Choosing a firm

Ten questions to ask an AI-native marketing firm before you sign

AI changes how marketing gets made. It does not change who answers for it: the brand does. These ten questions show whether a firm’s process protects yours.

By Lightvector 22 October 2026 7 min read

In brief

01Regulators in the UK and the US say the same thing: advertising rules apply however an ad was made, and the advertiser stays responsible for it.
02Most brands have not caught up. In one survey of large advertisers, 80% had concerns about how agencies use generative AI and only 36% had contract terms covering it.
03The ten questions cover four things: who checks the work, how claims are kept true, who owns what, and how results are proven.
04Good answers are specific: a named person, a written process, an example from a real client. Vague answers are the warning sign.

Why it matters: you stay responsible

When a firm makes your ads with AI, the law treats the result exactly as if a person had made it. The UK's Advertising Standards Authority has said so directly: its rules apply “regardless of how the ad was created”, and even where a campaign is generated entirely by automated methods, the advertiser keeps the primary responsibility for it.¹ In 2026 it repeated the point: if an AI tool produces something problematic, the advertiser remains responsible for any output used in their ads.²

The United States takes the same view. Announcing a set of enforcement actions in 2024, the Federal Trade Commission said there is “no AI exemption from the laws on the books”.³

The penalties are no longer small. Since April 2025 the UK's Competition and Markets Authority can decide for itself that consumer law has been broken, without going to court, and fine a business up to 10% of its global turnover.⁴ And a tribunal in Canada has already ruled on the obvious defence. When Air Canada's chatbot gave a customer wrong information, the airline argued in effect that the chatbot was responsible for its own words. The tribunal rejected that and ordered the airline to pay.⁵

80%

of large multinational advertisers had concerns about how their agencies use generative AI⁶

36%

had put terms into their agency contracts to govern it⁶

Brands know this is a risk and most have not yet acted on it. The World Federation of Advertisers found that 80% of the large advertisers it surveyed had concerns about agency use of generative AI, led by legal, ethical and reputational risk, while only 36% had contract terms covering it.⁶ The survey covered some of the world's biggest companies; smaller brands are unlikely to be further ahead.

So the useful questions are not about which AI a firm uses. They are about its process.

The work: who checks it

01Who reviews each piece before it reaches me?

AI can produce a hundred ads in an afternoon. Some will be off-brand, some will show what you sell wrongly, a few will say things you cannot say. The question is who catches those. “The AI checks itself” is not an answer, and neither is “you do”.

A good answer: a named, experienced person looks at every ad and email before you do, and can show you what they rejected last week and why.

02What can go live without my approval?

Speed is the point of an AI-native firm, and approval is where speed and safety meet. You should know exactly which actions need your yes, which need the firm's operator, and which the software takes alone. Pausing a losing ad is a very different thing from publishing a new one.

A good answer: nothing your customers see goes out without your sign-off at the start; anything automated later is agreed in writing, one step at a time.

03Does the ad show what I really sell?

Image models are good at making something that looks nearly like yours: a product with the label slightly wrong, premises that are not yours, a screen your software does not have. An ad showing something that does not quite exist is a misleading ad. Ask how the firm keeps the real thing in the picture.

A good answer: the work is built from your own photographs, screenshots and brand files, and they can show you a finished ad beside the material it came from.

The truth: how claims are kept honest

04Where does each product claim come from?

A language model will write “clinically proven” as readily as “new”. It does not know which one you can support. The firm needs a list of the claims your brand can make, the proof behind each, and a check that stops anything else reaching an ad. This matters most in health, beauty, food and finance, where the rules on claims are strict.

A good answer: every claim in the copy traces to a source you approved, such as your site, a study or your own guidance, and there is a written list of what must never be said.

05How do you handle reviews, testimonials and real people?

AI can invent a convincing customer quote in a second, which is exactly why regulators have moved. The FTC's rule banning fake reviews and testimonials, in effect since October 2024, names AI-generated fake reviews explicitly and carries civil penalties.⁷ The UK's 2025 consumer law bans posting and commissioning fake reviews.⁴

A good answer: only real reviews from real customers, quoted as written; and no real person shown or quoted without their actual words, photograph and permission.

06When do you label work as AI-generated?

