Every week someone sends me a list of forty AI tools that will change marketing forever. I have tested a painful number of them, first for my own companies and then with clients. Most did not survive a month. What follows is not a listicle. It is the shape of a stack that survived contact with real work, and the rule that decides what stays.
The rule
A tool earns a place if it removes a step from work we already do every week. Not "could be useful". Not "impressive demo". A named step, in an existing workflow, done faster or better. Tools adopted for their demo die within a month, every time, because nobody changes their behaviour for a maybe.
Where AI genuinely changed my work
Research and synthesis. First drafts of market overviews, competitor teardowns, and summaries of long documents. What used to be a junior analyst's week is now an afternoon of directing and checking. The checking is not optional: the tools remain confident when wrong, and the errors cluster exactly where you are least able to catch them, in the fields you know least.
Writing variations at volume. Not the strategy, not the core message. But once the message exists: ten versions for testing, adapted lengths, translated registers. This is the safest, highest-yield use in marketing today because judging a variation is fast and cheap.
Reporting. Turning raw numbers into a readable first-draft narrative. The judgment about what the numbers mean still has to be human, because the numbers mean nothing without knowing what the business was trying to do that month.
Meeting follow-through. Transcripts into action lists, briefs and recap emails. Boring, and probably the highest ratio of time saved to effort invested in the whole stack.
Where I stopped
Fully automated content pipelines. The output is grammatical, fluent and dead. Audiences might not consciously spot machine writing, but they stop responding to it, and the metrics say so within weeks.
AI-driven media buying beyond the platforms' own systems. Meta and Google already run serious machine learning on their side. Layering a third-party "AI optimizer" on top mostly added cost and a dashboard.
Anything that promises strategy. Ask a model for strategy and you receive the average of everyone's published strategy, which is precisely the thing a strategy must not be.
How I roll AI into a client's team
One workflow at a time, owned by a named person, measured before and after. The pattern that fails: buy licences for everyone, announce a revolution, watch usage collapse to two people after a month. The pattern that works: pick the ugliest recurring task, wire a tool into it properly, publish the before-and-after internally, and let the team pull the next one in themselves.
That difference between installing tools and changing how a team works is most of what my AI integration advisory is. The tools are the cheap part. The workflow is the work.