AI assistants (Claude, ChatGPT)
What it is
General-purpose assistants worked into a marketing team’s week: drafting, research, analysis of data you supply, and code.
What it’s good at
First drafts and restructuring: a brief becomes a working draft in minutes, a long document becomes a short one. Research and analysis with the checking kept in your hands: cited sources you open, and questions answered about data you paste in. And code, from a tracking fix to a full site.
What it costs to own
A written tone of voice: without one, every draft arrives in the default register and gets rewritten, spending the hours the tool was meant to return. Writing it takes days, once. Then a verification habit: anything factual is checked before it ships, and a named person owns the rule, which also covers what goes in; customer data and contracts need a policy before a deadline finds them.
When it’s a poor fit
It fits poorly when the answer cannot be checked. If no one can verify the output (a market size, a legal clause), the assistant’s confidence is the only evidence, and confidence is what it produces most fluently. And it fits poorly where no written voice exists: every draft comes back for rewriting, and the tool is an elaborate way to begin.
What it sits beside
It sits beside everything and replaces the blank page, plus a share of junior research hours. Two overlaps matter. Content drafted at volume reads generic without the voice document. And the assistants are becoming an audience as well as a tool: structured data and llms.txt are how they read your site.
My experience
I maintain this site through an AI assistant, and it has proved a strong partner for drafting and code and a weak source of fact.