Generating cheap AI images is easy, but pretty renders won’t fix broken listing conversion rates.

StorySell System is a brand-new system designed to transform raw product info, customer reviews, and competitor data into a high-converting customer journey.

Developed from 7+ years of Amazon CRO data across 500+ brands, StorySell locks your messaging strategy before you invest in production. You get a complete blueprint for your main images, A+ content, and video, with ready-to-paste briefs for your design team, agency, or AI tools.

Stop renting your revenue from PPC. Lock your strategy and start converting today at www.storysellsystem.ai

The platforms are writing your ads now. Guess whose name is on the summons.

For most of digital advertising's history, there was a simple arrangement between brands and the platforms they paid. You made the ad. They decided who saw it. Targeting was their job, the message was yours, and everyone politely pretended the attribution numbers were real. It was a functioning relationship, in the way a lot of relationships are functioning.

That arrangement is quietly dissolving. Over the past few weeks, the platforms that matter most to sellers outside Amazon have each moved further towards deciding what your ad says, what it shows and, occasionally, where the click lands. Meanwhile, legal responsibility for that output has been travelling in exactly the opposite direction. Towards us.

This isn't a story about whether AI can make a decent ad. It can, sometimes. It's a story about control, and about who pays when the machine gets creative with your product photos.

On 30 October, OpenAI switched on what it calls "creative enhancements" for self-serve campaigns in ChatGPT Ads. Advertisers got roughly ten days' notice to opt out (which, for anyone who has ever tried to get a brand team to read a platform email, is roughly ten days fewer than required).

If you didn't opt out, the system can now rewrite your headlines and descriptions, resize your images, translate your copy and perform something called "destination-URL optimisation". Translation: it may send the click to a page you didn't choose. Variations can be generated in real time, and none of them need your approval. You are, in the most literal sense, no longer the author of your own ads. More of a muse.

The scale is why this matters. OpenAI says ads now run in 89 countries, with over 100,000 advertisers on the platform, up from around 20,000 in late August. Company figures, so pinch of salt, but the direction is clear: a fast-growing channel with an editing layer switched on by default.

OpenAI is careful to point out that ads are separate from answers. That may well be true. It also sidesteps the more interesting point: the ad itself is now partly an answer, composed on the fly for whatever conversation it lands in.

If ChatGPT's move sounds abstract, Meta is helpfully providing a live demonstration.

Over recent weeks, advertisers and agencies have described pulling live campaigns after discovering AI changes nobody asked for. Bookshop.org found an automated update had enrolled it in a feature that dropped an AI-generated illustration into an ad its own team had made (a bold creative choice for a company whose entire identity is books written by humans). In June, REI found an AI image tool it hadn't knowingly enabled had distorted a vendor's bicycle in an Instagram ad. Agencies also report settings that switch themselves back on after being turned off, which is less a feature than a personality trait.

What changed the tone of all this is a revision to Meta's commercial terms, effective 30 October. Previously, advertisers carried responsibility for AI output when they chose to use Meta's generative tools. The new terms, as reported, quietly drop the "chose" part. They acknowledge the AI may make mistakes, and place responsibility for the results on the business running the ad.

To recap: Meta's AI may get it wrong, Meta may switch it on, and you're liable. It's a bit like a valet scratching your car and handing you the repair bill, with a helpful note that you could have parked it yourself.

Meta's position is that every enhancement can be switched off in Ads Manager. That's accurate. It's also the whole problem. Each enhancement is a separate setting, some are on by default, some live in collapsed "advanced" menus, and some can apply retroactively to ads built before the feature existed. Opting out is still possible. It has just become a recurring audit rather than a one-time decision.

Google's move is more tentative. In mid-October, it began testing ads in AI Mode for users who've connected their Google apps through Personal Intelligence, a feature that launched ad-free. The same insights that personalise AI Mode's answers may now help decide which ads those users see.

The detail matters, because the headline version has run ahead of it. Google says ads won't be based directly on the contents of your Gmail or Google Photos, that advertisers can't access connected apps, that sensitive information is excluded, and that it's opt-in. What's being used is a layer of inference drawn from those apps, not the raw data.

Whether consumers will appreciate that distinction is another matter. Someone who mentions a holiday in an email and then sees a suitcase ad inside an AI answer is unlikely to pause and reflect on the nuance between "your inbox" and "insights derived from your connected apps". Most people's reaction will be shorter, and considerably less printable.

While platforms loosen our grip on creative, legislators are tightening our accountability for it.

