Human on send: the AI approval rule most teams skip

Human on send means a person approves anything that leaves your outbox, goes live on your site, or spends your money. AI drafts and assembles. A person checks and sends. That is the rule, and it is the one most teams skip first when they want to move faster.

This is not a complicated framework. It is one rule with three applications: send, publish, and spend. If you get this right, the rest of your AI marketing system has a safety net. If you skip it, speed becomes the thing that causes the damage.

Why teams skip it

Speed theatre. The whole point of AI is to go faster. Adding a human review step feels like it slows things down. It does slow things down, by about the time it takes to read what the AI wrote. That time is where the mistakes get caught.

Connector write scopes. Most marketing tools offer API connections that can send emails, publish posts, or adjust ad spend. When you connect AI to those APIs with write access, the default is often full access. The tool can send without asking. Changing that default takes deliberate configuration, and most teams never bother.

“Set and forget” culture. Automation marketing has spent years selling the idea that you set up a system once and let it run. That works for rule-based automation (if cart abandoned, send reminder). It does not work when the content is generated fresh each time by an AI that has no concept of your brand’s boundaries.

What always needs a human

Send

Any email, DM, text message, or outbound communication that reaches a customer, lead, or partner. No exceptions. We documented this in detail in how to use AI for email marketing. The AI drafts the email. You read it. You send it.

This includes automated sequences. If the AI generates personalised emails for a sequence, each template needs human approval before the sequence goes live. Reviewing the first email and assuming the rest are fine is how brand-damaging messages get sent at scale.

Publish

Any content that goes live on your website, social media, blog, or any public-facing platform. We build publishing workflows with human review at Digital Bond, and every post (including this one) goes through a person before it is live.

In Codex vs Claude Code, we called the code version of this rule “human on merge.” Same principle, different surface. Whether the AI wrote a blog post or a pull request, a person reviews and approves before it ships.

Spend

Any change to ad budgets, bid strategies, audience targeting, or campaign activation. We covered the specific workflow for Meta ads in how to use Claude to manage Meta ads. The AI can draft ad copy, suggest targeting changes, and recommend budget adjustments. It does not make those changes. You do.

This also covers indirect spend: signing up for new tools, upgrading plans, or committing to contracts. If an AI agent can make purchasing decisions on your behalf, you have a spend control problem.

What AI can own

Not everything needs human review. Here is what the AI can do without a person checking each output:

Drafts. First drafts of any content. The AI produces them, they sit in a queue, and a person reviews before the next step.

Research packs. Competitor analysis, market research, customer review summaries. This is input for decisions, not output that reaches anyone.

Sequencing and scheduling. The AI can organise tasks, suggest send times, and build content calendars. None of that is consequential until something gets sent or published.

Data processing. Cleaning data, formatting reports, summarising analytics. This is internal work that informs decisions, not work that affects customers directly.

The line is clear: if it stays internal, the AI can own it. If it goes external, a person approves it.

How to install the human on send rule

This is a checklist, not a tool recommendation. The tools change. The checklist does not.

1. Audit your current write scopes. List every API connection, Zapier zap, Make scenario, or automation that can send, publish, or spend. For each one, confirm whether a human reviews the output before it executes.

1. Remove auto-send from AI-generated content. If any workflow sends AI-drafted content without human review, add an approval step. Most platforms support this through draft or approval statuses.

1. Name an owner for each AI loop. Every AI tool that produces external-facing output needs one person who is responsible for reviewing it. Not a team. A person with a name. If they are on leave, the loop pauses or someone specific covers it.

1. Set up a rejection path. The reviewer needs an easy way to reject, edit, or delay AI output. If the only option is approve or start over, the reviewer will approve things they should not.

1. Review the rule monthly. Check whether any new tools, integrations, or workflows have been added that bypass human review. Teams add automations constantly. The review step gets forgotten unless someone checks.

FAQ

Is human on send the same as human in the loop?

Similar, but more specific. “Human in the loop” can mean a person is vaguely involved somewhere in the process. Human on send means a person approves the specific action of sending, publishing, or spending. The approval happens at the point of no return, not somewhere earlier in the workflow.

Does this slow down my marketing?

It adds the time it takes to review AI output before it goes out. For most teams, that is minutes per item. The alternative is letting AI send unreviewed content, which is faster until something goes wrong. The recovery from one bad send takes longer than months of review steps.

What about AI teammates that run overnight?

The AI can work overnight. The send happens when a person reviews it in the morning. We covered how to set up AI teammates with proper approval in how to pick an AI teammate for marketing. The bot prepares everything. The human approves it when they are ready.

Does this apply to code too?

Yes. In coding, the equivalent is human on merge. The AI writes code, and a person reviews and merges it. We covered this in Codex vs Claude Code. No auto-merge from any coding agent, ever.

What is next

If you want to see how your current AI workflows handle approval, take the scorecard. If you want us to help build marketing systems with proper human review built in, visit AI Systems Lab or talk to us through our AI consultant page.

AI drafts. You approve. That is the rule.