You open your Google Ads account on a slow Tuesday and click into your search ad. Google now shows you a neat table, one row per headline. "Same-Day Drain Repair" sits at the top with nine conversions. "4.9 Stars From 600 Reviews" is near the bottom with one. The decision feels obvious. You delete the loser, write a new headline to replace it, and close the tab feeling like someone who just ran a test.
You did not run a test. You read a scoreboard for a game where the referee decided who got to play, how long they stayed on the field, and which teammates they played beside. Every number in that table is real. None of them answers the question you think it answers, which is whether one headline sells better than another.
This matters because ad copy is one of the few levers a small business pulls entirely on its own, and the reporting Google gives you for it is easy to misread. The good news is that a real test is built into the same account, two clicks away. Here is why the headline table cannot be your test, and what to use instead.
Google Retired the "Low" Label. The Habit of Deleting Headlines Stayed.
Until mid-2025, Google graded every headline and description in a responsive search ad with a label: Best, Good, Low, or Learning. A whole generation of advice grew up around those labels, and the most common piece of it is still circulating in tutorials today: go into your ad, find everything marked Low, delete it, and write something new.
Google has since retired that column. Its help page for the asset report now says plainly that the "Performance label" column "has been deprecated as full performance statistics for each asset are now available," with the new numbers reported from June 5, 2025 onward. So instead of a grade, every headline now carries its own impressions, clicks, cost and conversions.
That sounds like an upgrade, and in one way it is: more information is better than a vague label. But it also made the old habit more tempting. A label of Low at least looked like an opinion. A column that says one conversion next to a column that says nine looks like a result. Owners who used to prune by label now prune by conversion count, and they feel more scientific doing it.
The problem was never the label. The problem is what the numbers are measuring.
Every Headline Shares Credit With the Lines Shown Beside It
A responsive search ad is not one ad. You hand Google up to fifteen headlines and four descriptions, and Google assembles them, auction by auction, into whatever combination it predicts will work best for that particular search. A searcher might see three of your headlines and two of your descriptions. The next searcher sees a different three.
So when somebody clicks and books a job, who earned it? The headline in position one, the one in position two, the description underneath, or the combination of all five? Google's answer is to credit every piece that was on screen. Its help page explains that asset-level metrics are "attributed per instance of the asset served within an ad," which is why the headline numbers do not add up to the totals for the ad.
Think about what that does to your nine-versus-one comparison. "Same-Day Drain Repair" may have nine conversions because it is genuinely persuasive. Or it may have nine because Google kept pairing it with your strongest description and your best offer line, and it was simply standing in the room when the sale happened. The table cannot tell those two stories apart.
Google does not pretend otherwise. On the same page, it says ratio metrics at the asset level, things like click-through rate and cost per conversion, "should be used as directional indicators only," because "these ratios are influenced by the combination of assets served together." That is Google telling you, in its own documentation, not to treat the table as a verdict.
Google Decides Which Headlines Run Before Any of Them Can Win
There is a second problem, and it is bigger than shared credit. In a fair test, every contender gets a fair chance. In a responsive search ad, Google decides early which headlines it believes in and serves those far more often than the rest.
That is the system working as designed. The default ad rotation setting, which Google calls Optimize, "prioritizes ads that are expected to perform better," and as data comes in, serving "will become weighted more heavily in favor of" those predicted winners. The same logic runs inside each ad's mix of headlines. A line that starts slowly gets fewer chances to recover, which gives it even less data, which keeps it buried.
In one live account shown in a recent Google Ads training video, the lowest-ranked headlines had zero impressions. Not few. Zero. A headline that never appeared did not lose to anything. It was never in the race, and deleting it because its conversion column is empty is like firing an employee for missing a meeting nobody invited them to.
The same thing happens one level up, when you put two whole ads in an ad group and assume they are splitting the traffic. Paid Media Pros, a US agency that publishes its own account tests, pulled the numbers on one of theirs: three ad versions running in the same campaign over the same stretch of months had collected roughly 8,400, 3,000 and 200 impressions. The 200 version was never given a fair hearing. Even Google's other setting, "Rotate indefinitely," only promises that ads will "enter the auction more evenly," and Google itself warns that "ads don't always rotate evenly" because entering an auction is not the same as winning it.
Ad Strength Grades the Homework, Not the Sales
While you are in that part of the account, you will also see Ad Strength, the meter that runs from Poor to Excellent. It is tempting to read it as a performance score. It is not one, and Google is unusually direct about it: Ad Strength "isn't used to calculate Ad Rank, Quality Score, or auction wins."
Ad Strength checks whether you gave Google enough varied material to work with: enough headlines, different enough from each other, with your keywords in them. That is useful when you are writing an ad, and we covered what it is good for in the customer only reads your ad. But a perfectly varied set of forgettable headlines can score Excellent, and an ad that books jobs can sit at Average. It grades the homework, not the sales.
Google's Experiments Page Splits Searches Fairly. The Asset Report Never Did.