The rules differ by place. In the UK there is no blanket legal requirement to disclose AI in ads; the ASA's test is whether the audience is likely to be misled if it is not disclosed.⁸ Meta labels ads made with its own generative tools and said in 2026 that it would begin detecting ads made with third-party tools as well.⁹ In the European Union, the AI Act's transparency duties, which include disclosing deepfakes, began to apply in August 2026.¹⁰

It is also worth knowing how your customers feel. An IAB study in the United States found that 82% of advertising executives believed younger consumers feel positive about AI-generated ads, while only 45% of those consumers said they do, and that fewer than half of advertisers using generative AI always disclose it.¹¹

A good answer: they have a written policy, they know the rules for each platform and market you sell in, and they label wherever a viewer could otherwise be misled.

Ownership: what is yours

07Who owns the ad accounts, the email platform and the creative?

Your ad account holds years of performance history and audience data, and it is hard to rebuild. Some agencies run clients' advertising in accounts the agency owns, which makes leaving expensive. The same goes for your email platform and for the finished creative.

A good answer: everything runs in accounts you own, with the firm added as a partner; if you part ways you remove their access and keep the accounts, the history and the work.

08What happens to my data and my brand material?

An AI-native firm reads a great deal about your business: sales, customers, reviews, what works. Ask where that is stored, who can see it, whether any of it informs work for a competitor, and whether the AI providers the firm uses can train on it. In the WFA survey, only 29% of large advertisers had reviewed their agency contracts to add AI terms.⁶

A good answer: your material is used only for your work, it is not shared with other clients or used to train models, and the contract says so.

Proof: how results are shown

09How do you show the results are real?

Every ad platform reports its own results, and those reports tend to flatter the platform. A study of 663 advertising experiments on Facebook found that standard non-experimental methods could not recover the true effect of the advertising, and were often wrong by more than the size of the effect itself.¹² A firm that only shows you the platform's dashboard is passing on the platform's view.

A good answer: they report against your own sales or enquiry figures as well as the platform’s, they explain the difference between the two, and they run proper tests where the budget allows.

10What did you learn for a client last month, and what changed because of it?

This is the question that separates a firm that produces from a firm that improves. Making a lot of work is now easy. Getting better each month requires that results are read, understood and used. If the answer is a list of metrics, the learning is not happening.

A good answer: a concrete example: this is what we found, this is what we stopped, this is what we made next.

And read the contract

Before signing, check the minimum term, the notice period, what the price is based on, and what you keep when the contract ends. A firm confident in its results does not need a long lock-in.

This article is general guidance, not legal advice. For claims in regulated categories, take advice for your market.

How we answer them

We wrote this list expecting to be asked it. In short: a senior operator reviews every piece before you see it, and you approve the work your customers will see. Copy is written from a record of your brand's approved claims and real reviews. The work runs in your own Meta and Klaviyo accounts, and new ads are created paused. Results are joined to the piece that produced them, and what we learn is written down for your brand.

Ask us all ten on a call.

Sources

1ASA and CAP, Generative AI & Advertising: Decoding AI Regulation, 31 August 2023.
2ASA and CAP, AI and Deepfakes: Four Things Advertisers Need to Know Before They Hit ‘Run’, 11 June 2026.
3US Federal Trade Commission, FTC Announces Crackdown on Deceptive AI Claims and Schemes, 25 September 2024.
4UK Competition and Markets Authority, CMA to boost consumer and business confidence as new consumer protection regime comes into force, 7 April 2025.
5Dentons, Airline ordered to compensate a B.C. man because its chatbot provided inaccurate information, 15 February 2024. A law firm's account of Moffatt v. Air Canada, 2024 BCCRT 149, a customer-service case.
6World Federation of Advertisers, Eighty percent of brands have concerns about agency use of GenAI, 17 September 2024. 54 responses from 48 large multinational advertisers.
7US Federal Trade Commission, Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials, 14 August 2024.
8ASA and CAP, Disclosure of AI in Advertising: Striking the Balance Between Creativity and Responsibility, 29 May 2025.
9Meta, Expanding GenAI Transparency for Meta’s Ads Products, 3 February 2025, updated 1 June 2026.
10EU Artificial Intelligence Act, Article 50: Transparency Obligations for Providers and Deployers of Certain AI Systems, 2024. An unofficial rendering of Regulation (EU) 2024/1689. A 2026 amendment delays one marking duty for existing systems to December 2026.
11IAB, The AI Ad Gap Widens, 15 January 2026. United States; 505 consumers aged 16 to 43 and 104 advertising executives.
12Gordon, Moakler and Zettelmeyer, Marketing Science, Close Enough? A Large-Scale Exploration of Non-Experimental Approaches to Advertising Measurement, 2023.