The trust discount on AI-generated content

The cost of producing passable content has fallen to roughly zero. Anyone can now generate a competent-looking blog post, product description, or LinkedIn think-piece in the time it takes to write the prompt. For a while the obvious marketing move looked like volume: if content is nearly free, make more of it.

Audiences adjusted faster than the content strategies did. As the web filled with fluent, generic, machine-shaped text, readers started applying a trust discount — an instinctive markdown on anything that pattern-matches to 'this was generated, not known.' That discount is now a real cost, and a lot of content programmes are paying it without noticing.

Why fluency stopped being a signal

Good writing used to be weak evidence of effort and competence. It took time and some skill, so a well-written page implied someone bothered. Generative tools severed that link. Fluency is now free, which means it no longer signals anything about whether the writer actually understood the subject.

So readers look for what is still scarce: specificity that could only come from real experience. A concrete number, a named trade-off, a mistake the writer actually made, an opinion that risks being wrong. These are the things a model, asked to write generically, tends to smooth away — and their absence is exactly what makes content read as machine-made even when it is technically accurate.

The tell of AI slop is rarely an error. It is the absence of anything only a specific human could have said.

The discount is not evenly applied

Audiences are not rejecting AI involvement wholesale — most neither know nor care which tools touched a piece. What they reject is the feeling of content that no one stood behind. The discount lands hardest where trust matters most: anything advisory, technical, or reputational. It lands lightest on the purely functional, where nobody expected a soul in the first place.

That is the useful distinction for a marketer. Using AI to draft, edit, translate, or accelerate is not the problem. Publishing the ungrounded, unspecific middle of what a model produces — and hoping volume compensates — is.

What actually beats the discount

The teams keeping their content credible are not the ones avoiding AI. They are the ones using it to remove the grunt work so humans can spend their time adding the parts a model cannot fake:

• Proprietary specifics — your data, your results, your customers' patterns.
• A real point of view — a position that some readers will disagree with.
• Named accountability — a human byline that stakes a reputation on the claim.
• Lived detail — the texture of having actually done the thing.

The strategic inversion

The reflex — content is cheap, so make more — is precisely backwards. When production is free, production is no longer the scarce input, and flooding the channel with more generic material actively deepens the discount by making your brand look like part of the noise.

The scarce input is now knowing something worth saying and being willing to put a name on it. Generative tools are best pointed at everything around that core — the drafting, the formatting, the repurposing — so the human effort concentrates on the part that earns trust back. Spend the time you save on substance, not on volume. The market is quietly repricing content, and substance is the only thing holding its value.