Google Ads New Terms of Service: Who Owns the Risk When AI Writes Your Ads
What changed, and the one sentence to remember
As of the July 2026 update, Google rolled a new version of its Terms of Service onto every Google Ads account. There was no popup asking you to accept clause by clause. It just took effect, which is why a lot of sellers never registered that anything happened at all.
The one sentence worth memorizing: if AI generated it, you still own it. The revised terms state more plainly than before that advertisers stay responsible for reviewing, approving, editing, and removing their campaigns and ad assets, and they specifically pull in assets that Google’s own tools generate automatically. The old mental escape hatch of “I did not write that headline, the system did, so it is not on me” no longer holds.
One caveat up front: everything here is general information, not legal advice. My confidence on the specific legal interpretation is Low. Whether a clause reads a certain way, and whether your country layers its own advertising statutes on top, is a question for counsel who can read the actual terms for your jurisdiction. Trade press covering the update also mentioned changes touching arbitration and how liability gets allocated, but I am not going to parse those clauses for you, because getting them slightly wrong would be worse than useless.
To see why this clause suddenly matters, count how much of your account is now machine-written. AI Max handles text customization, Final URL Expansion (auto-rerouting where clicks land), and automatic format selection. Performance Max assembles its own image and text assets. Gemini reads your Merchant Center feed and writes a fresh product explainer for individual queries in real time. Across all of that, plenty of sentences, images, and landing-page pairings go live that you personally never laid eyes on. The terms now say: unseen still counts as yours.
Where the risk actually shows up
For e-commerce, the sharpest edge is inflated or regulated claims. To lift click-through, the system might take a plain sentence from your product page and rewrite it into something more absolute. Health, weight-loss, and skincare-efficacy categories are the danger zone. A machine-written “results in 7 days” or anything implying a medical effect is a non-compliant ad in many markets. Pricing and free shipping are just as exposed: you offer free shipping over a threshold, the AI writes an unconditional “free shipping on everything,” and now you have a false claim with your name on it.
The second bucket is trademark and competitor terms. Automated keyword expansion and copy generation sometimes pull a rival’s brand name into a headline or a match. You did not type it, but it ran in your ad, and the cease-and-desist letter will not stop to ask who wrote it.
Third is region-specific ad law. The EU, the UK, and Germany each set different rules on ad wording, comparative claims, and price display, and German competition law is especially strict. One AI-generated English asset can be fine in the US and cross a line in Germany. The system optimizes for conversions; it does not know which legal regime governs the market you are serving.
Fourth, and the quietest one, is Final URL Expansion. AI Max may decide a different page on your site converts a given searcher better and quietly send the ad traffic there. The trouble is that page might be an unpublished test page, an out-of-stock product, or a stale promo page with the wrong price. The landing experience stops matching the ad’s promise, which wastes budget at best and reads as misleading at worst.
Building the review gates
Do not respond by switching automation off. That throws away the performance Google has spent years tuning. The move is to put gates around automation so it runs while you can still see and stop what it produces. Start by sorting assets by who generates them and what you owe each one:
| Auto-generated asset | Generated by | Your review action |
|---|---|---|
| Assembled headline/description combos | AI Max text customization | Spot-check served combinations; add absolute and efficacy phrasing to a banned-terms list |
| Final URL Expansion destinations | AI Max | Pull the actual landing-page URL report weekly; check for out-of-stock, test, or mispriced pages |
| Auto-selected ad formats | AI Max | Review live renders so a required disclaimer never gets truncated |
| PMax image and text assets | Performance Max | Approve before launch; give regulated categories a separate image pass |
| Per-query product explainers | Gemini reading the feed | Fix it at the source by cleaning descriptions in Merchant Center |
The landing-page row deserves the most attention. Final URL Expansion is on by default, so go into campaign settings, find it, and manually exclude the paths that should never receive traffic, such as /test, /draft, and out-of-stock collections. When you audit, do not eyeball the homepage. Pull the report by landing page and see where clicks are really going.
