Do Google and Meta's Ad AI Goals Match Your Business Goals?

Google's and Meta's AI optimization goals conflict with advertisers and other third-party businesses in one narrow, predictable place: each system maximizes the conversions it counts against the target you typed in, and your profit is a different number. We think that is a structural conflict to manage, and a poor reason to switch the automation off. The company selling the impressions also designs the auction, counts the results and suggests the next budget, so the budget decision should rest on at least one number the seller did not produce. Vendor pages are quoted as we read them on October 7, 2026.

What are the two ad systems built to optimize?

Google defines Smart Bidding as "bidding strategies that use Google AI to optimize for conversions or conversion value in every auction", and the same page maps the business goal "Increase profit" to Target ROAS and Maximize conversion value. By our reading, profit enters only as a revenue ratio. The bidder works from the conversion values your tracking sends, so margin, returns and whether the buyer was already on the way stay outside the number unless you put them in. Our guide to ROAS and break-even math shows how a healthy ratio can still lose money.

According to Meta, "the winner of the auction is the ad with the highest total value, subject to a price floor (minimum price) that may affect whether an ad is shown and the price paid." Which conversions the system counts and optimizes for depends on the attribution model. A separate page from Meta says the model you choose will "inform ad delivery and determine how conversions can be credited to your ads." The counting rule is also the optimization target. Both vendors' descriptions stop at the conversion, and we found no mention of margin in either.

What the court record adds about the seller's goals

The seller has a goal of its own, and one of the fullest public accounts we found sits in a court opinion. On August 5, 2024 the U.S. District Court for the District of Columbia issued its memorandum opinion in United States v. Google LLC; we read the copy hosted by CourtListener. One finding of fact reads: "Google designs the auction and controls underlying inputs that can affect the ultimate price generated by the auction." A later one: "Google has used three primary pricing knobs to influence prices: (1) squashing, (2) format pricing, and (3) randomized generalized second-price auction." In its conclusions the court wrote that Google used those knobs to drive cost-per-click increases "often between 5% and 15% at a time", and added: "Many advertisers do not even realize that Google is responsible for the changes in price."

The limits matter. The findings cover general search text ads, and the pricing studies and launches they describe run mostly from 2017 to 2020. They make no finding about pricing under Smart Bidding, Performance Max or Meta's auction, and other determinations in the same opinion favored Google. Google has also appealed, and a brief the U.S. Department of Justice filed on July 28, 2026 lists the finding of monopoly power in general search text advertising among the issues that appeal raises. What the trial court found is this: "Google's records make clear that growing its revenue was a principal goal in launching these price tunings."

Meta's auction page makes a related disclosure in its own words. Meta may "make adjustments that could affect auction outcomes and prices" for purposes it lists, such as improving ad quality and optimizing the auction, with the limit that such adjustments "will not cause us to charge you more than your bid to show your ad."

Is this a conflict of interest, and how exposed are you?

Platform dependency risk with big tech ad AI comes down to a plain conflict of interest over optimization goals: one party sells the inventory, runs the auction, grades the results and advises the buyer on how much more to buy. Dependency gives the conflict its weight. The same opinion found that "Advertisers consistently testified that shifting significant ad spending from Google to Bing would be ineffective (and unwise) because of Bing's lack of scale." A buyer who cannot leave has to manage the seller's incentives from inside the account, and three gaps deserve attention.

Conversions are counted by the seller

Meta's standard attribution counts events within 1 day or 7 days after a link click, depending on the window you choose, and within 1 day after an impression (the page notes some accounts remain on prior versions of click-through during a rollout). Google's Smart Bidding page says of video ads optimized for conversions: "Viewing at least 10 seconds of a video ad counts as leading to a website conversion if the conversion happens within 3 days of the engagement." Credit follows contact with an ad. Whether the ad caused the sale is the incrementality question, and Meta's own incremental model "optimizes delivery for incremental conversions using models that predict whether a conversion is caused by an ad." We read that as the vendor treating them as two counts.

