Chat

Chat is the front door of The AI CMO – the first item under Create in the sidebar. Brief it the way you'd brief a marketer on your team: plain sentences about what you want. It writes, designs, analyzes, and publishes, always in your brand's voice, always showing you the work before it goes anywhere.

The first page

Chat doesn't open on an empty box. A new chat says, in one sentence, what is worth doing now for your brand, with at most two cards under it, taken from your own numbers, your plan and your work:

Money at stake first: revenue that dropped, customers who are slipping away, a campaign earning the most. These come from the same list Home leads with.
Then your own coached plan, when one is unfinished.
Then a paid campaign that paid back well in the last 14 days.
Then posts you've already made for a platform that isn't connected yet.

With nothing to go on yet, it suggests first jobs from what you told us when you set up your brand: a first post, and a read of your website with the three changes it needs.

Nothing on the page sends anything or uses credits. A card's button puts a written request in the message box, for you to read, change and send, or opens Coach me with the outcome already filled in. Under the cards, one line says what was made this week, and another shows a draft waiting for your yes. Like the rest of the app, the page is in English; its money is in your brand's currency. What chat writes for you when you send a card's request comes out in your brand's content language.

A card's button never wipes what you've typed: its request goes on a new line under your words.

After a day away (no message in chat on this brand for 24 hours), chat opens on this page instead of your last conversation, and Pick up where you left off reopens that conversation. A link to a plan or a task still opens what it names. Within a day, chat reopens your last conversation.

The Chrome extension's install line is in the globe menu inside the message box, on a computer in Chrome or another Chromium browser.

Briefing it

Just type. "Write the launch email for our spring collection." "Draft five ad variants for the summer sale." "Which customers are slipping away?"
Attach context with the paperclip: a document (PDF, Word, PowerPoint, Excel, CSV, text or Markdown) or an image, or pull from your image library or content library.
Paste a screenshot with Ctrl+V (Cmd+V on a Mac) straight into the message box – a snip or a copied picture attaches as if you had uploaded it. Text you copy, including cells from Excel or a paragraph from Word, still pastes as text. A very large screenshot is scaled down to the size the chat reads, and Send waits until the picture has uploaded.
Give it strategy context with the strategy selector in the footer, and chat will work inside the plan you've approved.

There is no mode to pick. Say what you want – "a 9:16 video", "a playful LinkedIn post", "an SEO article on…" – and chat picks the tool and the settings from your words. The studios hold the full controls when you want to work on one piece in depth.

Documents you attach

Attach a brief, a deck, a spreadsheet or a report and chat reads it before it answers – it quotes what the document actually says rather than what a document like it usually says. Files up to 10 MB.

The chip on the attachment tells you what was read: "14 pages", or "not readable" when there is nothing to read – a scanned PDF has no text layer, for instance. In that case paste the text or attach a text version; chat will say the same in its reply rather than guess.

Very long documents are read in part. Chat tells you when a document was cut and asks which section matters.

Every document you attach is saved to the Library under Files, for the brand you are working in, so teammates on that brand can open it too. Later conversations can come back to it: open the file from the Library and choose to discuss it in chat, or ask for it by name and chat finds it in the Library and reads it again.

Anything that goes out (publishing a post, sending, putting something live) waits for you: chat asks first, for example Publish this to LinkedIn? or Apply this change in Google Ads?, and does nothing until you press Publish or Apply (Confirm for other steps).

What it can produce

Nearly a hundred tools sit behind the conversation. The short version of what you can ask for:

Content – blog posts, SEO articles, product descriptions, PR releases, sales scripts, proposals.
Email and SMS – campaigns, subject line testing, full sequences.
Outreach emails – a cold email to a prospect, a follow-up after a call, an introduction, a partnership ask. These are one person writing to another, so they arrive as plain text – a subject, a few alternative subjects, the body, and a short note to send if there is no reply – with nothing invented: whatever only you know (a first name, a company, a recent launch) is left in [square brackets] for you to fill in. Copy puts the subject and body on your clipboard, ready for your own mail client. They are saved to the Library as outreach emails and are not sent through the platform, which only mails people in your customer base who have not opted out. Ask for several and you get one card each.
Landing pages – ask for one and it arrives as a real page in your brand's colours, built section by section from your own site's styling, which you can then talk to and change.
Ads – Google, Meta, and LinkedIn ad copy and campaign drafts, compliance checks, account audits.
Images and video – generation plus a full editing bench (remove background, upscale, resize, recolor), shorts and captions.
Social – posts and threads for every connected platform, and document carousels: ask for a carousel and you get the slides (a cover, one idea per slide, a closing ask) as a .pptx document plus the caption that carries it, with the slides shown in the chat and in the library. LinkedIn takes the document as a document post; Schedule on LinkedIn on the card hands both to the scheduler.
Answers from your data – "who are my champions?", "what's my repeat rate?", "which campaign drove revenue last month?" Chat reads your warehouse, web analytics, and customer intelligence directly. On plans with the data warehouse it also works out what your events and ad spend add up to – MRR and how it moved, churn, CAC, ROAS, new customers, funnels, whether people who cancelled used the product less beforehand and how many paying customers' use has fallen by half – week by week with the average and the spread (totals only, never a list of people), and says which definitions it used ("trials = signups") and what your data does not hold.

