What Google's Recommendations tab is actually optimizing for.

A few mornings ago I opened an email from Google. The subject line was about reaching more customers. Inside was a clean, well-designed pitch: upgrade your campaigns to broad match keywords, let Google's AI widen your reach, get more converting customers. There was a success story attached. An industry leader, it said, saw a 23% increase in revenue after switching to broad match.

The industry leader was CarParts.com.

CarParts.com is a national auto-parts e-commerce retailer. It sells millions of dollars of inventory a year to customers in every state. It has, for practical purposes, an unlimited number of products it can sell and an unlimited number of people it can sell them to. Broad match — which shows your ad on searches loosely related to your keywords — makes a great deal of sense for a business shaped like that.

I do not run ads for businesses shaped like that. I run ads for independent veterinary practices: single locations, a service radius measured in miles, a finite number of appointment slots in a finite week. The CarParts.com case study has almost nothing to teach a practice like that. And yet Google sent it to me, and Google sends versions of it to every advertiser, because the recommendation system is not built to know the difference.

This post is about that system — the Recommendations tab, the Optimization Score, the helpful emails — and what it is actually built to do. It is not a conspiracy piece. It's a piece about incentives, and about why a tool can be genuinely useful and still systematically point you in the wrong direction.

The incentive nobody states out loud

Here is the plainest version of it. Google's revenue is your ad spend. The two numbers are the same number, viewed from opposite sides of the transaction. When you spend more, Google earns more. There is nothing sinister in saying this; it is simply the business model, printed on the side of the box.

Now hold that fact next to the Recommendations tab. The tab surfaces a steady stream of suggestions, each one framed in the language of growth: reach more customers, capture more demand, improve your ad strength. Read enough of them and a pattern emerges. The recommendations that appear most often are the ones that increase spend.

Raise your budget. Add broad match keywords. Enable Search Partners. Expand to the Display Network. Add more headlines to your responsive search ads. Remove a negative keyword that's "limiting reach." Each of these is presented as a way to get more customers. Each of these also, by design, sends more of your money through Google's auction.

That doesn't make any single recommendation wrong. Sometimes raising the budget is exactly right. But it means the system has a direction it leans, and the direction is spend more. A recommendation engine built by a company whose revenue is your spend will, on average, recommend more spend. You don't need to assume bad faith to predict that. You only need to read the incentive.

Broad match is the clearest example

Of all the recommendations Google pushes, broad match keywords are the one I want to spend the most time on, because the gap between how it's sold and how it behaves is the widest.

A quick refresher on match types, in practice rather than in theory. Say you're an emergency vet in Missoula. If you add the keyword [emergency vet missoula] as an exact match, your ad is eligible to show on that search and a tight cluster of searches that mean the same thing. If you add emergency vet missoula as a broad match, your ad becomes eligible for a much wider universe of searches that Google's AI judges to be related.

What does "related" mean in that wider universe? In real vet accounts, I've watched broad match serve ads on searches like "is my dog throwing up," "pet insurance cost," "dog ate chocolate what do I do," "24 hour pharmacy near me," and "vet tech salary." Some of those are a person in genuine distress who might call. Most of them are not a pet owner ready to book an emergency visit. They are research, idle worry, the wrong kind of pharmacy, or someone looking for a job.

Here is the part that makes broad match so easy to sell: it will get you more clicks. That isn't in dispute. Google's AI is genuinely good at finding more searches and putting your ad in front of them. Your click count goes up. Your impression count goes up. The graph in the dashboard moves in the direction everyone likes.

The recommendation promises reach. It does not promise patients. And for a veterinary practice, those are not close to the same thing. A practice has a finite calendar. The goal was never "more clicks" or even "more website visitors" — the goal is a booked appointment with a pet owner who lives close enough to actually come in. Broad match optimizes the top of that funnel and leaves the bottom of it to chance.

None of this means broad match is never useful. In a mature account, with airtight conversion tracking and a long list of negative keywords doing constant cleanup, broad match can be a controlled tool. But that is the opposite of how the recommendation is pitched. It's pitched as a switch you flip for easy growth. Flipped without the tracking and the negatives underneath it, it mostly just widens the hole your budget drains through.

Why the average advice is wrong for your practice

Come back to CarParts.com, because the contrast is the whole argument.

An e-commerce retailer of that size lives in a different universe of numbers than a vet practice. Its conversion rate might sit somewhere around two percent — and two percent of an enormous traffic number is still a great deal of revenue. Its inventory is effectively unlimited; it can always sell one more part. Its market is the entire country; a customer in any state is a good customer. For a business with that shape, broad match casting the widest possible net is a reasonable strategy, because almost any related search has some chance of ending in a sale, and the ones that don't cost very little relative to the ones that do.

