The Coffee Writer

There is a conversation around AI that gets told so often now it has started to feel like fact. There’s this presumption that the tools are so good that almost anyone can produce professional-quality content with the right prompt. The implication (sometimes stated, often implied) is that the expertise of professional writers has become largely redundant.

For the entrepreneur and the C-suite, it’s a neat story and leads to boasting dinner conversation. It is also wrong. Not wrong in a defensive, wishful-thinking way. Wrong in a verifiable, functional way that becomes apparent the instant you actually try to use these tools at a professional level.

Here is the thing that the neat and tidy story leaves out: the quality of AI-generated content is a direct function of the knowledge and expertise of the person directing it and managing what it creates. The tool is only as good as the knowledge behind the human using it.

WHAT AI TOOLS ARE ACTUALLY DOING

Large language models, aka the technology behind ChatGPT, Claude, and the now hundreds of other writing tools built on top of them, are sophisticated pattern-matching systems. They were trained on an enormous body of text and learned, at a statistical level, what good writing in various contexts looks like. When you give them a prompt, they predict what a good response would look like.

The prediction can be remarkably fluent. But fluency is not the same as accuracy. Fluency is not the same as insight. Fluency is not the same as knowing whether the argument you are making is the right one for this specific audience at this specific moment. The model produces text that sounds like it knows what it is talking about. Knowing whether it actually does requires a human who actually does.

THE PROMPT IS THE TEST

The simplest illustration of that knowledge-dependency of AI output is the prompt itself. Write a vague prompt such as ‘write a blog post about content strategy for small businesses’ and you will get vague, generic output. Every sentence will be technically true. None of it will be specific enough to be useful.

Now write an expert level prompt, one where you specify the keyword, the search intent, the audience’s knowledge level, the angle, the H2 structure, the tone, the specific questions the post needs to answer, the internal links to include, the length, and the voice. The output will be categorically different. Not because the AI became smarter. Because the expertise in the room increased.

The person who can write the second kind of prompt that produces genuinely useful, strategically sound, audience-specific content has the same domain knowledge, the same audience understanding, and the same editorial judgment that the professional writer always brought to their work. The expertise did not disappear. It moved upstream, from the execution of the work to the direction of it.

PRO TIP 

The quality gap between vague and expert prompts is roughly the same as the quality gap between a vague brief and a complete one. If you have ever tried to write from a client’s brief that said only ‘write something about our new product launch,’ you understand exactly what the AI is working with when it receives a vague prompt.

WHAT GETS LOST WITHOUT THE KNOWLEDGE BASE

The consequences of using AI without the knowledge to direct it are visible, even if they are not always immediately obvious.

Hallucination: AI tools fabricate facts with confident fluency. They invent citations, misstate statistics, and produce plausible-sounding claims that are simply false. A writer with topic expertise catches these made-up “facts” immediately. A non-expert often does not.

Generic positioning: Without deep audience knowledge, AI output defaults to the median. It covers what everyone else covers, in the same way everyone else covers it. The specific angle, the surprising insight, the claim that only an expert in this niche would think to make are all absent. It adds to the noise cluttering the internet and will get lost behind every other page out there when a search engine comes along looking to answer a user’s search query.

Voice erosion: Brand voice is the accumulated product of deliberate choices made over time. AI does not know your client’s brand history, their previous messaging, or their customer’s vocabulary. It produces something that sounds like the category, not like the brand.

Strategic misalignment: The most expensive errors in content are not typos. They are strategic screw-ups. Writing for the wrong search intent. Targeting the wrong keyword. Structuring a piece for the wrong funnel stage. These are judgment calls that require expertise to make correctly.

THE PRACTICAL IMPLICATION

For experienced freelancers, the implication is both reassuring and demanding. Reassuring because the expertise that makes your work valuable has not been automated. Instead, it has become the essential first step for getting useful output from the tools. Demanding because it requires you to be clear about what that expertise actually is, and to position it explicitly in a market that sometimes confuses the output with the work.

The tool is not the threat. The conflation of the tool’s output with the writer’s work is the threat. And the answer to that conflation is clarity: about what you bring, why it matters, and what happens when it is absent.

The takeaway:
AI tools amplify existing expertise. They do not create it where none exists. The knowledge base that makes professional writing valuable has not become less necessary. It has become the thing that makes AI useful.

A skilled writer using AI produces better, faster work. A non-expert using AI produces confident-sounding text that lacks the judgment that makes content effective.