Ask a chatbot to write you a campaign brief and you get something that sounds like a campaign brief. Competent, plausible, and completely disconnected from your brand, your last three campaigns, and the reason the client said no in March.
That is not the model being weak. That is the model working blind.
For a long time my workflow was: open a chat, then spend the first ten minutes pasting in context. The tone guide. Last quarter's numbers. The product description. The email that performed well so it had something to copy. By the time it had enough to be useful, I had done a fair chunk of the thinking myself.
What has changed is that Claude can now be connected to the places that context already lives. Your files. Your email. Your drive. And the difference in output quality is much bigger than I expected.
The gap is context, not intelligence
Here is the thing marketers underestimate. When AI output feels generic, the instinct is to write a better prompt. Usually the real problem is that you have asked it to work from nothing.
Every good brief you have ever written drew on things you did not consciously list: how the brand sounds, what the audience has already been told, what got rejected last time. You did not prompt yourself with those. You just had them.
Give the AI access to the same material, and a lot of the prompt engineering stops being necessary. "Write in our tone of voice" is a weak instruction. "Here are eleven things we published this year" is a strong one, and now you can just point at the folder.
What actually changes day to day
Three things got noticeably better once I stopped working from a blank box.
It stops asking me for things I already have. The brand guidelines are in the folder. The past campaigns are in the folder. It reads them instead of asking me to summarise them.
The first draft starts closer to right. Not finished. Closer. When it can see how we normally write, the draft needs a trim rather than a rewrite, which is the difference between saving twenty minutes and saving five.
The boring cross-referencing gets done. "Have we already said this?" "Which of these decks covers pricing?" "Pull the three case studies that mention retail." That is dull work I would put off, and it is exactly what something with access to the whole folder is good at.
The part I am careful about
Connecting AI to your actual work is a permissions decision, not just a productivity one. I treat it that way.
I connect a folder for the project I am working on, not my entire drive. I keep it in the mode where it asks before it does anything that touches the outside world. And I do not connect anything holding client data or commercially sensitive numbers that I would not be comfortable having read.
Being useful and being careless are not the same thing, and the setup lets you have the first without the second. But only if you actually make the choice rather than clicking through it.
It is not doing the marketing
Worth saying plainly, because the hype gets this wrong.
None of this decides what the campaign should be. It does not know which audience matters, which trade-off is worth making, or why the thing that worked last year will not work again. That is still the job.
What it removes is the tax you pay before the job starts. The pasting, the re-explaining, the hunting for the file. On a normal week that is a surprising amount of my time, and I would rather spend it on the part only I can do.
If you want the practical version of this, I have written a step-by-step for connecting Claude to your own files and tools: start here.