12 questions before an AI agent spends your media budget

AI agents start to book, change and buy on behalf of advertisers. Before one of them touches your budget, these are the questions your team, your agency and your vendors should be able to answer. Print it, take it to the next meeting.

  1. Which campaigns may an AI agent book, change or stop — and who signed that permission?
  2. Can you show afterwards what the agent was allowed to do, as a signed record and not only as a log line?
  3. When an agent buys on your behalf, does the shop or ad network see that it is an agent, and under whose mandate?
  4. How do you tell a visit or click by your own agent from a fraudulent bot?
  5. Which billed events (visits, impressions, purchases) can you prove yourself, and which does only your vendor report?
  6. Who holds the evidence — you, your agency or the platform — and for how long?
  7. If a creative leaks or is cloned, can you tell which agency's or placement's copy it was?
  8. Which of your creative formats survive the re-encoding of an ad server? Has anyone measured it on your own creatives?
  9. Does your verification vendor score traffic, or can it prove a single event? Both are useful — they are different jobs.
  10. What happens when the agent meets a manipulated landing page or a prompt injection?
  11. Is there a budget limit the agent technically cannot exceed?
  12. Who can check the agent's receipts — and can a third party check them without your software?

Where we can help today

In our lab, not yet at a paying customer: questions 2, 5 and 12 (signed receipts for visits and purchases by agents, lab only); question 8 (a key in social-format creatives, measured on our own creatives after simulated JPEG/WebP re-encoding and scaling, not through a real ad server; banners cannot be marked); question 7 only partly — telling several placements apart is not measured yet. Questions 4 and 9 stay with your verification vendor.

Talk about your answers