“Good enough” ad copy can look harmless in a draft. The risk shows up later, when the copy has to compete beside every other result on the page. Search Engine Journal’s test puts that problem in a useful place by comparing what happened when human-written and AI-generated ads ran against the same offer, audience, and format.
CTR and CPC show whether the copy earned attention at a cost the campaign could justify. With the average CPC for Google Ads being $4.66, vague copy shows up in the spend. Copy that sounds fine in a draft has to prove itself beside every other result on the page. The search page strips away the extra context a campaign usually provides.
What Happened When Generative AI and Human Copy Ran Side by Side
Search Engine Journal tested two versions of the same Google Ads campaign. One used human-written ad copy, and the other used ChatGPT-generated copy.
Both campaigns used the same offer, audience, and ad format, which makes the comparison useful. I wouldn’t treat one small test as proof that human-written ads will always beat AI-generated ads. But it does give you and your team something worth paying attention to because both versions had to compete in the same place.
Results: The Human-Written Ads Won the Click and the Cost Test
The human-written ads performed better across the metrics that matter in paid search:
- Human-written ads generated 65 clicks from 1,306 impressions.
- AI-generated ads generated 26 clicks from 713 impressions.
- Human-written ads had a 4.98% CTR.
- AI-generated ads had a 3.65% CTR.
- Human-written ads had an average CPC of $4.85.
- AI-generated ads had an average CPC of $6.05.
The AI-generated ads still drove activity, but they lost once both versions reached the market.
Analysis: Why the Human-Written Copy Had the Edge
The human-written ads didn’t win by a little. They brought in more than twice as many clicks and did so at a lower average cost. That doesn’t prove a person will beat AI in every campaign, but it does show where generic ad copy starts to hurt.
The AI version still got clicks, so the issue isn’t that the copy failed. The point is that it underperformed in a setting where “fine” copy can still cost more than it should. On the page, the AI-generated ad had to sit next to other options and give someone enough reason to stop there. The human-written version appears to have done that better.
That’s what appears to have happened here. The AI copy had enough to get some attention, but the human-written version gave people more reason to click. In paid search, that difference shows up quickly because the audience is already comparing options.
Use AI Where it Strengthens the Process
AI can quickly produce a long list of ad options, but the tool won’t know which are safe to run. That part still has to happen before the copy goes live. Before prompting, the team should know what the ad is allowed to promise. The landing page has to support it. The offer has to match it. The claim has to be specific enough that a reviewer can say yes or no, not just “sounds good.”
Give AI the Details That Shape the Message
If the prompt you feed to AI uses only the product page as an example, the output will just repeat the same approved language. That can work for a rough first pass, but it won’t help the ad say anything new.
Better material is usually sitting in places you already have access to. For example, you can:
- Look at the searches people used before they clicked.
- Read the objections sales keep answering.
- Check the ads competitors are running, too, because you don’t want the prompt repeating the same claims in slightly different words.
Customer language is useful when the messaging doc starts to stray from how people describe the problem. They might give you the phrase the product page can’t. Even if that line never makes it into the ad, it can change the prompt enough to keep the first draft from sounding like a product summary.
Don’t Let Clean Copy Hide Weak Claims
Vague ad claims don’t always look obvious in a draft. “Save time,” “reduce costs,” and “improve results” all sound fine at first. Before a line like that goes live, you or someone from your team needs to push on it a little. What kind of time? Which costs? What result improves?
If your landing page doesn’t answer those questions, the ad you’re working on may be asking for a click, but the offer can’t really support it. That doesn’t mean the line is false. It may just be too broad for paid search, where even a harmless claim can waste budget if it doesn’t give the searcher enough to act on.
The Human Judgment Behind Better Ad Copy
SEJ’s results point to a problem with AI-generated ad copy. The copy only has to miss the better angle, blur the promise, or sound enough like the next result that the campaign pays more for weaker attention.
Someone still has to look at the draft and ask whether the promise is specific enough, whether the landing page can support it, and whether the ad gives the searcher a reason to pick that result over the others on the page.
Last updated on 5/22/2026