What is agentic marketing? 5 questions before you sign

Every marketing vendor now says agentic. A platform that drafts posts calls itself agentic. A scheduling tool with an AI assistant calls itself agentic. A department that plans, writes, designs, publishes and reports calls itself agentic. The word covers all three, and the three do wildly different amounts of your work.

Agentic marketing is marketing run by software that takes a goal and carries the steps out itself: it plans, produces, checks and finishes the work, then hands it to you. You set the outcome and the limits. It works the hours in between. That is the whole definition, and it is also the whole problem for a buyer, because the word alone does not tell you which of the three you are looking at.

So stop asking what the word means. Ask these five questions, in order, and any agentic marketing pitch turns into a comparable answer sheet. Use them on the demo, the trial and the sales call. They work on every vendor, including us.

What agentic marketing actually means

Strip the word down and it means one thing: the software holds a goal across more than one step. A generative AI tool waits for a prompt and returns a draft. An agent holds the goal, decides the steps, calls the tools that change things in the world, and checks each result before choosing the next move. A copilot sits beside you. An autopilot flies the route and checks in at the points that matter.

In practice, an agentic marketing product builds and stages a campaign from a plain-language brief, builds audiences, answers performance questions against your own data, moves spend toward what converts, and watches for problems before you go looking. Steve Muylle of Vlerick Business School writes that AI agents perform many marketing tasks on your behalf with minimal intervention, and that you remain accountable for the outcomes and for how the agents are used.

Where its work ends and yours begins is the first thing a pitch should make clear. Question one is how you find it.

Question one: what exactly does it do every day, without being asked?

Ask for a day in the life. The feature list can wait. What does it do on its own on a Tuesday when you have said nothing to it?

Answers come in three shapes. Some products act only when asked: you prompt, they draft, they wait. Some run on schedules and triggers: they check performance, they draft, they queue. A true agentic product holds a goal and works toward it across steps, and can name those steps: it notices something, decides what to do, does it, checks the result.

If the salesperson cannot describe what happens on a day you gave it no instructions, the product is a prompt tool with agent branding. That may still be worth buying, but it is a different purchase at a different price, and you should know which one you are signing for.

Question two: what does it need from you before it starts?

An agent is only as good as what it can read and reach. This is the question the demo never answers on its own, because the demo runs on a prepared account.

Ask what the product reads before it works. Your website? Your campaigns, audiences and product data? Data from other tools, and how it gets there? Ask whether it reads live data or a synced copy, because an agent reasoning over yesterday's snapshot reasons confidently about customers who have already changed. Ask what connections it has on day one and what you have to build: the integrations page of any vendor is where that answer is either published or missing (as ours is at https://nouna.io/integrations).

Then ask who does that setup. Some products learn from your website in minutes. Some need a data project, an integration budget and a team. The gap between those two is often the gap between a monthly subscription and an enterprise rollout, and the price list will not tell you. The setup burden is the real price.

Question three: what still waits for a person's yes?

This is the question that separates a real product from a risk. Ask which actions publish without your approval and which ones stop and wait.

You want a specific answer: what the agent does on its own, and what stops and waits for you. The products worth trusting say the second list out loud. Ads, website changes and anything that spends money should wait. Content should be visible before it goes out, and asking for a change should mean typing what you want and getting a new version back, with no support ticket in between.

If the answer is everything is autonomous, ask what happens when a piece is wrong. If the answer is nothing is autonomous, you have bought an assistant and should price it as one. Either is defensible. Vagueness is not. Every vendor claims speed; the question is whether you can see how a piece was made and stop it before a customer sees it.

Question four: how does it prove the work it did?

An agent that acts while you are not looking needs to leave a record. Ask what it shows you: what it did, why it did it, and what changed because of it.

You are listening for three things: a log of the actions it took, the reasoning behind each recommendation so you can question it, and results measured against your own numbers. Benchmark screenshots do not count. And you want all of it on demand, because a monthly PDF means your questions wait a month.

This matters more in agentic marketing than in any tool category before it, because the agent makes more decisions per week than a person you could ask at the coffee machine. A recommendation you cannot question is a guess with good posture.

Question five: what does it cost to leave?

Nobody asks this one and it is where the worst surprises live.

Ask whether your data, brand documents, content and connected accounts export in a usable form. Ask whether the contract auto-renews, and on what notice. Ask what happens to the work it made for you on the day you cancel: if six months of posts, pages and campaign structures leave with the vendor, you rented your own marketing instead of building it. A vendor's help centre usually answers cancel and exit questions before sales does; ours is at https://nouna.io/help.

The exit question doubles as a character test. A vendor who answers it quickly and in full has thought about what its product is worth against what you could move elsewhere. A vendor who stalls is telling you the answer.

The five questions, side by side

Scope: what does it do every day without being asked? Inputs: what does it need from you before it starts? Control: what still waits for a person's yes? Proof: how does it show the work it did? Exit: what does it cost to leave?

Five questions, one conversation, and you leave with a comparable answer sheet no two vendors can fake, because each answer is either specific or it is a pitch. Ask them in order: scope tells you what you are buying, inputs tell you the real price, control tells you the real risk, proof tells you the real accountability, and exit tells you the real relationship.

One more test that costs nothing: after the demo, put your own notes through the same five questions. A product that answers all five in specifics, needs little from you, keeps you in control, shows its work and lets you leave cleanly is rare. Anything that clears all five belongs on your shortlist.

What the answers told you, and where Nouna lands on them

Score the answers against the job you actually have. If you need a copilot that drafts when asked, most of the category qualifies and price should decide. If you need daily production handled, you are looking for full answers on scope and control, not the biggest feature list.

Ours, in the same order: Nouna is a subscription AI marketing department, a CMO and eight channel specialists. They plan, write and design daily across social, email, blog and search, and publish to the social and blog channels you connect (nouna.io/product). Email arrives as Klaviyo drafts and search work arrives as fixes to your site, so Google and AI answers like ChatGPT find and quote you. It learns from your website, so setup is one step: your site. Ads, new pages and website rewrites always wait for your approval, and every piece is visible before it publishes. Work is metered in credits with published volumes, so the output is countable. Plans are monthly, published in full at nouna.io/pricing, and you can leave.

That is the answer sheet filled in. Run the checklist on us, then on anyone else. The free start is where the questions stop being hypothetical: start free at https://nouna.io/start, no card, and see what the department makes from your own website in the first week.