Ads in AI Search: Announced Formats and Tests to Run Now

Ads are moving deeper into AI assisted research, where a person may describe a problem, compare options, and refine constraints before clicking. That changes the context around an ad: the placement may need to answer a specific question or explain a product's fit, not merely match a short keyword. The trend is real, but the operational details require restraint. Google has announced tests, pilots, launches, and coming formats with different scopes. Those words are not interchangeable with global availability. Marketers should separate what Google has actually announced from what they choose to test in their own accounts. What Google has actually announced On May 20, 2026, Google said it was testing two ad formats in AI Mode: Conversational Discovery ads and Highlighted Answers. Google described an independent AI explainer that evaluates product or service information and appears alongside advertiser creative; the formats remain labeled Sponsored. Conversational Discovery is intended to tailor creative to a specific question. Highlighted Answers makes eligible ads part of a recommendation list when Google considers them relevant and high quality. The same announcement said AI powered Shopping ads and Business Agent for Leads were coming to Search in the following months. It also described an expansion of the Direct Offers pilot, including more promotion types, native checkout for eligible Universal Commerce Protocol merchants, and planned travel offers. These are announced product facts, documented in A new generation of ads for the AI era of Search https://blog.google/products/ads commerce/google marketing live search ads/ . The boundaries matter. Google's language includes testing, pilot, coming soon, and launching. The announcement does not establish that every advertiser, market, query, or account can use every format on demand. It also does not promise incremental profit. Treat account eligibility, reporting, controls, and availability as facts to verify in the live Google Ads interface or with an authorized account representative. The strategic shift behind the formats Classic paid search often compresses intent into a keyword, an ad, and a landing page. Conversational search can expose more context: intended use, constraints, desired features, uncertainty, and follow up questions. An ad may therefore be selected and explained against a richer request. That raises the value of accurate machine readable inputs. Google's announcement tells advertisers to build a foundation with AI Max for Search, AI Max for Shopping, and Performance Max. The AI Max for Search documentation https://support.google.com/google ads/answer/15910187?hl=en describes search term matching and asset optimization as core features, while preserving settings and controls that advertisers should review before activation. For retailers, product data is part of the creative system. Google's Product data specification https://support.google.com/merchants/answer/7052112?hl=en requires fields such as product links and main images, and requires submitted availability and price to match the landing page, checkout, and relevant structured data. A fluent AI generated explanation cannot repair a stale price, the wrong variant image, or an unavailable item. Recommended experiment 1: repair the evidence layer This section is a recommendation, not a Google announced requirement for access to every AI ad format. Choose one commercially important category and audit the data that could support a tailored answer. For each item or service, check naming, identifiers, variant attributes, dimensions, materials, compatibility, price, availability, shipping, returns, and claims. Compare the feed, ad assets, structured data, visible landing page, and checkout. Record every disagreement and assign an owner. For products, use clear main images that accurately show the item. Add useful secondary views for scale, use, controls, texture, or included components. Do not assume more attributes are better if they are unverified. The experiment is successful when the selected category has complete, consistent, current evidence across surfaces, not when a team merely fills every optional field. Create a preflight check that can run repeatedly. Flag price or availability mismatches, broken image URLs, missing critical attributes, disapproved items, and landing pages that redirect to a generic category. This work improves conventional Shopping and Search campaigns even if a new AI placement is unavailable. Recommended experiment 2: continue the conversation Build two landing page variants for one narrow intent cluster. Keep the offer, audience, bid strategy, geography, and conversion definitions stable so the page is the meaningful variable. The control can be the current category or product page. The treatment should open with the same problem and constraints that brought the visitor from search, then provide a direct fit explanation. Include a comparison, limitations, total cost details, evidence for claims, and an obvious next action. If the product is unsuitable for a common use case, say so and route the visitor to a better option. Do not imitate an AI answer with a wall of generated prose. The page should help a person verify the recommendation. Use original images, specifications, demonstrations, policies, and decision support that an ad cannot contain. Test mobile load, accessibility, form behavior, stock changes, and deep links before traffic arrives. Judge the treatment by qualified outcomes: completed purchases, accepted leads, returns or cancellations, margin where available, and downstream lead quality. A higher click through rate with worse post click outcomes is not evidence of a better AI search experience. Recommended experiment 3: test controlled expansion If AI Max or another relevant feature is available, start with a bounded campaign or portfolio that has reliable conversion data and enough budget to run without constant manual disturbance. Document the initial settings, included URLs, brand and location controls, exclusions, assets, bidding objective, and conversion values. Use a holdout where the platform and account design allow a credible comparison. If a clean holdout is not practical, establish a pre test baseline and annotate every concurrent change. Do not switch bidding, creative, landing pages, conversion definitions, and budget at once; that makes the result impossible to diagnose. Review search term and landing page behavior for relevance, not just aggregate totals. Look for expansion into needs the business can actually satisfy. Add exclusions or tighten inputs when the system reaches unsupported claims, unsuitable destinations, low value informational demand, or locations the business cannot serve. Set a decision date and minimum evidence threshold before launch. Possible outcomes are expand, revise, continue gathering data, or stop. Avoid declaring a winner from a brief fluctuation or from platform reported conversions that have not been reconciled with business records. Measurement and governance for a moving product Create a change log that separates platform events from advertiser actions. Platform events include an eligibility notice, format rollout, reporting change, or policy update. Advertiser actions include feed repairs, new assets, budget changes, exclusions, and landing page releases. Without this separation, normal product rollout can be mistaken for campaign optimization. Define guardrails before testing: approved claims, prohibited audiences, geographic limits, price tolerances, inventory rules, brand terms, and escalation owners. Review generated or assembled creative in the actual placement when previews are available. Preserve screenshots and configuration exports because interfaces and eligibility can change. The practical conclusion is neither to ignore AI Search ads nor to reorganize the entire media plan around announcements. The responsible move is to strengthen the data and destination experience now, then run narrow experiments as eligible formats become inspectable. Announced facts establish the direction. Controlled tests determine whether that direction creates value for a particular account. Sources A new generation of ads for the AI era of Search https://blog.google/products/ads commerce/google marketing live search ads/ How AI Max for Search campaigns works https://support.google.com/google ads/answer/15910187?hl=en Product data specification https://support.google.com/merchants/answer/7052112?hl=en Merchant Center product data specification update 2026 https://support.google.com/merchants/answer/16989427?hl=en Cover image credit Cover image by Zuko.io Images , made available under CC BY 2.0 . WIKIVISE cropped and converted the source image.