Previews you can trust

Outputs render the way they'll actually look: Google ads as a search result, Meta and LinkedIn ads in their feed layouts, social posts in authentic platform frames, outreach emails as the plain text they are, and designed emails, articles, landing pages, press releases and journeys on the canvas beside the conversation. What you approve is what ships.

The canvas

The bigger things chat makes open beside the conversation instead of filling it. The conversation keeps its column on the left; the piece slides in on the right, at full size, with its own actions. In the conversation it is a small card – what it is, its title, its first lines, a miniature where there is something to see – with Open in canvas.

What opens there:

Journeys. The journey itself – the real draft, step by step, the same steps you see in Journeys. Each email step shows the email it sends: a miniature, its subject and its pre-header. A step that is written each time the journey runs says Written when it runs, with the brief it is written from. Click a step for its details; click its email to read it at full size, at desktop or phone width. Open in Journeys takes you to the builder to change anything. See Journeys built in chat.
Designed emails. The email at full size with its subject, its pre-header, the other subjects the writer offered, and which fields fill in for each person – first name, company – shown as labels where the person's own details will go. Copy HTML, Export, Design and Open in Studio are there, and the screen and phone buttons switch between desktop and phone width.
Articles and blog posts, at reading width, with Copy HTML, Export, their picture slots and Edit in Writing Studio to keep working on them.
Landing pages, at desktop or phone width, with Open in Page Builder.
Press releases, as the release document with Copy, Export and Edit in Writing Studio, where the release opens with its kit beside it – the pitch e-mail, pull quotes, suggested visuals and keywords.

An article or press release you changed in Writing Studio shows on the canvas as it was saved there – "as saved in Writing Studio" beside its word count – so the conversation never shows an older version. To change it again, use Edit in Writing Studio. See Writing Studio.

Social posts, ads and pictures stay in the conversation – they are short, and seeing them there is the point. Outreach emails stay there too.

The canvas opens by itself when something finishes while you are watching. Coming back to a conversation later, or reloading the page, leaves it closed – open it from the card. Close it with ×; later pieces in the same conversation then wait on their cards instead of opening it again. When a conversation has made several things, in this chat at the top of the canvas lists them all. On a narrow window the canvas opens over the conversation instead of beside it.

Coach me

When what you want is an outcome rather than one piece of work, ask to be coached. Chat writes a plan toward the outcome and takes you through it one step at a time, in the same conversation.

Starting one. Press Coach me under the message box, or pick Coach me in the command palette (⌘K). When you ask chat for a plan or for the steps to something, it also offers a small card under its answer: Work through this step by step? Either way, one card asks three things: the outcome in your own words, by when (2 weeks, 1 month, 3 months, or ongoing), and, only if you want to choose them, the channels. Leave the channels alone and the plan chooses them from the outcome; Choose them myself suggests the ones that fit it, and you can add your own. The card says what planning costs before anything is charged. Press Write the plan.

While it is written. A plan takes a couple of minutes. A card in the conversation says what is happening; you can keep chatting or leave, and the bell tells you when the plan is ready. Planning is charged once the plan exists. On the admin portal's Usage page its line reads Coached plan, and it is not counted towards the daily and weekly limits under Agents › Office, so a plan of yours never uses up what your AI team works within. If it could not be written, nothing is charged; Try again opens the card again with what you answered, and Remove takes the unwritten plan away. Sometimes a plan comes with one thing to read before you start, such as how workable it looks: you can start anyway or Change the outcome.

