ChatGPT will rewrite your ads by default. Here's how a CPG brand should decide

On October 20, OpenAI emailed advertisers to say that a feature called creative enhancements would be switched on by default for new and existing ChatGPT ad campaigns. Search Engine Land and PPC Land both covered it the next day. Once it's on, ChatGPT can write new versions of your headline and description to fit the conversation the ad appears in, and those versions run without anyone at your company seeing them first.

For most advertisers this is a performance question. For a CPG brand that sells supplements, functional food and drinks, baby products, skincare or cleaning products, it's a claims question. This piece covers what the feature does, the dates, and a simple way to decide campaign by campaign.

What creative enhancements can change

OpenAI's help page lists four kinds of change, and three inputs it uses to make them: the context of the conversation or the user's interests, your existing ad, and content from your landing page.

What changes	What OpenAI says it may do	Why a CPG brand should care
Headlines and descriptions	Create new or updated text to connect your offering to the conversation or user interests	New wording is where implied claims creep in
Language	Translate your ad text into another language	Claims and required qualifiers have to survive translation
Images	Resize images to fit the ad format	Low risk, unless cropping cuts off a disclaimer on the image
Destination URL	Send people to a different relevant page on your website	A shopper could land on a page you didn't pick for that audience

Two lines on the page matter most. Variations "may be prepared in advance or created in real time based on the conversation," and when the feature is enabled, "these ads can run automatically without requiring you to review or approve each version." In its FAQ, OpenAI adds that "the experience does not provide a preview of every variation."

The help page's own example is a grocery delivery ad. "Fresh groceries delivered" becomes "Groceries for quick weeknight dinners" during a conversation about weeknight meals. That's a harmless change, and it shows why the feature exists: the rewrite is more relevant than the original. The question is what happens when the product is a probiotic and the conversation is about bloating.

The dates, and why they're fuzzy

OpenAI's email gave October 30. When we read the help page this weekend, it said creative enhancements would be enabled automatically starting October 31, and PPC Land reported that the page had earlier shown October 29. Treat the deadline as this week and act before Thursday.

The default applies to new and existing campaigns. Opt-outs you've already set at the account or campaign level are kept, and OpenAI says campaign-specific settings take priority over the account default. The control is in Ads Manager under Manage settings.

There's one toggle. Neither the help page nor PPC Land's coverage describes a way to allow some kinds of change (translation, say) and block others (text rewrites). It's all four or none, per campaign.

Date	What happened
January 16, 2026	OpenAI announces it will test ads in ChatGPT
February 9, 2026	US ads pilot begins
May 5, 2026	Ads Manager opens to all US businesses
October 20, 2026	OpenAI emails advertisers about creative enhancements becoming the default
End of October 2026	Creative enhancements switch on for campaigns that haven't opted out

The scale is no longer small. PPC Land reported that OpenAI said on October 21 that ChatGPT ads are available in 89 countries and that more than 100,000 advertisers have used them. And shoppers are using assistants: Salesforce's 2026 holiday predictions say 21% of shoppers have used a third-party AI chat assistant while shopping.

Why this is a claims problem, not a copy problem

The Federal Trade Commission's Health Products Compliance Guidance is direct about three things that apply here.

First, you're responsible for what an ad implies, not only what it says. "Under FTC law, a marketer is equally responsible for the accuracy of claims suggested or reasonably implied in advertising," the guidance says. "Marketers can't suggest benefits, safety, or other characteristics about their product indirectly that they couldn't claim directly."

Second, the FTC judges the whole ad, not phrases in isolation. Marketers should assess "the 'net impression' conveyed by all elements of the ad," and the determination is "consumer-driven," based on what reasonable consumers take from it, not on what the marketer intended.

Third, responsibility is broad. "Anyone participating in deceptive marketing is potentially liable under FTC law," and the guidance lists ad agencies, distributors and retailers among those the FTC has acted against. It doesn't mention AI-written ads, and we won't speculate on how the FTC would treat one. But nothing in the guidance suggests "the platform wrote it" moves responsibility off the brand whose product is being advertised.

