Tag: AI Advertising

If you run Google Ads for a large product catalog, your next growth problem may not be finding more keywords. It may be helping Google’s systems understand which products fit searches that are longer, more specific, and harder to classify.

That changes the work. You need product data that makes relevance clear, a controlled way to give overlooked SKUs another chance, and measurement that distinguishes genuine discovery from automated spend.

The opportunity has shifted from keywords to interpretable intent

A conventional product query might name a category and little else. A conversational query can include the shopper’s use case, constraints, preferred features, and stage of decision-making in one sentence. That extra context is commercially valuable if the ad system can interpret it and find a suitable product.

Google says AI Max can match ads to complex or ambiguous searches that traditional keyword targeting could not readily monetize. The company described this as billions of additional potential ad-bearing searches. AI Max had also moved out of beta and reached more than 500,000 advertisers by Alphabet’s Q2 2026 earnings call.

The scale is notable, but it shouldn’t be mistaken for a performance guarantee. Google attributes an average 15% lift in conversions or conversion value at a similar return on ad spend to advertisers using AI Max or Performance Max. It also says Gemini has improved Shopping-ad relevance for complex queries by about 20%. These are aggregate, vendor-supplied figures. Your result will depend on your catalog, margins, tracking, offers, product information, and the demand available in your market.

The important distinction is that better matching creates reach; it does not manufacture qualified demand. A shopper still needs a real problem, and your product still needs to solve it at an acceptable price. Treat AI-powered reach as an opportunity to enter more relevant decisions, not as proof that every new impression is valuable.

Google is also testing AI Mode formats that move ads closer to an answer experience. Highlighted Answers can place labeled sponsored links in AI-generated lists, while contextual sitelinks and Direct Offers are intended to respond to information surfaced during a conversation. These formats indicate where discovery could go, but they are still developing. Build your strategy around accurate product evidence and sound economics, not an assumption that any particular experimental placement will become material.

Give Google a product record it can match to real needs

When matching moves beyond literal keywords, the quality of your inputs matters more. Google needs enough consistent information to connect a shopper’s stated need with the product that can satisfy it. A generic title, thin product page, recycled image, and incomplete feed leave the system very little evidence to work with.

Translate conversational intent into product evidence

Start with the language of a decision, not a list of keyword variants. A useful intent statement combines the product, the intended use, and the constraint that will decide the purchase. For example, a shopper may need an item for a particular environment, compatible with equipment they already own, within a size limit, or suitable for a specific recipient.

For each important intent, create a short query-to-evidence record:

1. Write the shopper’s need in plain language.
2. Identify the product fact that proves suitability, such as dimensions, material, compatibility, capacity, fit, intended user, or supported use.
3. Confirm that the fact is accurate and present in the feed where an appropriate attribute can carry it.
4. Show the same fact in the creative when it is visually or verbally important.
5. Make the proof easy to find on the landing page, close to the price and purchase decision.

This isn’t a keyword-stuffing exercise. Repeating a phrase doesn’t establish relevance. A precise compatibility statement, measurement, material, or use limitation gives the system and the shopper something concrete to evaluate.

Your feed, visible product page, and Product structured data should also agree. Check prices, availability, variants, identifiers, names, and decisive attributes across those surfaces. If they conflict, you are asking automated systems to resolve uncertainty at the moment they should be deciding whether to show the product.

Use the same standard for creative assets. The image and copy should distinguish the SKU rather than merely represent its category. If two products solve different problems but use interchangeable descriptions and images, the system has weak evidence for choosing between them.

Apply an eligibility gate before buying more reach

Not every low-traffic SKU deserves more exposure. Before a product can enter an AI-powered discovery or recovery campaign, verify that it is:

• Currently sellable, correctly priced, and available to the intended customer.
• Economically viable under the budget and loss limits you are prepared to accept.
• Represented by accurate feed data, useful creative, and a functioning landing page.
• Distinct enough that you can explain why someone would choose it over nearby products in your own catalog.
• Appropriate for the current season and market rather than temporarily irrelevant by design.
• Measured by a conversion action that reflects business value, not merely an easy on-site interaction.

This gate prevents a common misreading of automation. More reach can reveal latent product demand, but it can also expose weak merchandising faster. If a SKU is unavailable, poorly differentiated, or uneconomic, the right action is to repair or exclude it rather than pay an algorithm to rediscover the same problem.

Create a recovery lane for products the algorithm stopped testing

Large catalogs develop a performance feedback loop. Products with strong history keep winning impressions and conversions. Products with little history receive less traffic, which leaves them with even less evidence to compete for future traffic. A viable SKU can become invisible without ever receiving a clean test of demand.

A recovery campaign interrupts that loop. It moves eligible but underexposed products into a dedicated Performance Max campaign, where they can receive another opportunity to generate impressions, clicks, and conversions. The goal is not to force every product to spend. It is to separate lack of opportunity from lack of demand.

Define a recovery SKU with rules you can audit. Its status should mean that the product is sellable and strategically eligible but has fallen below your business’s floor for meaningful opportunity during a chosen lookback period. Align that period with your buying cycle and seasonality. A universal impression or click threshold would be misleading because catalog size, price, purchase frequency, and demand differ.

Your operating rules should cover five decisions:

• Entry: What combination of low impressions, low clicks, or absent conversion opportunity qualifies an otherwise viable SKU?
• Exclusion: Which products are intentionally paused, out of season, unavailable, disapproved, unprofitable, newly launched under a different process, or missing required data?
• Isolation: How will you remove the product from its original Shopping campaign while it is in recovery so the campaigns do not overlap?
• Graduation: What evidence means the product has earned a return to its original campaign?
• Retirement: When should repeated spend without useful progress end the test?