When it asks first. Now and then the plan needs one thing only you can say before it is written, such as which customers you are after. The card then shows the question with a box under it. Type your answer, loosely is fine ("both", "mostly small companies", "you decide"), and press Answer and write the plan (or ⌘ Enter, Ctrl Enter on Windows). Your answer stays in the conversation under the question it answers. You can also just answer in the conversation: chat recognises the answer, says it is writing the plan, and starts it with your own words. Nothing is charged for the question; planning is charged once the plan is written, as the card says. Once you have answered, the plan does not stop to ask again: anything still open it decides for you. To change what the plan is for, its end date or its channels, use Change the outcome; an answer does not change those. Remove is there too.

Working through it. Start step 1 begins. A line above the conversation names the plan and the step you are on, a row under your opening message keeps the plan in the conversation, and Plan opens the whole plan on the canvas beside the conversation: finished steps ticked, the current one with what it involves, later ones with their dates. When a step's work is done, a card under the answer says so, and Mark step done moves the plan on while chat takes you through the next step. You can mark the current step done from the plan as well. Each step's work is priced as you do it, like anything else you ask chat for.

Finding it again. The conversation is the plan's own. It is in History, where it says where the plan stands, and under the Plans tab there; Home lists your plans in progress too, and a plan's dated steps on the Calendar open its conversation. Everyone on a brand shares its plan and its progress, each in a conversation of their own.

A plan that goes quiet. When nothing has happened on a plan for a week, the bell says so, once, and opens its conversation. If another week passes with nothing happening, its next steps leave the Calendar and Home and History say it is quiet. It is not ended: pick it up in the conversation (write in it, or mark a step done) and it is back on the Calendar at once.

Ending it. The ⋯ menu on the plan has End plan, which stops the coaching and keeps the plan and the conversation, and Delete plan, which removes the plan and keeps the conversation.

A goal you want the AI team to run by itself is given on Agents instead.

Hiring an AI teammate from the chat

Ask for a teammate – an agent for SEO, someone to run the ads every week – and the reply names the one that fits, with a card under it: what it does and needs, how freely Agents run on the brand, what setting up and its ideas cost, and Hire. Nothing is hired until you press it, and the card follows the new teammate until it is on the team or has a question for you. The card hires for the brand it was offered on, even if you switch brands afterwards. See Hiring from the chat.

When a platform isn't connected

When you ask chat for something that goes through a platform that isn't connected yet (publish a post on LinkedIn, read your Google Analytics traffic, read your Meta Ads results), chat stops that step before anything runs and says so in a small card: what didn't happen, that the step wasn't charged, and a Connect button. An audit of an ad account only finds out when it reaches the account; the same card follows, and the credits for that step are returned. The button opens the same sign-in as the Integrations page. If your account runs several brands, chat first asks which brand the account is for, with the brand you're working in already selected.

Once you've signed in, chat picks up where it stopped. A post goes straight to its publish question, so publishing takes two presses: Connect, then Publish. A step that only reads, such as your traffic, waits for one press on Carry on, and the card says what it costs. Nothing runs without a press.

If a saved login stopped working, the card says so and offers Reconnect. Marketing email, SMS, WhatsApp, the support inbox and the app store keys need a few details, so their card opens Integrations in a new tab at the right form; come back to the chat when it's done and the card notices. An app that isn't one of the Integrations page's own cards opens Integrations with its name already in the search. When several platforms are missing for one request, one card lists them, each with its own button.

Not now closes the card. For a step that was stopped it leaves one line saying what didn't happen, and if you ask again later chat offers again. When chat offers a connection on its own under one of its answers, Not now keeps that offer away for two weeks in this browser; if you ask to connect it, chat offers it again. Chat never connects anything by itself and never opens a window without your press.

Doing things with the output

Every result carries its actions: Copy, Export (Markdown, PDF, Word, plain text, CSV, or HTML), and Regenerate – and for articles and press releases, Edit in Writing Studio on the canvas to keep editing. There is no Save: what chat makes is saved to the brand's Library as it is made.

Sharing a conversation

Share chat (under ⋯ beside History) makes a link to a read-only copy of the conversation, taken at that moment – what you say afterwards is not in it. You can add a password, and the link expires. Whoever opens it sees the conversation as the chat shows it: your messages and what you attached, the replies, and what the chat made – pictures, e-mails, pages and articles, posts, ads, journeys, site audits and carousels – with who shared it, the brand and the dates at the top, and Print or save as PDF. They never see the reasoning, the steps behind a reply, credits, or anything that needs an account to use. Turning the link off removes the copy.

