When should a content engine run without a human?

What is a content engine?

A content engine is a system that plans, researches, drafts, publishes and measures marketing content on a fixed rhythm. It combines AI generation with written rules: what the brand may say, which sources count as evidence, who approves what and how performance is read back into the next cycle.

The question is never whether an engine uses AI. It is where in the cycle a person must decide, and where a machine may act alone. Michael S.U. Hudson draws that line separately for each client and each channel, and writes it down before the engine runs.

What can a machine check on its own?

A machine can check rules that are explicit and testable. Has this post been approved before? Is it inside the publishing window? Does it repeat a claim already cleared? Is the email list authenticated, and is the complaint rate below the threshold? When every rule in a step is of this kind, the step can run without a person.

A machine cannot check whether a new claim about a product is true, whether a sentence will read badly in a politically sensitive market, or whether a chief executive would be comfortable seeing it under the company name. Those are judgements. They need a person, and the engine should stop and wait for one.

What do the platforms allow?

Each channel sets its own limits, and an autonomous engine has to work inside all of them. Google describes scaled content abuse as producing many pages mainly to manipulate search rankings, whether by people or by automation, and treats it as spam. There is no page quota; the test is whether the pages add value. An engine watches how much of what it publishes Google actually indexes, and slows down when that share falls.

Gmail asks every sender to authenticate mail and keep the spam rate reported in Postmaster Tools below 0.3 per cent. Bulk senders, those sending around 5,000 or more messages a day to Gmail addresses, must also publish DMARC and offer one click unsubscribe. An email engine that passes those limits stops itself.

Meta’s spam policy covers posting, sharing and engaging at very high frequency, whether by hand or by automation. LinkedIn’s User Agreement prohibits bots and other unauthorised automated methods to create, comment on, like or share content. YouTube’s monetisation rules exclude mass produced and repetitive content. Read together, these rules permit automated publishing through approved tools and forbid automated engagement that pretends to be a person.

Where must a person stay in the loop?

Three decisions stay with people in every engine Michael designs. The first is any new claim: a product capability, a price, a partner or a result. The second is anything that represents a person in conversation, such as a reply, a comment or a connection request. The third is anything hard to reverse, from a paid campaign to a statement in a regulated market.

For Geepas, a home appliance brand, people approved every start frame and every final edit of AI generated video ads, because the product had to be exact in every frame. AI video tends to invent buttons and redraw logos; only a person comparing each output with the reference photography could clear it.

For an industrial client in a politically sensitive market, every public asset is approved by the chief executive. That creates a bottleneck, so the engine runs under a written rule: approved evergreen content keeps publishing when approval is delayed, while every new claim waits. The engine stays steady without ever publishing something no one has cleared.

How do you decide for a new client?

Start with the channels and list each platform’s written limits. Then list every step in the cycle and mark each one as machine checkable or a judgement. Where a step is checkable and a mistake is cheap to reverse, automate it and monitor it. Where it is a judgement, put a named person at that point, with a time limit and a fallback.

Review the line each month. As evidence builds that a type of content is safe, more of it can move to autonomous publishing. As platforms change their rules, some of it may need to move back. The engine is reported weekly, improved monthly and approved by people.