Fix the cadence so you are not only looking after something breaks. Run one gate at launch, whenever new assets or copy go up. Run a second gate weekly, pulling the Final URL Expansion report and the disapproval log. Raise the bar for regulated categories (supplements, cosmetic-efficacy, anything finance-adjacent): those assets get human eyes before launch even when a machine wrote them.
Keep the claims and brand-safety check as a short list you run against every new campaign:
- No absolute or efficacy language (cure, permanent, 100%, guaranteed)
- “Free shipping/returns” is genuinely unconditional, or the threshold is stated
- Price, discount, and stock match the current product page
- No competitor trademarks in headlines or match terms
- Final URL Expansion excludes out-of-stock, test, and mispriced pages
- Target-market ad rules confirmed separately (especially EU/UK/Germany)
Keeping the upside while capping the downside
Reviewing does not mean pushing every asset past a human. Ten people could not keep up, and you would erase the speed that made automation worth it. The lever is sampling, not full inspection. Layer it by category and risk: sample heavily, near full coverage, for high-risk lines (regulated, high ticket, dispute-prone), and skim maybe ten percent of low-risk everyday goods. When the sample shows the system favoring one type of mistake, encode that into a rule instead of catching it by eye every time.
Three tools do the containing. Grow a negative-keyword list continuously, adding terms that pull irrelevant traffic or trigger risky matches. Keep the URL exclusion list, as above, so wrong pages never receive clicks. Third, monitor disapprovals. Every time Google rejects an ad it hands you a reason, and those reasons are free compliance signals rotting in the back office if nobody reads them. Block time weekly to scan the disapproval log; the same reason recurring means a class of your auto-generated assets is systematically hitting the same rule, and that gets fixed at the source.
One thing sellers skip: leave a paper trail. Note when you excluded a URL and when you added a term to the banned list. If a compliance dispute ever lands, “we run active review and fixed the issue the day we found it” puts you somewhere very different from “we handed it to the system and checked nothing.” The terms assigned you the responsibility, so records that prove you met your review duty are worth keeping.
If you sell into more than one country
One compliance bar cannot cover every market. The same line of copy and the same image can clear in the US, break German competition law, run into UK advertising standards, and miss a required price or environmental label in the EU. Do not expect a single review standard to protect every placement.
The practical fix is to split your review standard by market. At minimum, separate the lenient markets from the strict ones, and give the strict group (Germany, France, the EU broadly) its own tighter banned-terms list and higher sampling rate rather than sharing the US setup. AI-generated multilingual copy needs extra care: a machine-translated claim can be literally accurate and still land squarely on a local advertising prohibition you cannot see because you do not read the language. For those markets, a reviewer who knows the local language and rules is worth far more than the labor cost you would save skipping them.
To repeat the disclaimer: this is operational guidance, not legal advice, and my confidence on specific statutes and liability is Low. Before you go live, have local counsel review both the actual terms and your live assets.
Related Articles
Meta Pixel AI Enrichment and One-Click CAPI: What DTC Brands Need to Do Now
Meta launched AI-enhanced Pixel and a free one-click Conversions API in April 2026. Here is what changed, what to watch for, and how to set it up for your store.
TikTok Shop Creator Picks: How AI Affiliate Matching Changes the Seller Playbook
Creator Picks is an AI recommendation engine inside TikTok Shop Seller Center that matches products with creators based on audience purchase behavior and category conversion data rather than follower count. This guide covers how to evaluate AI-recommended creators, set targeted commission plans, manage sample budgets, and guard against low-quality AI-generated affiliate content.
AI Ad Creative Tools Compared 2026: Meta Advantage+ Creative vs Google Asset Studio vs TikTok Symphony
Three platforms take different technical approaches to AI creative: Meta uses conversational signals to auto-generate variants, Google Asset Studio generates video and image assets with Veo/Imagen, and TikTok Symphony repurposes existing content through CapCut integration. Which one fits your workflow depends on your primary platform and where your creative production breaks down.