Everything above has the same root cause. In the normal account view, Google is both running the ads and deciding who gets seen, so you can never separate "this copy is better" from "Google showed this copy more." A real test needs one thing the normal view cannot give you: a fair split, decided before anyone sees an ad.
That is exactly what the Experiments page in Google Ads is for. It sits in the left-hand menu under Campaigns, and it gives you two ways to test copy.
Ad variations are for changing one line. You tell Google which ads to target, then choose a change: find a phrase and replace it ("Book Online" becomes "Same-Day Appointments"), add, remove or pin a headline, or swap the landing page URL. Google then runs the original and the variation side by side. Its help page notes that the variation uses cookie assignment so that a person "may discover only one version of your ad, regardless of how many times they search," which is what keeps the comparison clean.
A custom experiment is for testing something bigger, like a whole new set of headlines built around a different idea. Google makes a copy of your campaign, you put the new ads in the copy, and the two run against each other on a split of traffic and budget you choose, usually 50/50. When one wins, Google lets you apply the winning version to your original campaign.
Two pieces of Google's fine print are worth knowing before you start. First, a 50/50 split "controls auction eligibility but doesn't guarantee equal impressions or spend," so check that both sides are actually serving. Second, Google says to "allow 7-14 days for the treatment arm to stabilize," which means the first week or two of results is noise. Do not peek on day three and call it.
And judge the result on the number that pays your bills. Clicks are easy to win with vague, curious copy. Conversions are better. Booked jobs or sales are best, which is only possible if your conversion tracking counts real leads, something we unpacked in your conversion number is going up, your sales aren't.
An Experiment Can Tell You Your New Ad Copy Lost. The Asset Report Can't.
Here is the clearest way to see the difference between a scoreboard and a test. A scoreboard always crowns somebody. A test can come back and tell you that your new idea is worse.
Paid Media Pros ran exactly this kind of experiment on one of their own accounts, testing the same new ad copy against their existing ads in two campaigns at once. In the non-brand campaign, the new copy won clearly: about 28% lower cost per conversion and 40% more conversions. In the brand campaign, the same new copy lost: cost per conversion went up about 22% and conversions fell about 14%. Their best brand ad, by their own account, is still one they wrote in March 2023, and nothing they have tested since has beaten it.
That is what a useful answer looks like. The same words worked for people who had never heard of the business and failed for people searching for it by name. No asset table would ever have shown that, because the asset table does not compare your new copy to your old copy. It only ranks the lines Google already decided to show.
Google's Fine Print Assumes 100 Conversions a Day
Now the part most testing guides leave out. On its custom experiments page, Google says that "for reliable results, ensure your base campaign meets minimum requirements, such as more than 100 daily conversions."
Read that again with your own account in mind. Take a local service business getting 30 conversions a month, roughly one a day. Google's own bar for a reliable experiment is about a hundred times that volume. Split that one conversion a day 50/50 and each side gets about 15 a month. A real difference will take months to show up, and a small one may never show up at all.
That does not mean a small account should skip testing. It means it should test differently. The same arithmetic shows up on every platform, and we have worked through it for LinkedIn ad tests and for Google's own Demand Gen tests. For search copy in Google Ads specifically, three adjustments do most of the work.
Run one experiment at a time, and run it long. Every test you add splits your conversions again. One experiment, left alone for six to eight weeks past the settling period, will tell you more than three tests that each end in a shrug.
Test the line that carries the most weight. The first headline slot is filled every time your ad appears, and it is the first thing a searcher reads. So the most valuable test a small account can run is often a pinning test: an ad variation that pins your strongest offer, say "Flat-Rate Price, Quoted Before We Start," into headline position one, against the same ad left unpinned. Pinning narrows the combinations Google can try, so you are trading some of its flexibility for a guarantee that your best argument is always the first thing people see. That trade either pays off in booked jobs or it does not, and an experiment is how you find out.
Change the idea, not the wording. "Call Today" against "Call Now" will never produce a difference you can see at this volume. An offer-led ad against a trust-led ad might. The reasons small numbers mislead, and why checking early makes it worse, are the same ones we walked through for website tests in you won the A/B test, you probably just got lucky.
The Asset Report Is a Maintenance List, Not a Verdict on What Sells.
None of this makes the headline table useless. It just has a different job than the one most owners give it.
Use it to find headlines with zero or near-zero impressions. Google is not using them, so they are dead weight, and replacing them with a genuinely different idea gives the system something new to try. Use it to spot lines that are off-message, outdated, or promising something you no longer offer. And use it to see which themes Google keeps reaching for, which is a hint, in Google's own word a "directional" one, about what your searchers respond to.
What you should not do is delete a headline because its conversion number is lower than its neighbor's. That number was shaped by who it was paired with and how often Google let it play. Keep a simple log instead: what you changed, when, how you tested it, and what happened to booked jobs. In six months, that log will be worth more than every asset table you have ever scrolled past.
If you would rather spend your Tuesdays running the business than reading experiment reports, testing ad copy properly is a routine part of our Google Ads management. And if you want to run it yourself, you now know where the real test lives.
