AI for media planning is easiest to use well once you are clear about the split. The model is strong on the repetitive, data-bound work before a plan exists — audience pulls, channel sizing, budget scenarios, a first-draft rationale — and weak on the judgement that decides the plan. The practical question is not whether AI can plan. It is which steps to hand over, how to check them, and where to start.
Most teams point AI at the plan itself and get a faster version of the same deck. The better use is to point it at the steps that never needed a human, free the planner's time for the trade-offs, and keep every number traceable to an input. That is the difference between planning with AI and letting AI produce a plan.
The work before the plan
Media planning is mostly the work nobody sees. Pulling audience data, sizing channels, building budget scenarios, checking reach and frequency assumptions, assembling a rationale that survives a client question. That work is structured, repetitive, and data-bound, which makes it exactly the kind of work a model compresses well.
The planning that matters — the choice of where to play, what to trade, what to leave out — is judgement, and judgement is not what the model brings. So the honest frame is not "AI does planning." It is "AI clears the grunt work so the planner can do more planning." Teams that get this backwards end up with faster decks and the same weak strategy.
What AI is genuinely good at
Four planning jobs suit models well. Audience discovery: turning a brief into candidate segments and the data that describes them. Channel sizing: estimating reach and cost across options from platform and third-party data. Scenario building: generating budget splits across channels and objectives. And first-draft rationale: assembling the argument from the inputs so the planner edits rather than starts blank.
In each case the model is accelerating a step that was already defined. That is the right test for where to apply it. If you can describe the step as a repeatable input-to-output job, AI can help. If the step is "decide what matters," keep it human and use the freed time for it.
What it is not good at
AI is weak at the things that define a good plan. It does not know what the client's business actually needs this quarter unless you tell it. It will happily produce a plausible plan for the wrong objective. It has no feel for the politics, the historical baggage, the category norms, or the one constraint that makes three of its options impossible.
It is also confidently wrong in the places planning cannot afford it: invented benchmarks, stale channel assumptions, and reach estimates treated as facts. A planner who pastes model output into a plan without checking the inputs has outsourced the judgement, not the labour. The output is a hypothesis, and every number in it needs a source you trust.
How to tell which side a task falls on
Not every planning step announces its side, so run each one through a single test: could you hand the input to a new hire and describe, without judgement, what the output should be? Audience pulls, channel sizing, and scenario generation pass — the job is defined before it starts. A trade-off between two channels, or a call on what the client should stop doing, fails — the answer depends on context the model was not given.
A second signal: ask whether the output can be checked against a source you trust. If you can point to the data behind every number, the step is safe to hand over. If you cannot, it is judgement wearing an arithmetic costume, and it stays with the planner.
Designing for defensibility
A plan has to survive a client asking "why this and not that." AI-assisted planning makes that question harder, not easier, because the model can generate a rationale for any option. The defence is to keep the inputs visible: which data fed the sizing, which assumptions the scenarios rest on, and which choices were judgement rather than arithmetic.
This is the same discipline I set out in AI media planning and measurement design, and it matters more at this layer because planning sits upstream of everything else. A plan built on unexamined model output propagates its errors through activation, reporting, and measurement. Traceability at the planning stage is what keeps the rest of the campaign honest.
The planner's new job
The role does not disappear; it concentrates — the same shift I described years ago in how media planners would lose their jobs, when the automation was a research tool and a spreadsheet. Less time assembling and sizing, more time deciding which scenario to back and explaining why. That is a harder job than the old one, and a more valuable one, because the thing clients cannot get from a tool is the reasoning behind the plan. Planners who lean into that become the ones the client calls before a budget is set.
It also raises the bar on inputs. If the planner's value is judgement, the judgement is only as good as the data it is judging. That is why the memory and data layers underneath planning matter so much: a model working from your client history and results gives the planner better material to reason over than one working from a blank brief.
A practical starting point
Pick the single most repetitive planning step — usually audience sizing or scenario building — and rebuild only that with AI, keeping the output traceable to its inputs. Run it alongside the old method for a plan or two and compare. If the AI-assisted version produces more scenarios without producing more errors, expand to the next step. If it produces confident numbers you cannot source, stop and fix the inputs first.
Do not try to automate the whole plan in one move. Planning is a chain of judgement calls, and the value of AI is at the joins, not as a replacement for the chain. Rebuild one join at a time and keep the planner in the loop at each.
Where AI planning goes wrong in practice
The failures are predictable. Teams feed the model a brief and accept the first plan, skipping the comparison that would have shown three better options. They let the model size channels from stale assumptions and never check the inputs. They use it to write the rationale last, as an explanation for a decision already made, which is the reverse of how planning should work. And they treat its confidence as accuracy, when the model is equally fluent about the option it is wrong about.
The correction is procedural, not technical. Always ask for several plans, not one, and force the model to state its assumptions alongside each. Make the planner choose and justify, so every decision carries a human name. Keep the inputs visible, so any number in the plan can be traced to a source. None of this is exotic; it is the discipline planning always needed, applied to a faster tool. One more habit helps: keep the old method running alongside for a cycle or two, and compare the plans. When the AI-assisted version and the manual version disagree, the difference is the most useful thing you will learn all quarter, because it points to the exact assumption the model is resting on. If that assumption is one you would not defend, you have found the flaw before it reached a client.
FAQ
Will AI replace media planners?
No — it takes the assembly, not the judgement, and the full answer explains what actually changes, while AI media planning and measurement design covers how to redesign the plan and its measurement together.
Can the model size channels accurately?
It can estimate from the data you give it, and it can be confidently wrong. Treat every sizing number as a hypothesis with a source, not a fact. The model is useful for exploring assumptions, not for replacing the vendor and third-party data you would normally trust.
Where should a team start?
With the most repetitive, well-defined step in the planning process, run in parallel with the current method so you can compare. Starting with the hardest judgement call is the fastest way to conclude that AI does not work for planning when actually you pointed it at the one thing it is worst at.
Does AI planning change the plan's structure?
Not much. The plan still needs an objective, a strategy, a channel mix, a budget, and a rationale. What changes is how many options the planner explored before choosing and how quickly they could explore them.
The short version
AI for media planning compresses research, sizing, and scenario work so planners spend more time on trade-offs and less on assembly. Use it for the repeatable steps, keep the judgement human, and keep every input traceable. The Execution guide covers the wider workflow, and the for team leads track works planning through the operating model.
If you run planning with AI today, I would like to hear which step you rebuilt first — and which you would never touch.
Cheers, Chandler