What an AI marketing employee should never do

Every AI marketing tool sells you on what it can do. Draft your ads, write your posts, watch your spend. Fine. But if the thing is going to act like an employee, the hiring question is different: what will it refuse to do, even when a prompt or a bug asks it to?

We manage real ad spend for our clients through Spuro every day, which concentrates the mind. This is the list of nevers we built into it, and the list we think you should demand from anything with its hands on your marketing.

Never publish without a human yes

Drafts are cheap. Publishing is not. An ad that goes live in your name, a reply sent to your customer: each is a promise made in public. In Spuro, AI drafts and a person approves. Nothing crosses from draft to live without a human tap, and we treat that boundary as load-bearing, no matter how good the drafts get.

Never move money on its own judgment

Automation should act inside rules you wrote, at thresholds you set, and log what it did. It should never improvise with budget. Our rules engine pauses, scales, and alerts against your targets, the numbers you typed, and the log of every action reads like a shift report. If a tool cannot show you the rule that caused an action, you have handed a stranger your card.

Never invent a number

AI systems are fluent, and fluent things are tempting to trust with arithmetic. But a marketing number that cannot be traced to a real report is worse than no number. Spuro's figures come from the platforms' own reporting, carried through untouched. When a number is a model's guess, labeling it that way is part of the design. The rule inside our team is simple: no stat without a source, even in a draft.

Never hold your data hostage

Your ad accounts stay yours. Your content is written to live on your own site and build your domain's authority, and Spuro hands it to you ready for your CMS. Your customer data stays in your platforms. If you leave Spuro, you leave whole. A tool that gets stickier by making departure expensive has stopped earning its fee and started charging rent on your exit.

Never perform confidence it has not earned

The most human failure an AI employee can copy is bluffing. A system that always sounds sure is lying at least some of the time. We would rather Spuro say a plain thing you can check than an impressive thing you cannot. That standard is in this journal too, which is why the claims here are about incentives and design, things you can verify by using the product, and never about magic.

The AI can still draft, watch, and suggest all day. These rules only cap what it can do to you. That distinction, we think, is what will separate tools people use for a quarter from tools people staff their business on for years.