Recommendations lean toward volume

Google calls optimization score "an estimate of how well your Google Ads account is set to perform." For a Target ROAS campaign the score "is geared towards driving more conversion value at a similar ROAS, like expanding targeting and ensuring the budget isn't constraining additional high value conversions." Opportunity score, which Meta labels "in development", is "based on how many Meta Ads Manager recommendations you apply". The same page says: "Opportunity score (including a high score) itself does not reflect your actual or future performance." We found no input for profit margin in either page's account of how the score is built, which is why a budget raise needs the marginal-return check in our piece on scaling paid ads without killing ROAS.

Brand demand looks like new demand

Google offers brand exclusions that stop a Search or Performance Max campaign from serving on queries tied to brands you list, and recommends using brand settings only for campaigns that really need them "so as not to limit your conversion opportunities." In our view the caution should run the other way. A search for your own name comes from someone who already knows you, so a campaign free to buy it can report strong returns on demand it did not create. New-customer CAC is where that shows up.

What goes wrong when the platform AI becomes the strategy?

The risks of relying on Google and Meta AI platforms for business strategy show up fastest in one month of numbers. The example below is hypothetical: the 40% incremental share and the 45% gross margin are assumptions chosen to show the arithmetic.

At a 45% margin the break-even return is 1 divided by 0.45, about 2.2, and the ROAS calculator will run that for your own margin. The dashboard clears the bar at 5.0 and the holdout misses it at 2.0. A suggestion to add budget "at a similar ROAS" is sound on the dashboard's terms and loses money on the holdout's. Accept a year of such suggestions and three decisions have changed hands without anyone choosing it: what counts as a result, how much to spend, and which customers are new.

The checks that keep the budget decision with you

Three account checks are worth running before the next budget conversation.

• State your own objective. Google's optimization score page says "you can override the system's inferred strategy." It also lets you choose which conversions are used for optimization, so keep low-intent actions out of that set.
• Treat both scores as a review queue. Google says applying or dismissing recommendations changes the optimization score. Meta says a dismissed recommendation leaves opportunity score unchanged, so if business constraints keep you from applying some recommendations, the score "will always be lower than 100 points."
• Compare counting rules on Meta. The attribution page warns that "comparing across different attribution models will lead to inaccurate conclusions." It adds: "If using external analytics tools, evaluate performance in those tools."

For a measurement the seller did not grade, run a holdout. Google describes incrementality experiments as "the way to measure the causal impact of ads." It also states that "Conversion Lift isn't available for all Google Ads accounts" and routes access through a Google account representative. Meta offers Conversion Lift in Experiments at no additional cost. As a guide, the ad account needs a campaign that started in the past year with $5,000 USD or more in spend and at least 500 conversions, and self-serve tests carry signal quality requirements. The platform being tested still runs both tests, so pair them with a geographic holdout read in your own order data and with new-customer counts from your books.

Where we land

Platform AI should execute and the advertiser should decide. Let Smart Bidding and Meta's delivery system set bids across more auctions than a team can watch, and keep three things outside them: the definition of a result, the proof that results are incremental, and the size of the budget. Any recommendation that touches one of those three gets tested against a number from your own books before it gets a yes. Reviewing where an account has handed those decisions over is the kind of work our paid media team does.

Sources

• Google: About Smart Bidding
• Google: About optimization score
• Google: About brand settings for Search and Performance Max
• Google: About Conversion Lift
• Meta: About ad auctions
• Meta: About attribution models and attribution settings
• Meta: About opportunity score in Meta Ads Manager
• Meta: About Conversion Lift
• CourtListener: United States v. Google LLC, No. 20-cv-3010 (D.D.C.), Memorandum Opinion, filed August 5, 2024
• U.S. Department of Justice: Response Brief and Opening Brief on Cross-Appeal for the United States and Co-Plaintiff States, United States v. Google LLC, Nos. 26-5023, 26-5047, 26-5049 (D.C. Cir.), filed July 28, 2026