Running ads without generative AI.

We learned this properly working with Jim'll Paint It. When the whole value of a brand is that a real person sat down and made the thing, using a model trained on other people's work to advertise it is not a bad look. It is a contradiction the audience will spot immediately, and artists' audiences are, in our experience, the most switched-on audience on the internet when it comes to spotting AI slop.

So we made it a standing option, and for some clients a standing rule. If you want that written into the engagement, we put it in writing. It costs us nothing to commit to, because it is how we prefer to work anyway.

Why artists are right to be wary

The objection is not technophobia. It is specific, and it holds up.

Training data. Image models were trained on scraped work, largely without consent, credit or payment. Using one to advertise an artist's work launders that problem straight into their brand.

It undercuts the offer. If your pitch is "a human painted this", an AI-generated ad image tells the buyer the human part is negotiable.

Audiences retaliate. Art communities are fast and vocal. One AI-looking ad in the feed can turn a comment section into a pile-on that costs more than the campaign earns.

Voice matters more than volume. Most artist brands sell on a distinctive voice. Generative copy flattens it into the same cheerful mush every other brand is running.

What "no generative AI" actually means in practice

It is worth being precise, because "no AI" is a broad claim and the platforms are full of AI whether you like it or not. Here is where the line gets drawn.

What we do not do. No AI-generated or AI-extended images, backgrounds or product mockups. No AI-written ad copy, headlines, captions, emails or blog content for the brand. No AI-generated avatars or synthetic spokespeople. No AI voiceover pretending to be the artist. No feeding the artist's catalogue into an image model for "variations".

What we do use. The platforms' own delivery algorithms, because Meta and Google are machine learning end to end and you cannot advertise there otherwise. That is bidding and targeting, not generation. Advantage+ and Performance Max with the generative features switched off. Analytics, forecasting, spreadsheets and dashboards, which is maths rather than generation. Ordinary photo editing: cropping, colour correction, retouching a dust speck, nothing that invents content that was not photographed.

The setting that quietly switches itself back on

This is the practical bit that catches people out. Meta and Google will auto-generate text variants, expand images and swap backgrounds unless you tell them not to. We switch it off per campaign, document it, and re-check it after every platform update, because it has a habit of switching itself back on.

If you do not audit this monthly, you do not actually have a no-generative-AI account. You have one that was true for about six weeks.

Where the creative actually comes from

This is the part people assume is the problem. It is not. Artists have an advantage most e-commerce brands would pay a fortune for, and it is an endless supply of genuinely interesting raw material.

Process footage. The single best-performing format for artist brands is watching the thing get made. Phone on a tripod, timelapse, thirty seconds. It outperforms polished studio work consistently, it is free, and it is the one thing a model cannot fake.

The artist talking. Unscripted, badly lit, one take. Why they made it, who asked, what went wrong halfway through. The rougher it is the more it reads as real, which is exactly the signal this audience scans for.

The archive. Most artists are sitting on years of work, commission briefs, sketchbooks and rejected drafts. That is a creative library. We audit it, catalogue it and build a rotation from it before discussing anything new.

Real customer content. The work on real walls, in real houses, with a real caption from the person who bought it. Ask at delivery, offer nothing more than a repost, and most people say yes.

Copy from the artist's own words. Their captions, their emails, their replies to commission enquiries. The voice already exists. The job is editing, not inventing.

Does it perform worse?

No, and that is the practical argument rather than the ethical one. On the accounts we run this way, human creative built from process footage and the artist's own voice consistently beats slick generic assets.

It stops the scroll differently, because it does not look like an ad. It carries proof: watching a piece get painted is evidence, where a rendered mockup is only a claim. And it fits the buying decision, because nobody buys original art on price. They buy on trust in the maker, so the creative has to build trust in the maker.

Where the rule does cost you

It costs you volume, and I would rather say so than sell it as costless. You cannot spin out forty creative variants a week from a prompt. So the testing plan has to be tighter: fewer concepts, held longer, judged on contribution margin rather than on how many hooks you burned through.

We plan around that from day one. Fewer, better, longer. It also happens to be cheaper.

How we set the account up

Creative audit first: everything already shot, painted, filmed or written gets catalogued into a usable bank. Platform settings lockdown, generative enhancements disabled across Meta and Google, documented, re-audited monthly. Then a capture routine rather than a shoot, which is one timelapse, one talking clip and one finished-piece photo a week. Ten minutes, not a production day.

Profit-first measurement, because originals, prints and commissions have wildly different margins and scaling prints should not quietly cannibalise the commission pipeline. And slow, deliberate testing: fewer concepts, longer read windows, decisions on profit rather than on ROAS theatre.

Who this is for

Artists, illustrators, printmakers, ceramicists, custom commission studios, small handmade brands, and anyone whose audience would be actively offended by a generated image. It works particularly well for people who start on Etsy or the marketplaces and build their own shop from there.

It is also for brands who simply do not want their name attached to it. You do not need an ethical manifesto. "I don't want it, don't use it" is a complete instruction.

If you want help with this

It is not complicated to do badly and it is not complicated to do well. What is costly is finding out six weeks later that the platform quietly re-enabled the thing you switched off, or that your creative library was never built, and you have been running generic assets the whole time.

That work runs through Magic Dolphin, where I run the ads and the measurement side. This is a fixed-scope engagement rather than something I will try to sell you AI for. If you would like someone to audit an account and tell you exactly what is live on it, book a call and say you want the no-generative-AI audit. The full write-up of the method is on the Magic Dolphin blog, and the client results are on this site under Magic Dolphin.