Isolation is essential. If a recovery SKU remains active in its original campaign, you won’t know which environment produced its new opportunity, and the two campaigns may compete to serve the same product. The label that admits a SKU to recovery should also trigger its exclusion from the original campaign.

At catalog scale, automate the movement rather than relying on periodic manual cleanup. One working pattern uses BigQuery to evaluate each SKU, a Google Sheet to carry eligible IDs, Feedonomics to apply a custom label, and Google Ads to route labeled products into a dedicated Performance Max campaign. When a SKU no longer meets the recovery criteria, the label is removed and the product returns to its original campaign.

You don’t need that exact technology stack. You do need one authoritative SKU list, deterministic entry and exit logic, an automated feed label, mutual campaign exclusions, and a log of every movement. Without those controls, a useful recovery strategy becomes a recurring campaign-maintenance task with unreliable measurement.

The potential is visible in an early two-week implementation involving 13,829 previously overlooked SKUs. Those products moved from zero activity to 198,774 impressions, 1,617 clicks, $5,072.17 in cost, 24.42 conversions, and $5,161.70 in conversion value. That produced 101.77% ROAS during the recovery period.

Those figures demonstrate that an isolated campaign can restart data collection; they are not a general benchmark for profitability. The result came from one early implementation, and its stated objective was rehabilitation rather than maximizing immediate ROAS. The decisive test comes later: whether graduated products retain useful performance after returning to their normal campaign structure.

Measure discovery separately from harvest performance

A mature Shopping campaign usually optimizes around revenue, conversion value, or ROAS. A product-recovery campaign has an earlier job: determine which neglected SKUs can attract qualified attention and build enough evidence to rejoin the main system. Applying only the mature campaign’s efficiency target can recreate the same feedback loop you are trying to break.

That does not mean cost is secondary or unlimited. Automation can spend quickly, so define the campaign budget, the maximum acceptable loss, and the conditions for stopping an unproductive SKU before launch. Discovery is a learning objective, not permission to buy data indefinitely.

Track each entry cohort through a measurement ladder:

1. Eligibility: How many products passed the data, availability, margin, and operational checks?
2. Activation: What percentage of entering SKUs received at least one impression?
3. Engagement: What percentage received at least one click, and how much did that engagement cost?
4. Commercial evidence: Which SKUs generated conversions or conversion value while in recovery?
5. Graduation: What percentage met the exit condition and returned to the original campaign?
6. Post-return performance: Did graduated SKUs continue receiving impressions, clicks, conversions, and value after re-entry?
7. Incrementality: Did the process produce more total catalog value, or merely redistribute traffic that other products would have captured?

Keep the cohort log at product level. At minimum, record the SKU, entry date, reason for entry, prior campaign, recovery impressions, clicks, cost, conversions, conversion value, exit date, exit reason, destination campaign, and post-return results. This record becomes more important as AI matching reduces your visibility into exactly how every query was interpreted.

Four simple derived metrics make the operation easier to manage:

• Activation rate = SKUs with an impression divided by SKUs entering recovery.
• Engaged-product rate = SKUs with a click divided by SKUs entering recovery.
• Graduation rate = SKUs meeting the exit rule divided by SKUs entering recovery.
• Cost per graduated SKU = total recovery spend divided by the number of graduates.

These metrics won’t replace revenue or ROAS. They tell you where the recovery mechanism is working or failing before you evaluate downstream commercial value.

Treat these as diagnostic starting points, not automatic conclusions. Several causes can produce the same pattern. A click without a conversion, for example, could reflect the offer, the landing page, measurement, or simply insufficient evidence. Inspect the full path before changing bids or removing the SKU.

If you need to estimate incrementality, keep a comparable group of eligible products outside the recovery campaign or introduce cohorts in stages. Compare total catalog outcomes, not only the isolated campaign’s dashboard. Without a comparison, a rise inside the recovery campaign cannot tell you how much demand was genuinely added versus shifted from another product or campaign.

Key takeaways

• AI Max expands the range of Search intent Google may be able to monetize, while a Performance Max recovery campaign can give overlooked products a separate route back into consideration.
• Better matching begins with discriminating product facts carried consistently across the feed, creative, visible landing page, and structured data.
• A low-traffic SKU is not automatically a bad product. Separate products that lack opportunity from products that are unavailable, uneconomic, seasonal, or genuinely unwanted.
• Use explicit entry, exclusion, graduation, retirement, and loss rules. A recovery campaign should be a controlled system, not a permanent holding area.
• Measure activation, engagement, graduation, and post-return performance before deciding whether the process creates durable value.
• Google’s aggregate lift figures are directional context, not targets for your account.

Your practical next step is to export product-level performance for a lookback period that fits your purchase cycle. Filter for sellable SKUs that received no meaningful opportunity, inspect their product records, and admit only the clean, viable candidates to a bounded recovery cohort. Give every SKU an entry reason, an exit condition, a loss ceiling, and a scheduled post-return review. That is how AI-powered reach becomes a product-discovery system you can govern rather than another opaque campaign setting.

References

• Search Engine Land — Google says AI Max unlocks billions of new monetizable searches
• Search Engine Land — How to revive overlooked ecommerce SKUs with Performance Max