An engine that writes whatever you ask is working against you

Sooner or later every client asks the engine for a sentence it should not write. It might be a superlative nobody has measured, a green claim with nothing behind it, or a feature from the roadmap described as if it had already shipped. An engine that complies is not being helpful. It is putting the client's name on a claim the client cannot defend, and doing it every morning at scale. Refusing is part of the product.

The request that should come back refused

The awkward requests are rarely outrageous. They come from someone who believes them. The founder is sure you are the fastest onboarding in the Nordics. Sales has been calling the platform carbon-neutral for a year. The product lead says the integration is "basically done" and wants the post out before a trade fair.

Each of those claims might even be true. That is not the test. The test is whether the brief contains anything that would let you show it was true if a customer, a competitor or a regulator asked. When it does not, a fluent paragraph asserting it is a liability with good grammar.

We have already argued that claims about your company come from the brief, and that is still the rule. This post is about the harder case: the brief itself asks for something the evidence does not support, and the person asking is the one paying.

As of today, some of these refusals are not a matter of taste

Directive (EU) 2024/825, the Empowering Consumers for the Green Transition directive, applies across member states from 27 September 2026. It adds to the list of commercial practices that are banned outright. A generic environmental claim such as "eco-friendly" or "climate friendly" is out unless the trader can show recognised excellent environmental performance. So is a claim about the whole business that is really about one part of it. Carbon-neutral claims resting on offsetting are prohibited, and claims about future environmental performance need a published implementation plan checked by an independent expert.

Two concessions belong here. The directive governs business-to-consumer marketing, so a firm that sells only to other firms is not directly in scope, although the underlying ban on misleading claims has older and wider roots. And we are not your lawyers. An engine cannot decide whether a sentence is lawful.

What an engine can do reliably is recognise the kind of sentence. Superlatives, comparisons with named rivals, environmental claims and promises about the future are easy to detect. They are also where nearly all the risk sits. The directive's own recital shows the fix: "climate-friendly packaging" is generic, while "100 % of the energy used to produce this packaging comes from renewable sources" is specific. The first needs proof nobody has. The second needs one fact the client can check.

A brief has to carry evidence, not just assertions

This changes what a good brief contains. It is not enough for the brief to say you are the market leader. It has to say by what measure, as of when, and where that figure is published. A claim with a source attached gets written. A claim without one gets narrowed until it is specific enough to stand on what the brief actually holds.

The same logic covers claims that are true today and will be indefensible next quarter. "Free migration for every new customer" is fine while the offer runs. Six months later it sits in a dated post, still indexed, promising something sales has stopped doing. We have written about how silence lets a brief go stale. Time-limited claims are the version you can see coming. So the refusal can take a gentler form: write the offer, date it in the sentence, and let it age honestly.

Google's guidance on helpful, reliable content asks whether a page contains easily verified factual errors. An unbacked superlative is exactly that, waiting for the one reader who checks. Google has never cared much who wrote the page. It cares a great deal about whether the page is true.

Refusing has a commercial cost, and we should admit it

None of this is free. A client who asked for a bold sentence and got back a hedged one feels corrected by their own vendor. Some will take it well. Some will take it as a sign we do not understand their market, and occasionally they will be right: they know something that should have been in the brief and was not.

So a bare refusal is the wrong output. What goes back is the rewrite plus the one missing fact that would unlock the original: send the benchmark and the superlative returns. Most of the time the specific version is also the stronger sentence, because a reader believes a number sooner than an adjective. When the client supplies the evidence, the claim goes in and the brief is better for it.

A blog that cannot say no is not yours

An engine that publishes anything its client asks for is not loyal to that client. It is only obedient, and obedience at a rate of one post a day turns small overstatements into a searchable archive of them, all carrying the client's name.

The refusal is where the engine earns its keep. It protects the thing an automated blog is supposed to build, a record your customers can quote back to you without it becoming a problem. Write down in the brief which claims need evidence attached, and expect the engine to hold that line even when the person pushing is you.