The FTC has settled or adjudicated more than 200 cases about health product advertising since 1998, according to the same document, and it says the principles apply to foods as well as supplements. Its examples include a children's drink ad whose imagery implied protection from colds, and an herbal supplement whose name and pictures implied cold prevention without saying it.

Now picture a rewrite engine whose whole job is to make your ad fit the conversation. Someone asks ChatGPT about bloating after meals. Your approved headline says your probiotic "supports digestive health." A version that connects that line to the conversation could easily drift toward relieving bloating, which is a different claim with a different evidence bar. OpenAI says enhancements use your existing ad and landing page, and its example "assumes the original ad and landing page support the product and delivery details shown." That's a sensible design. It isn't a claims review.

A campaign-by-campaign rule

Most CPG brands don't need a long policy. They need a quick sort of their campaigns into three groups. The table below is our suggested starting point, not legal advice, and your regulatory counsel should set the final lines.

If the campaign's ad or landing page...	Suggested setting	Examples
Makes a health, structure/function, nutrition content, safety or performance claim	Off	Probiotics, protein, "clinically tested," "dermatologist recommended," "kills 99.9% of germs," "high in fiber"
Targets children, pregnancy, allergies or a medical condition, even without a claim	Off	Baby food, allergen-free snacks, products for sensitive skin
Sells on taste, price, convenience, bundles or availability only	On, with monitoring	Variety packs, a seasonal flavor, free shipping, a new retailer listing
Runs in markets where you haven't reviewed translated claims	Off until you have	Any campaign outside your home market

Three practical notes. Set the account default to off if most of your catalog falls into the first two rows, then switch it on for specific campaigns; that way a new campaign starts safe. Check the landing page as well as the ad, because the landing page is an input, and a claim buried in a product page FAQ can surface in a rewritten headline. And remember the URL change: if a campaign is on, make sure every page the system might pick on your site is one you'd be comfortable sending a stranger to.

Tip 35 has the short version of this, with the four clicks to make in Ads Manager.

If you leave it on, watch it

OpenAI's help page points advertisers to Ads Manager reporting at the campaign, ad group and ad level: impressions, clicks, spend, CTR and attributed conversions if you've set up measurement. It also says plainly that "changes over time do not, on their own, establish the effect of creative enhancements." So don't read a week of better click-through rate as proof the rewrites work.

The bigger gap is that you can't see every version. For campaigns left on, do three things:

1. Keep your source ad and landing page tight. The help page asks for current prices, promotions, product details and availability. Remove or qualify anything on the landing page you wouldn't want quoted in a headline.
2. Ask ChatGPT, from a test account, the kinds of questions where your ad might appear, and note any versions you see. It won't be complete, but it's a sample.
3. Put a review date in the calendar. OpenAI has changed the dates on this page within a week, and the controls may change too.

What it doesn't change

OpenAI says ads don't influence ChatGPT's answers: "Ads are separate and clearly labeled." That means the setting has nothing to do with whether ChatGPT recommends your product in its own answer. That depends on the same things it always has: clear product facts on your site and in feeds, consistency across retailers (tip 5), and third-party proof such as certifications (tip 21). If you're also selling through ChatGPT's shopping results, our feed piece covers that side, and our holiday playbook puts it in the season's calendar.

It's worth keeping the two apart when you check your visibility, too. A paid placement and an organic mention look different in the interface, and they should be tracked separately, whether by hand or with a tool such as SearchPanel from XENA Intelligence, which asks assistants shopper-style questions and keeps the answers.

The short version

Creative enhancements are a reasonable idea for a lot of advertisers, and for a sampler-pack promotion they may well help. For any CPG ad that carries a claim, a system that rewrites your words after your legal review is a system that undoes your legal review. Make the call per campaign, before the end of October, and write down why.