Google’s AI Finally Has to Show Its Homework
On October 8, Google did the most honest thing an ad platform has done all year: it showed its homework. Buried in the Asset optimization section of Google Ads, advertisers can now generate up to 10 sample headlines and descriptions from AI Max’s text customization engine without turning the feature on. Enter a final URL. Wait about 30 seconds. Read what Google’s AI would say about your brand if you let it loose. (Search Engine Roundtable, Oct 9, 2026)
The real story is not the preview button. It is what the preview reveals. Every sample is generated from your landing page, your existing keywords, and your current ad copy. Which means the preview is a mirror. And most brands are about to learn that the reason Google’s AI writes boring ads about them is that their own website is boring.
What it is
Text customization asset previews. The feature lives inside Asset optimization, both when you build a new Search campaign and when you edit a live one. Provide a final URL and Google AI automatically generates a representative sample of up to 10 example headlines and descriptions. No opt-in to text customization required. Google says examples land in under 30 seconds and work in every language text customization supports: English, Spanish, Portuguese, Italian, French, German, Japanese, and Dutch. Most business categories qualify. Adult content does not. (Search Engine Roundtable, Oct 9, 2026)
Two details matter. First, these are examples, not the exact assets that will serve. A taste of the machine’s phrasing, not a contract. Second, the preview respects your text guidelines, the rules you set to exclude certain terms or themes. Set the guidelines too tight and block all output, and Google prompts you to loosen them. So the preview also tests whether your guardrails strangle the machine or steer it.
What changed
Context makes this feature sting. On September 1, Google started automatically upgrading Search campaigns that used campaign-level broad match or automatically created assets into AI Max. There was no opt-in step. The only escape hatch was switching those features off before the deadline, and the deadline passed. Google notified affected advertisers by email on August 5, signed by the Google Ads Team, with no blog post and no announcement. If it landed in a shared inbox nobody checks, nobody at your business saw it. (NewsXtend, Sept 2026)
Here is the part that should make your stomach drop. Campaigns using Automatically Created Assets were upgraded with search term matching AND text customization switched on. Read that again. Google’s AI has been writing your ad copy for over a month, and the preview button to see what it might write arrived this week. The blindfold came off after the race started.
Common Thread Collective called the missing preview “the single most cited concern from advertisers reluctant to let AI generate their copy.” Their analysis, published the same day as the rollout, told ecommerce brands to check the samples for product accuracy and message consistency before Q4 ramps up. (WebProNews, Oct 8, 2026)
What works
The preview is useful as a diagnostic in a way Google probably did not intend. Strong samples reference specific product names, unique selling points, or pricing pulled straight from the site. Weak samples feel generic. Or worse, they stretch claims your brand would never approve.
Flip that around. When the samples come back vague, the problem is not the model. It is your raw material. Google’s AI can only remix what your page gives it. If your landing page says “innovative solutions for modern businesses,” congratulations, your AI-generated ads will say “innovative solutions for modern businesses” too, because that is all you told the machine. The preview is a free brand-voice audit, and most companies are about to fail it.
Use it as one. Run the preview on your five highest-spend landing pages this week. Grade the samples the way you would grade a junior copywriter’s first draft. Where the copy is sharp, your page is doing the work. Where it is mush, rewrite the page, not just the ads. Every hour you spend sharpening the landing page pays twice: once in conversion rate, once in every AI-generated asset Google builds from it going forward.
The practical value is real too. Two minutes per campaign to generate and review samples beats the old loop of launch, monitor, panic, pause, rewrite. Generated assets show up marked in reports, they do not count against your responsive search ad limits, and you can remove individual examples (bulk editing stays limited). And the upside case is not imaginary: Google’s own data says campaigns running the full AI Max combo, search term matching plus text customization plus final URL expansion, average 7% more conversions or conversion value at similar cost per acquisition or return on ad spend compared with search term matching alone. (WebProNews, Oct 8, 2026)
What breaks or stays fenced
Now the fine print, because there is always fine print. The preview shows examples, not the exact assets that will serve. The system keeps generating new variations over time as user behavior evolves. So the preview is a snapshot, not a contract. The copy you approve on Monday is not necessarily the copy running on Friday.
The performance record is mixed, and you should know the other side. A July analysis by Brad Geddes found that AI-generated copy often lifted click-through rates while conversion rates sometimes fell, particularly in B2B scenarios where nuanced audience prequalification matters. The machine is great at writing clickable ads. It is less great at writing ads that filter out the wrong clicks, which is half the job in considered purchases. Messaging restrictions helped steer outcomes, but demanded significant upfront work. (WebProNews, Oct 8, 2026)
Then there is the compliance angle. A financial services firm could watch the preview suggest rate claims that cross regulatory lines. A pharma brand could see efficacy language it would never clear with legal. The 30-second check gives you time to adjust landing pages, tighten guidelines, or simply decide against enabling the feature. But it only helps if someone actually reads the samples, and “someone reads the AI’s homework” is a process most teams have not built yet.
And keep perspective on what the preview does not do. It does not show you what the AI will write next month. It does not explain why one variation wins. It does not fix the underlying black box. Google shipped companion transparency this same week: new reporting columns for brand inclusions, locations of interest, and optimized targeting, plus AI-generated experiment summaries, all covered in Anu Adegbola’s October 2-8 roundup for Search Engine Land. A unified journey report connecting query, creative shown, and landing page is slated for later this year. (Anu Adegbola, Oct 2026) The direction is clear. The box is opening, one small window at a time.
Who it is for
Three teams should care most. First, ecommerce brands heading into Q4: check every product category page the AI will draw from, because holiday volume multiplies every weak sample into real spend. Second, anyone who was auto-migrated in September and has been running text customization without ever reading what the machine writes: you have a month of AI-written copy in your account and this is your first chance to audit the raw material. Third, B2B marketers: the Geddes finding is your warning label. If your sale needs prequalification, the preview is where you catch the machine sanding off the edges that keep tire-kickers out.
What to do this week
Do not just toggle the preview on and nod. Make it a ritual.
One, run asset previews on your top five landing pages by spend. Screenshot the samples. You now have a before picture.
Two, grade them brutally. Circle every sample that could belong to a competitor. Those are the sentences your landing page failed to prevent.
Three, set your text guidelines BEFORE you preview again. Exclude the terms, themes, and claims you never want near your brand. Then run the preview and confirm the machine respected them.
Four, fix the page, not just the ad. The strongest lever in AI Max is the landing page itself, because the page is the prompt. Specific product names, real prices, concrete differentiators. Feed the machine better raw material and every future variation gets better.
Five, open the ad-level asset report and look at what is already serving. The preview is new. The AI writing your copy is not. Find out what it has been saying about you for the last five weeks.
Google finally let you read the first draft before the machine spends your money. The uncomfortable part is what the draft says about you. Your landing page is the prompt. The preview just made that literal.