Pictures in articles

An article comes with a few picture slots, placed where a picture helps – in chat, in Writing Studio, and in the articles your AI teammates write. Each is made at the size the article shows it (the article column, sharp on high-resolution screens – 1520 × 848 for a wide picture), with no words painted into it. In Writing Studio you can choose another size first: a common one or your own. The full-size original stays in your Library; the article loads small web copies of it (WebP, in a few widths), so it opens fast wherever it's published. Articles made before this carry the full-size file – Make pictures web-sized in Writing Studio's toolbar (or on the picture itself) gives them web copies. Nothing is generated or billed for that.

Fixing one part of a picture

When a picture is almost right – one word misspelled, a line of text running over a photo, an object you'd rather not have – choose Edit an area under the picture (or in its full-size view). Paint over the part to change (three brush sizes, with Undo and Clear if you overshoot), say what should change there and choose Edit area; the button shows the price. Only the painted part is redrawn: everything else is the original picture, pixel for pixel, at its full size. The result appears in the chat like any other picture and lands in your Library, and Regenerate on it tries the same painted part again. A failed edit is refunded.

To change the whole picture – a new style, a new scene – ask in the chat as usual.

Chat also publishes directly: an article can go to any connected CMS (Webflow, WordPress, Ghost, and more) as new content – a draft where the platform supports it – and posts can go to your connected social accounts. When there's more than one destination, chat asks – "Which collection should this go to?" – and publishes where you say.

Your website's events

Ask the chat to set up tracking and it works with the Website SDK, never a third-party tag manager. It checks what your site already reports – whether the tag is installed and sending, which event keys are registered, which keys the site sends that nobody registered – then proposes one event per business moment in your brief, registers the keys (the same registry as Data → Events), and hands your developer one block per event: the track call, the no-code data-aicmo-event attribute, and the identify or trackConversion call that belongs beside it. Signups and purchases go in as conversions with a value. The rules are the ones on the Website SDK page: keys are lowercase letters, digits and underscores, they must be registered before the site sends them, and they land in your warehouse as sdk_<key> where journeys and reports can use them.

Your brand, always

Chat works from your active brand profile – name, audience, offer, differentiators, and brand voice ride along on every request, which is why nothing comes back generic. Switching brands in the icon rail switches everything.

If a brand sells in another language, set Content language on the brand itself – Brand → Profile → Market. Strategies, ads, emails, the text inside generated images, video scripts: all of it comes out in that language for that brand, while the app itself stays in English. Brands that don't set one follow the Default AI content language in Settings, so an account running brands in several countries gets the right language per brand rather than one setting for all of them.

Currency sits next to it, and works the same way: revenue and customer value are labelled with it, and written offers use it instead of dollars.

Getting around

⌘K (Ctrl+K) opens the command palette from anywhere: search tools, pages, projects, clients, and your recent content, or fire a quick action like New chat, Coach me, or New journey.

History, under the message box, lists your conversations on the brand by the day of their last message, newest first; opening one does not move it. Under each title is what the conversation came to: what it made and whether that is in your Library ("2 images and 1 email · in your Library"), where it stopped and why ("Site audit stopped: the site blocked the crawl"), or what it is waiting on you for ("Waiting on you: connect Google Search Console"). A green dot marks work done, a gray one a conversation waiting on you, a red one a step that stopped. Search looks through the titles, what you asked and those lines. Each conversation's ⋯ menu renames, pins, archives or deletes it, on a phone too, and a deleted conversation can be brought back with Undo for a few seconds; it is gone once that time is up, or when you close History or leave the page. Deleting the conversation you have open starts a new one, as New chat does, unless you open another conversation from History first: that one opens. The arrow keys move through the list and Enter opens a conversation. A conversation about one item in its side panel (a Library item, a journey, an ads audit) stays with that item and is not listed here. The newest 50 come first, and the list says how many more there are, with a button at its end to show them. When you have a coached plan's conversation, however old, a Plans tab lists them.

Credits

Every message is priced on what it uses. The first message after an hour away loads your brand's full context and comes to around 12 credits; every message after it within the hour reuses that context – a reply is usually 1 to 3 credits, a message that makes something with a tool a few more.

Everything the chat produces runs on credits too – text draws the least, richer media like video the most. Reading your own data, publishing, and browsing are production work, so they draw credits as well. Failed runs are refunded, and every charge shows in Usage.