Google Ads headlines now come with a preview, and a review step you should not skip

What the preview shows, and the google ads headlines it cannot

Automated google ads headlines have always asked advertisers for a leap of faith. This week Google handed over a torch. Its help page now says that when you give a final landing page, Google AI automatically generates a representative sample of up to 10 example assets of headlines and descriptions. You do not have to switch text customization on to see them. For anyone who has had to explain a machine-written headline to a compliance team after the fact, that is a real change of order: review first, opt in second.

Trade coverage dates the rollout to 8 and 9 October 2026. MediaPost says the pre-activation preview tool allows advertisers to generate and review up to 10 sample headlines and descriptions, while Search Engine Roundtable adds that the preview panel automatically filters the generated assets to meet your guidelines. Your own rules shape what you are shown, which makes the preview a test of your guidelines as much as a test of the model.

Two details in that chart repay a second look. The first is that the preview is a sample, not a rehearsal. ContentGrip is blunt about it: it is still a sample, rather than a complete view of every asset or combination that a live campaign will serve. The second is that generated assets are separate from your own. Google's page says customized text assets don't contribute towards the 15 headlines and 4 description lines for responsive search ads. Your google ads headlines and Google's can sit side by side in one ad.

That second point is the one that surprises people. A set of google ads headlines was a closed set that you wrote. With text customization it becomes an open set that you wrote and a model extended. The advertiser still owns the outcome, and the preview is the first honest look at what that extended set might say.

Why ai ad copy is a brand risk before it is a lever

Performance teams will read the preview as a lever. Compliance teams should read it as a risk register. Machine-written google ads headlines can invent a benefit, a price, a guarantee or a qualification, and an advertiser cannot hand that to the machine. The UK's advertising regulator is explicit on the point: if an AI tool produces something problematic, the advertiser remains responsible for any output used in their ads. That is not a footnote. It is the rule.

Google's own policy points the same way. Its misrepresentation policy prohibits making inaccurate claims or claims that entice the user with an improbable result, and generated google ads headlines are held to it exactly like handwritten ones. In regulated categories the exposure is sharper still, because a single invented claim about a treatment, a return or a bonus can trigger a policy strike, a regulator's letter or both.

The wider legal backdrop is shifting, too. The European Commission says the AI Act entered into force on 1 August 2024 and became applicable on 2 August 2026, with some exceptions. An advertiser's use of ad tooling is not the same as a high-risk deployment, and nothing on that page says the preview feature is caught. It does mean transparency expectations around generated content are rising, which is why logging what you reviewed matters.

The study is worth citing precisely because it is narrow. The authors report that the generated advertisements received approximately the same CTR as human-authored ads. That settles whether a model can write a clickable line. It leaves open whether it can write a lawful one, which is the only question a regulated advertiser is paid to ask.

Text guidelines: the 25-term fence around the machine

The preview only matters if you have something to preview against. Google's text guidelines are the fence. The help page says you can add up to 25 term exclusions per campaign to refine your ad, and Google's own February announcement framed the feature in plain terms: you can now steer Google AI by defining specific terms to exclude or concepts to avoid. The MediaPost write-up adds that those inputs can be expressed in the user's own words.

Twenty-five is a lot until you list the claims a regulated brand cannot make in google ads headlines. Cure, guaranteed, risk-free, best, number one, free bet, instant approval: that is seven before you reach the product names, the competitor names and the jurisdiction-specific phrases. A fence with room for twenty-five posts rewards a short, ruthless list over a long, hopeful one.

The upgrade clock matters here. Google's April post says starting in September, legacy features like Dynamic Search Ads will automatically upgrade to AI Max, and the page was updated in June. If your account carries legacy automation, the guidelines are not an optional extra. They are the control that decides whether the upgrade is a convenience or a surprise. Our earlier notes on the AI Max auto-upgrade and the migration audit cover the mechanics.

Finally, find the label. The ad-level report marks generated assets: Google customized text assets can be identified through the designation Google AI in the Added by column. That one filter is your audit trail. If a regulator or a client asks what ran, it is the difference between a screenshot and a shrug.

Reading a preview the way a reviewer would

A good review of google ads headlines is a short, repeatable ritual, not a sniff test. Read each sample for four things: a claim you cannot prove, a promise you cannot keep, a word on your banned list and a phrase that sounds like nobody at your company. The last one is the one the scent of the fox finds first. Voice drift is not illegal, but it is a tell.

The third row deserves its own sentence. A bad preview is often a page problem in disguise, because the sample is generated from your landing page. One practitioner on X put it in six words: treat a bad preview as a page problem first. It is a good rule. If the machine cannot find a safe sentence on your page, it is not going to invent one that is both safe and persuasive.

Adalysis, a PPC tooling firm, gives the iterative version of the same advice: if the output contains assets you don't like, then you'll need to refine your guidelines. Notice the verb. Refine. The preview is not a one-off gate you clear and forget. It is a loop you run whenever the page, the offer or the rules change.

A pre-flight for the ai max text customization switch

Before anyone flips ai max text customization on for a live campaign, run a pre-flight. It takes an afternoon, it needs a checklist and it should end with a name against it, because the feature is exactly the kind of change that nobody owns until something goes wrong. The preview has made a pre-flight for google ads headlines possible. It has not made it automatic.

This is also where an AI ad copy generator mindset goes wrong. A tool that writes copy on demand invites volume. A preview that you review invites judgement. The two are not the same job, and the advertisers who confuse them will find out in the account's policy centre. For ai generated headlines, the question is never how many. It is which ones you would put your name to.

For regulated advertisers the stakes are not abstract. A healthcare clinic, an iGaming operator or a fintech lender cannot treat google ads headlines as disposable. That discipline sits inside our PPC management, and the preview makes it cheaper to practise. If you want it set up, start with a conversation. The fox does not trust the first path that looks clear, and neither should an account manager at a moonlit deadline.

Frequently asked questions