms - a brand deck, a strategy doc, a past agency report, a tone-of-voice guide. The Brain extracts the useful parts and folds them into its knowledge, so every downstream surface (reports, Creator's Studio, recommendations) gets sharper.

What you can send:

• PDFs, Word documents (.docx), and plain text or markdown files
• Images - PNG, JPEG, GIF and WebP

Files can be up to 25MB each. The file is read as soon as you send it, and the chat shows what it is doing while it reads.

Adding what the Brain found

Reading a file does not add anything to your Brain by itself. When the Brain has finished, a card appears under the upload listing the points it found, each one ticked:

• Add them - press Add N to your brain. Untick anything that is wrong or not worth keeping first; an unticked point is turned down, and the rest are added.
• Not now - close the card. Nothing is lost and nothing is added.

Once you add them, the card says Adding N points to your brain while the Brain takes them in, and N points are now in your brain when it has. That usually takes a minute or so. If it takes longer the card tells you it is still finishing in the background - you don't need to wait on it.

Until you press Add, those points are not part of your Brain - they can't shape a report or a recommendation. The conversation you sent the file in can still see the file itself, so you can ask about it straight away. That is the reason for the extra tap: it's much easier to leave five wrong points out than to take them back after they've been used.

Turning down a point the Brain already added

You can change your mind about a point after it has been added, from the document's review list: choose Undo (reject) on the point.

A point that has been in your Brain for a while may have been used: the Brain keeps its own record of it, and may have worked other things out from it. So before the point is turned down you are told what else will happen - for example "Turning this down also retires 1 record in the brain that copies it and asks you about 2 more that rest on it." - and nothing happens until you press Turn it down.

• Records that simply copy the point are retired with it, straight away.
• Records that rest on it but may rest on other things too are never retired for you. They are listed in one review with tick boxes, alongside your other Brain questions, and nothing in it is retired until you tick it. See Brain questions & feedback.

Nothing is deleted. If you restore the point later, the record that copied it comes back with it, and a settled review has Undo this review.

A point that was never added - one you unticked on the card, or turned down while it was still waiting for review - has not been used by anything, so turning it down has no knock-on effect and you won't be asked.

Sending an updated version of something

Send the exact same file twice and nothing happens - the Brain recognises it, keeps what it learned the first time, and doesn't read it again.

An updated version is different: the file has changed, so it's genuinely new to us, but it's still about the same meeting or the same plan. If the name looks like a new version of a document you've already sent - Kick-off call.pdf followed by Kick-off call v2.pdf, or the same name re-saved - you'll be asked which you meant:

• Replace it - the new version becomes the live one, and what the Brain learned from the older version is retired. Use this for a corrected transcript, a revised deck, a second draft. The new version's points still wait for you to add them (see above), so add them straight after - until you do, the Brain is holding neither version. If the Brain had worked other things out from the older version, you'll also get one review listing them, so you can decide which still stand.
• Keep both - they're separate documents that happen to have similar names. Both stay.

It's worth answering rather than dismissing. Keeping two versions of one meeting leaves the Brain with two accounts of it, and it will weigh both when it answers you.

Nothing is deleted either way. A replaced document stays in your list, marked as replaced, and you can still open it. Points you retire yourself - from the Explorer, or from a review of older records - show as Retired in that document's entry list with a Put back link, and the count beside the document says how many were retired rather than hiding them.

Screenshots, photos and scans

You don't need a file to be text to send it. The Brain reads images and scanned documents as well as ordinary files, so all of these work:

• A screenshot of a dashboard, an ad, or a competitor's landing page
• A photo of a whiteboard, a printout, or a page from a deck
• A scanned or design-heavy PDF with no selectable text, such as a ticket, an invoice, a brochure or a brand manual

Anything visual that isn't text - a chart, a logo, a photograph - is described alongside the text, so the Brain can answer questions about the whole thing rather than just the words.

Image-based documents are read a section at a time, up to 50 pages, and the chat shows how many pages it has read so far. A long one takes a minute or so. If yours is longer than 50 pages, send the section that matters and the Brain will read that.

If something you upload makes a Brand Identity field look out of date, the Brain may propose a refresh on that field for you to review - it never changes it on its own. See Brand Identity.

Tips for good answers

• The more context the Brain has, the better it gets - connect your platforms and add the odd document or note.
• Ask follow-ups. The chat keeps the thread, so you can drill in.
• If the Brain isn't sure, it'll tell you, or ask a question back - see Brain questions & feedback.