Do not start by asking for more headlines. The more consequential change is that ChatGPT Ads is about to switch on a set of runtime creative controls by default. The system may revise headlines and descriptions, translate copy, resize images, and optimize destination URLs. Some variations can be created in real time, and advertisers will not receive a preview of every version before it runs.

OpenAI’s October 22, 2026 update to “Creative enhancements for ChatGPT Ads” says the feature will be enabled automatically for both new and existing campaigns starting October 29. Advertisers can opt out before that date or change the setting later. Existing account- and campaign-level opt-outs remain in place, while a campaign-specific setting takes priority over the account default.

That inheritance rule turns a creative feature into an operating-control issue. A brand now needs to know which accounts are default-on, which campaigns override them, which ads and landing pages become source material, who may opt out, and who checks the setting after an agency or market handover. The delivered output can change without a new production order.

Default-on is not a universal recommendation

The official documentation establishes what OpenAI currently offers. It does not establish that creative enhancements improve every campaign, nor that the advertising industry has completed the same transition. OpenAI also cautions that Ads Manager metrics such as impressions, clicks, spend, CTR and attributed conversions describe overall performance. A change over time does not, by itself, isolate the effect of creative enhancements.

October 29 should therefore be treated as a configuration-audit date, not an instruction to scale. Start by classifying four kinds of change:

ChangePotential efficiencyControl required before launch
Headline and description editsFewer manually written variantsLock numbers, comparisons, performance claims, prices and qualifiers
TranslationWider language coverageMap market-specific claims, warranties, promotions and product versions
Image resizingMore format coverageKeep the product, pack text, demonstration and required disclosure inside a safe composition
Destination-URL optimizationA more relevant post-click pathAllowlist eligible pages, keep price and availability current, preserve tracking

If a team can approve a master ad but cannot assign ownership for these four change classes, opting out temporarily is more controlled than leaving the switch on “to see what happens.”

The ONCE SWITCH handover

Our earlier executable brand-guidelines framework focuses on trustworthy inputs, invariants and launch controls. Default-on behavior introduces a different upstream question: where is automation active before anyone reviews an output? ONCE uses a six-part SWITCH handover to make that scope auditable.

S — Scope

Record the account default, every campaign override, audit date, operator, and a screenshot or export of the setting. “The account opted out” is not sufficient if a campaign-level choice can supersede it. New, copied and agency-migrated campaigns should be checked again rather than inheriting a verbal assumption.

W — Working sources

List the original ads, landing pages, SKUs, markets and languages that the system can draw from. OpenAI says enhancements may use the existing ad and landing-page content. An expired promotion, discontinued model, stale availability statement or page that combines several markets can therefore become a bad source. Page ownership and refresh frequency belong in the media handover.

I — Invariants

Separate facts that must not drift: numerical specifications, price and validity dates, performance or comparative claim conditions, certifications, warranty territory, intended users, pack quantity and mandatory qualifiers. This is not another brand-voice paragraph. It is a product-truth register that creative and media teams can actually check.

T — Test baseline

Keep a stable, enhancement-off baseline and predefine the question, window, market, budget and decision rule. Do not change bidding, audience, landing page and enhancement settings at the same time and then attribute an overall CTR movement to the creative feature. OpenAI’s own measurement note says the overall trend is not causal evidence.

C — Containment

Create an allowlist of destination pages, define the markets eligible for translation, provide crop-safe source images, and set stop conditions for price or stock errors, complaints, rejection patterns and unexpected redirects. The goal is not to predict every failure. It is to narrow the blast radius and restore a known approved baseline quickly.

H — Handover owner

Assign an actual role—or named person—across the brand, agency, production partner, web team and legal reviewers. The handover should contain the current switch state, campaign exceptions, source pages, invariants, baseline version, last audit, next review and authority to pause. During an agency transition, this record matters more than an email saying “creative approved.”

Buy safe coverage units, not a count of ads

Automation may reduce manual exports while increasing source audits, configuration checks, translation rules, page QA, baseline design and rollback work. A quote based only on headline or image counts hides that labor.

A practical internal denominator is:

Fully loaded cost per safe coverage unit = (source audit + invariant mapping + configuration and baseline design + monitoring and rollback) ÷ approved market × offer group × campaign combinations.

This is not an industry benchmark. Its purpose is to show where scale is real. Products that share stable claims, page structure and market rules can reuse control work. Fast-moving promotions, materially different SKUs or jurisdiction-specific qualifiers should not be merged just to improve the denominator.

The production scope should include crop-safe masters, do-not-alter regions, product-fact mapping, translatable and non-translatable copy, an eligible-page list and a recovery baseline—not only finished images and videos. The media or marketing-technology scope should include setting records, override relationships, test design, monitoring and the actual opt-out action. Those responsibilities should not disappear inside a single “managed service” line item.

Keep enhancements on—or opt out?

A low-risk pilot is more reasonable when product and price facts are stable; landing pages have clear owners; the offer does not depend on complex legal qualifiers; translation is limited to reviewed markets; imagery can tolerate resizing; and a team retains an enhancement-off baseline with someone monitoring URLs and anomalies.

Opt out first if any one of these conditions applies: legal requires exact-copy preapproval; pages mix markets, models or promotions; price and availability change faster than the site is maintained; translation can change a material qualifier; products or pack information cannot survive a crop; several agencies manage campaigns without a shared setting register; or nobody can pause and restore the campaign quickly.

Opting out is not a rejection of automation. It is a judgment that the current inputs and accountability structure are not ready. OpenAI allows the setting to be changed later.

The real definition of “final” has changed

Traditional production treats an approved image or signed-off paragraph as the final. With default-on creative enhancements, the final operating object also includes the setting, source content, override hierarchy and rollback authority. Without those elements, the brand has approved only the raw material that a platform may continue to adapt.

For international brands working with Chinese product and production teams, the useful next move is not another batch of variants. Complete one SWITCH handover: inspect every account and campaign, treat landing pages as creative inputs, freeze product invariants, retain a clean baseline, and give one owner the authority to opt out. Automation can then reduce repetitive production without quietly replacing brand judgment with an inherited default.

Sources and scope

The SWITCH handover, cost equation and enablement criteria are ONCE’s analysis. They are not OpenAI requirements, legal advice or a performance guarantee.