AI Marketing Tool With an Approval Workflow, Step by Step

Key points

• Draft, edit, approve, preview and confirm are separate steps, so no single click takes a post from draft to published.
• The preview shows the exact text, the account and the character count. If the text changes before you confirm, nothing is posted and you see a new preview.
• Only the workspace owner can approve, publish, confirm a code change or approve an Instagram reel. Other members can read and edit drafts.
• Nothing posts twice. If a result is unclear, Kanso flags it instead of retrying.
• Kanso never merges code, never posts to Hacker News, and you can disconnect any account at any time.

Kanso is an AI marketing tool with an approval workflow: nothing publishes without the owner's yes

Kanso's specialists and your own Claude prepare drafts, replies, fixes and reel plans, and each one waits for the owner. The owner can publish through a connected X, LinkedIn or Reddit account, or copy the text and post it by hand.

That is the difference between a tool that prepares and a tool that posts. For the wider case for keeping a person in the loop, the human-in-the-loop marketing guide and the guide to the limits of AI in marketing go further.

To see it on your own business, start free and review your first draft with your website address. No credit card is asked for. For the product itself, what Kanso is and how Kanso works cover the rest.

How the approval workflow runs: draft, edit, approve, preview, confirm

A post passes through five separate steps, and no single click takes it from draft to published. They all happen on the Content page, which lists everything your Claude and the specialists saved, so you can review AI drafts before publishing.

1. Draft. A specialist or your own Claude saves the work. Nothing is public yet.
2. Edit. Every edit is saved as a new version, and you can compare versions word by word.
3. Approve. The workspace owner approves the draft. Nobody else can.
4. Preview. Publishing opens a preview of the exact post and the account it will go from.
5. Confirm. A separate click publishes it. Cancel at the preview and nothing is posted.

Each step is a decision a person makes. Approving a draft is not the same as publishing it.

What does the preview show before anything goes out?

The preview shows the exact post, the account it will go from and the character count, and nothing is posted while you look at it. For X and LinkedIn it is drawn the way the post will look, with its picture if the draft has one.

The preview says in plain words that this exact text is what goes out. If the text changes before you confirm, nothing is posted and you see a new preview.

Posting a Reddit reply from the Reddit Agent works from the reply you are looking at. Post to Reddit publishes the reply exactly as it appears at that moment, so read it first. Copy and open Reddit is the manual alternative.

Which actions only the workspace owner can take?

Only the workspace owner can approve a draft, publish it, confirm a code change or approve an Instagram reel, while other members can read and edit drafts. The Content page tells other members this directly.

Owner approval for AI posts covers every channel agent. The X Agent, the LinkedIn Agent and the Instagram Agent prepare work, and a reel goes out only after the owner approves it.

Why nothing posts twice

Nothing posts twice, because a post moves from draft to posting to posted and an unclear result is flagged instead of retried. Each approved Instagram reel is posted once.

Say the connection drops halfway. Kanso tells you it lost the answer and cannot tell whether the post went out, and it does not try again.

Press Check status and Kanso looks at the post again and shows whether it is live. Until you have checked, do not publish it a second time.

What your Claude can and cannot do through the connector

Your own Claude can read, draft and suggest through the Kanso connector, and it cannot publish unless the owner has turned that on. Reading and drafting are on by default, while publishing is off by default and limited to the workspace owner.

Your Claude can

• Read your business documents, today's ranked list and the channel's strategy.
• Read your Search Console and Analytics data and the saved page-speed audit.
• Save drafts and articles, and make one picture for a LinkedIn or X post.
• Read what has been saved, and update a task's status.
• Suggest corrections to your business documents.

Your Claude cannot

• Change your business documents on its own. It can only propose, and you accept or reject.
• Publish by default. If you turn publishing on, it still works only after you have seen the exact text and said yes.
• Post an Instagram draft.

How evidence is labelled: measured, estimated, not available

Kanso marks what it states as measured, estimated or not available, so you can tell a figure from a guess. A number a platform did not return is never shown as zero.

• Recommendations show their page, their evidence and the limits of that evidence.
• The GEO Agent's view of how citable your pages look is labelled as an estimate, and suggested wording is labelled as an AI-generated suggestion, never fact.
• Competitor findings quote the lines on the competitor's own page. Nothing is invented.
• Verified means Kanso saw the change on the live page. It does not mean your rankings improved.

Drafts are told never to invent statistics, customers or personal stories, and Kanso refuses a draft that breaks the “never say” rules you set. Reading it is still your job. In finance, law, health or insurance, a qualified person should review every draft.

How code changes stay small and reviewable

A code change stays small because the Coding Agent can only prepare a bounded change to website text files, show you the exact lines, and open a pull request when the owner confirms. It never merges.

1. Connect. The owner installs the Kanso GitHub App through GitHub's own sign-in and picks the one repository the project uses.
2. Prepare. Your own Claude reads the repository's file list, finds where the page lives and prepares the change.
3. Check. Kanso checks that it is the right repository, that no protected file is touched, that the change still matches the file as it is now, and that it is small. Keys and environment files, automation, dependency files, hosting and build configuration, scripts and binaries are refused.
4. Confirm. Kanso shows the exact lines. Only the owner can confirm, and then Kanso opens a pull request on a new branch, marked as a proposal for human review, and stops.

Your developer reviews and merges on GitHub under your own rules. Kanso never merges, approves, deploys, builds or tests your code.

How to disconnect an account at any time

The owner can disconnect an account at any time from Settings, and Kanso asks for a confirmation before anything is disconnected. Connections use each platform's own sign-in, Kanso does not see your passwords, and credentials are stored encrypted.

Without a connection you lose nothing you need to stay in control: drafts still wait for you, and you can copy the text or open the platform with it filled in. The privacy page covers how Kanso handles your data, and the documentation explains how connections and approvals fit together.

Things Kanso never does

Kanso never publishes without approval, and the list below is the part of that promise that is easiest to check.

• It never posts, comments, replies or votes on Hacker News, and it never logs in there.
• It never posts the same thing twice, and it never retries an unclear result.
• It never changes your business documents on its own.
• It never merges, approves, deploys, builds or tests your code.
• It never reads likes, views, impressions or followers on X or LinkedIn. It reports only whether a post went out.
• It never runs paid ads, manages local listings, checks compliance or gives legal, financial or medical advice.

AI marketing tool with an approval workflow: frequently asked questions