A single-location veterinary practice inverts nearly every one of those numbers. The conversion that matters is a phone call or a form submission, not an instant checkout. The "inventory" is appointment slots, and there are only so many in a week. The market is not the country; it's the set of pet owners close enough to drive in. And the cost of a wasted click is proportionally much higher, because the budget is smaller and every dollar spent on an unqualified search is a dollar not spent on a qualified one.

Google's recommendation engine cannot see any of this. It is tuned to the average advertiser across every industry that buys Google Ads — retailers, software companies, travel sites, lead generators, and somewhere far down the list, veterinary practices. The advice it produces is advice for that average. A suggestion that genuinely helps a national retailer with infinite inventory can actively damage a practice that can only see twelve new patients a week. The recommendation is not lying. It's just answering a question about a different business.

The recommendations that are worth applying

I want to be careful here, because the easy version of this post — "ignore everything Google tells you" — is also wrong, and it would be its own kind of lazy advice.

Some recommendations are genuinely good, and you should apply them promptly. The Recommendations tab will sometimes flag a disapproved ad you didn't know about, and fixing it is pure upside. It will point out conflicting negative keywords — cases where a negative you added is silently blocking a keyword you're paying for. It will catch a broken or redirecting final URL. It will prompt you to add sitelink, callout, and structured snippet extensions, which cost nothing and genuinely improve your ads. It will sometimes notice that your conversion tracking is missing or misconfigured, which — as I've written about at length — is the single most important thing to get right, and which usually comes down to a tooling problem you can fix yourself.

The common thread among the good recommendations is that they improve the efficiency or correctness of what you're already doing, rather than simply widening the funnel. They fix something broken or add something free. They don't quietly ask for more money.

A simple test for any recommendation

When a recommendation appears — in the tab, or in one of those well-designed emails — I run it through four questions before deciding anything. You can use the same four.

One: does it increase spend, or improve efficiency? A recommendation that makes your existing spend work better is usually safe to apply. A recommendation whose mechanism is "spend more" deserves real scrutiny, because that's the direction the system always leans.

Two: is the success story attached to it from a business that looks like yours? If the proof point is a national retailer, a software company, or any business with unlimited inventory and a nationwide market, the case study is not evidence about your practice. It's evidence about them.

Three: does applying it make your targeting tighter or looser? Tighter targeting — more specific keywords, more negatives, narrower geography — tends to help a small local advertiser. Looser targeting tends to help Google's auction volume. When in doubt, a vet practice usually wants tighter.

Four: would you do it if Google hadn't suggested it? Strip away the framing and the urgency and the score impact, and ask whether the underlying change is something you'd independently choose. If the only reason to do it is that Google asked, that's not a reason.

The Optimization Score is the same story

Near the top of the Recommendations tab sits a number: your Optimization Score, presented as a percentage out of 100. It is colored and styled to make 100% feel like the obvious goal, and applying recommendations is how you raise it.

It's worth being precise about what that number measures. The Optimization Score measures how closely your account aligns with Google's recommendations. That is all it measures. It is not a measure of account health, it is not a measure of return on ad spend, and it is not a measure of how many patients you booked last month. An account can sit at 100% and quietly waste half its budget. An account can sit at 70% — because the owner declined the broad match push and the budget increases — and run with real discipline.

I am not telling you to ignore the score. I'm telling you to read it for what it is: a measure of agreement with a system whose incentives you now understand. A lower score reached through deliberate choices is not a problem to fix. It's frequently a sign that someone is paying attention.

What to do if you run your own ads

If you manage your own Google Ads account, here is the practical version of all of this.

First, find the auto-apply settings inside the Recommendations tab and look at what's switched on. Google offers to apply certain categories of recommendation automatically, on your behalf, without asking each time. For a small practice account, I'd turn auto-apply off entirely. The whole argument of this post is that these recommendations need a human judgment applied to each one; auto-apply is the setting that removes exactly that judgment.

Second, keep reading the recommendations and the emails. Don't delete them unread. They genuinely do surface real problems — disapproved ads, broken URLs, tracking gaps — mixed in with the spend-more suggestions. Read them, run the four questions, and apply the ones that pass. In a typical month, on a typical vet account, that's somewhere around one in five.

Third, hold onto the frame. The Recommendations tab is an input, not an instruction. It is a list of things Google would like you to consider, generated by a system that earns more when you spend more. Treated as one input among several, it's useful. Treated as a to-do list, it will slowly reshape your account into something that serves the auction better than it serves your practice.