In short
- Most agencies now use AI somewhere. The risk is not the tools; it is unapproved, untested, undisclosed use.
- Get the answers in writing, ideally in the contract.
- Good answers are specific: named approvers, a written AI-use policy, tests, a watchdog, dated results.
- Under the EU AI Act’s transparency rules, deep fakes must be disclosed, and AI-generated public-interest text needs disclosure unless a person reviews it and holds editorial responsibility.
Watch: the ten questions
Hiring a marketing agency that uses AI? Ask ten questions first. Who approves? Who owns the system? What is tested, and what happens when it breaks? Which AI tools see your data? Is AI use in the contract? How is synthetic content labelled? Does it automate engagement? Are results dated? And what happens when you leave? Get the answers in writing, ideally in the contract.
Before you hire a marketing agency that uses AI or automation, ask ten questions: who approves, who owns the system, what is tested, what happens when it breaks, which AI tools see your data, whether AI use is in the contract, how synthetic content is labelled, whether it automates engagement, how results are dated, and what happens when you leave.
Most agencies now use AI somewhere, and most should. The risk is not the tools. It is content published without anyone checking it, words attributed to people who never said them, your data sent somewhere you didn’t agree to, and systems nobody watches. These ten questions find out, quickly, whether an agency has thought about that. Get the answers in writing.
We have added our own answers, so you can see what a specific answer looks like. Judge ours the same way.
1. Who approves each item before it is published?
A good answer: a named person on your side, for every item, with a record of who released what and when.
Our answer: nothing goes out until someone on your side says yes. You name the editor and the approvers; any named approver can release an item; the register records it.
2. Who owns the system that produces the content?
A good answer: a clear statement of what you own — content, accounts, data, and ideally the process — and what happens to each when you leave.
Our answer: you own your server, your accounts, your data, your content, and the right to keep running the system we built for you, for good.
3. What is tested before it runs?
A good answer: specific automated checks, with a number and a date, and a rule for adding tests when something breaks.
Our answer: every engine ships with automated checks, and we add a test for every fault we find. In our oil and gas operation that is 285 automated tests (test suite as at 21 Sep 2026).4
4. What happens when it breaks?
A good answer: how they find out, how fast they respond, and an example of a real incident and what changed.
Our answer: if publishing stops, we start fixing it within one working day of the alert or your report, and critical security patches are applied within 72 hours. In September 2026 a settings change stopped one engine’s timed workflows for five days and nothing alerted us. No content was lost; we fixed the cause, added three tests and added a daily watchdog that runs outside the system it watches. It is on our status page.
5. Which AI tools see our data?
A good answer: a list, by task, with what data each one sees and on what terms.
Our answer: AI models are chosen per task and per data-sensitivity, and listed in your Diagnostic. Your data stays on your accounts, and you can opt out of any tool.
6. Is your use of AI written into the contract?
A good answer: yes, as a policy you can read before you sign.
Our answer: yes. Our AI-use policy is public and part of every contract.
7. How do you label synthetic content?
A good answer: they know the rules for your markets and platforms, and label by default. In the EU, the AI Act’s transparency rules (Article 50) require deep fakes to be disclosed, and AI-generated text published to inform the public on matters of public interest to be disclosed unless it has been through human review and someone holds editorial responsibility for it.12
Our answer: synthetic or substantially altered images, audio and video are marked as such, and every item has named editorial responsibility on your side.
8. Do you automate likes, comments or connection requests?
A good answer: no. Automated engagement breaks most platforms’ rules and puts your accounts at risk.
Our answer: never. We also never generate testimonials or invent people.
9. How do you report results?
A good answer: every figure has a period or a date, a source, and a note on what the agency did and didn’t cause.
Our answer: monthly, with dates on every number, and including what didn’t move. In our oil and gas case we say plainly that website conversions and search clicks did not rise over the period.
10. What happens when we leave?
A good answer: the notice period, what you take with you, and how the handover works.
Our answer: one month’s notice at the end of any term, and you keep everything, including the system.
What if the answers are vague?
Vague answers aren’t proof of bad work, but they are a risk you will carry. Ask for the answer in writing; ask for an example; ask to see a real incident report. Search engines don’t object to AI-assisted content as such — Google’s published guidance (February 2023) rewards helpful content however it is produced3 — so the question isn’t whether an agency uses AI, but whether a person stands behind everything it publishes.
Our services start at £38,000 in year one in the UK, including the build and excluding VAT. If you need a few posts a month, a freelancer is a better fit. If you need content going out every week, checked by your own people, book a 20-minute intro call.
Questions
Is it a problem if a marketing agency uses AI?
Not in itself. The problems come from AI use that is unapproved, untested or undisclosed: content published without a person checking it, invented quotes or testimonials, or your data sent to tools you didn’t agree to.
Do we have to label AI-generated marketing content?
In the EU, the AI Act’s transparency rules (Article 50) require deep fakes to be disclosed, and AI-generated text published to inform the public on matters of public interest to be disclosed unless it has been through human review and someone holds editorial responsibility. Platforms also have their own labelling rules. Take legal advice for your case.
Should AI use be written into the contract?
Yes. A written AI-use policy that forms part of the contract tells you what the agency’s tools may do, what they never do, which data they see, and how you can opt out.
How much does a content operation with these safeguards cost?
Beeyawn’s services start at £38,000 in year one in the UK, including the build, excluding VAT. If you need a few posts a month, a freelancer is a better fit.