Pennsylvania has signed an AI Deceptive Advertising Act requiring clear disclosure when an ad uses AI-generated or significantly modified content in a way that creates a false impression of the product. The disclosure has to appear at first view, in the ad's language, and can't be easy to obscure. There's no small-business exemption and no requirement that the deception was intentional. Fines are capped at $3,000, which is modest. The attorney general's attention tends to be less so.

Lawmakers indicated the core worry was AI making the product look better than it is, rather than the person holding it. Which describes a distorted bicycle and a regenerated background around your product with almost uncomfortable precision.

It's one state, and the law has gaps (chief among them: nobody has properly defined "false impression"). But it sits alongside disclosure rules in New York and California, deepfake laws in Illinois and Minnesota, and EU AI Act transparency obligations that began applying in August. The details vary. The direction doesn't.

Which creates a rather elegant problem. If an ad is generated in real time for a single conversation, how exactly do you put a compliant disclosure on a version you never saw? The platform made it. The law, and now the platform's own terms, say it's yours.

Platforms justify these defaults on performance, with some evidence. One agency test on Meta found a variant with an AI-generated background got twice the spend of the original and produced five purchases against none over a month. Small numbers, but not nothing.

Notice, though, who allocated the spend. When the platform generates the variant and decides which variant gets the budget, it's grading its own exam, against metrics it picked.

Independent research suggests the cost may sit somewhere the dashboard doesn't look. A Syracuse University study with Ipsos showed 3,000 US consumers 20 video ads, half made by people and half generated entirely by AI. Viewers struggled to tell them apart. Yet the AI versions scored 14% lower on short-term sales potential and 17% lower on long-term brand equity. AI did best on conventional, product-led work, and worst where an ad needed humour, emotional specificity or an unexpected idea (so, roughly the bits that make advertising worth watching).

The study tested fully AI-made ads, not platform tweaks, but the logic transfers. A system optimising headlines for click-through won't notice brand distinctiveness slowly leaking away, because that loss never appears in the report it sends you.

The easy framing is advertisers versus platforms. The more useful one is that these defaults risk annoying everyone, including the customer.

For sellers, it's concrete. Approval workflows built over years assume the approved asset is the live asset. That's no longer true. A rewritten headline promising a benefit the product page doesn't mention, or a URL swapped to a page without the offer, creates a gap paid for in wasted clicks and customer service tickets. Regulated categories like supplements and cosmetics carry extra exposure. And for Amazon sellers advertising off-platform, an AI-polished image that drifts from the listing photos sets an expectation the actual box can't meet.

For consumers, the damage is subtler. Quietly improved imagery produces returns and poor reviews, and the reviewer blames the brand, not the platform that touched up the photo. Participants in the Syracuse study were mostly relaxed about brands using AI in ads, but a large majority said it should be disclosed. Default enhancement makes that harder to guarantee.

The platform is the only party whose incentives line up neatly. More variants mean more inventory, more data and a better growth story for investors (OpenAI is reportedly building its ad business ahead of a possible stock market listing, which I'm sure is entirely coincidental). Nothing sinister. It just explains why the defaults point where they do.

The answer isn't to reject AI creative. It handles product-led, conventional formats reasonably well, which describes a fair chunk of ecommerce advertising. The answer is to treat AI enhancement as a decision, not a default:

None of this is free. It's process work smaller sellers rarely have capacity for, which is precisely why defaults carry so much power.

For years, the deal was simple: you made the ad, they ran it. The new deal is that they make the ad, they run it, and you answer for it.

The bigger questions don't fit in a settings menu. If platforms generate the ad and the law holds the advertiser liable, does the advertiser's job shift from creator to auditor, and is the channel priced for that? Will regulators eventually decide that a platform rewriting an ad in real time is, in any meaningful sense, its publisher? And if customers start doubting that what they see is what arrives in the box, whose credibility goes first: the platform that made the change, or the brand whose name is on it?

Jo

P.S. One report puts Meta's settings review deadline at 20 November. Put it in your calendar now, ideally somewhere the AI can't reschedule it.

P.P.S. If your ads start describing your products as "revolutionary", "game-changing" and "a must-have for 2026", that wasn't your copywriter. Probably.

🌐 Code Design - Build your website by writing a simple text prompt and host it in minutes using AI.

🤖 Shortwave AI Assistant on iOS - Write, search, and analyze your emails with AI on the go.

🕺🏽 Alta - Your personal AI stylist that truly gets you - from the clothes in your closet to the looks you love.

🩻 Glif - Remix any image on the web.

🧱 Blocks - An AI-powered email builder that will help you make stunning emails in minutes!

😍It was great, all I needed

🙂Good, not great

🙄It sucked