If your AI output is generic, the fix is the brief, not the tool. Here is the intern test: how to brief AI so it stops giving generic answers, step by step.
Everyone has had this moment. You ask an AI to write something, it hands back a paragraph that is smooth, confident, and completely forgettable, and you think, that is not it, without being able to say exactly what "it" would have been. So you try again, tweak a word, get another forgettable paragraph, and quietly conclude the tool is overhyped.
The tool is fine. The brief was thin. And the fastest way to fix a thin brief is to stop treating the machine like a search engine and start treating it like an intern. Here is how to brief AI so it stops giving you generic answers, using a test you already know how to apply, because you have briefed a junior before.
Before you send the prompt, ask yourself: if I gave this exact instruction to a capable intern on their first day, could they produce what I actually want.
Almost always, the answer is no, and the reason is instructive. You would never tell a new hire "write our brand's launch email" and walk away, because you know they do not have the context to make it yours. They would write a launch email, a generic one, the average of every launch email they have ever seen. Which is precisely what the AI does, for precisely the same reason. It is not a mind reader and neither is your intern. Both are only as good as the brief.
So the intern test is not a metaphor. It is a checklist. Whatever you would have to tell a smart new hire to get good work, you have to tell the machine, in the same words, for the same reason.
A capable intern, and a capable model, needs four things before they can produce anything but the average. Leave one out and the output slides straight back to generic.
Context: what is actually going on. Who the audience is, what the situation is, what has been tried, what matters here. The intern who knows the launch is a relaunch after a rough year writes a different email than one who thinks it is a debut. Give the machine the situation, not just the task.
A point of view: what good looks like here. This is the one people skip, and it is the one that kills the generic. "Make it feel calm and certain, not excited" tells the machine which of a thousand good emails you actually want. Without a point of view, it defaults to the most average acceptable version, because average is the safest guess when no one has said what good means.
Constraints: what to leave out and what to avoid. Length, tone, the words you never use, the claims you cannot make. Constraints are not limitations, they are what turn an infinite blank into a specific target. An intern told "under 120 words, no exclamation points, never call the customer a user" produces something usable. Told nothing, they produce something safe.
An example, if you have one. One piece of work you would be happy to have produced. Nothing sharpens a brief like a "more like this" reference, for a person or a machine. It communicates in ten seconds what a paragraph of adjectives cannot.
Give it those four and the output stops being generic, because you have removed every reason for it to guess.
Take the generic version first. "Write a welcome email for new subscribers." That is a search query, not a brief. The machine has no audience, no point of view, no constraints, no reference, so it returns the internet's average welcome email, and you are disappointed.
Now the briefed version. "New subscribers just downloaded a free tool for brand people who are short on thinking time. They are a little sceptical of marketing emails because they get too many. Write the first welcome email. Make it feel like a note from a peer, not a brand: warm, direct, slightly understated. Under 150 words. No exclamation points, do not tell them they made a great choice, do not sell anything yet. Just deliver the thing and set up what comes next."
Same tool. The second one produces something you can almost ship, because you did the thinking a good brief requires and handed the machine only the writing. You made the decisions. It did the typing. That is the whole trick, and it is the same trick whether the worker is human or not.
The intern comparison holds in one more way worth being clear about, because the hype tends to blur it. The intern only knows what you tell them in the room. The machine only knows what you paste into it. It is not quietly reading your files, watching your brand, or remembering last week unless you put those things in front of it. Being honest about that is not a limitation to apologise for, it is how you get good at briefing: you stop assuming shared context that does not exist, and you start supplying it, which is exactly what makes the output stop being generic.
Pick the one thing you ask AI to do most, the email, the summary, the caption, and write it a real brief using the four elements above. Save that brief. Now you never write the thin version again, and the task that used to disappoint you produces usable work every time.
The free operating pack is essentially a set of these briefs, already built for the tasks brand people repeat every week, in paste-and-keep form, so you can borrow good briefs before you are fluent at writing your own. It is free, so start with the task you are most tired of re-briefing, and let a good brief do the work a thin one never could.
If you want the thinking behind this, read Better questions beat better prompts. The free packs are in the library.
← Back to Thinking