Should a Roofer Move Google Ads Budget to ChatGPT Ads?

No, not yet. A study of U.S. traffic found ChatGPT ad visits earned $0.89 in revenue each, against $2.45 for paid Google search visits. Most advertisers in the comparison saw no ChatGPT sales at all. Keep your search budget where it is, and test ChatGPT ads with a small, tracked amount.

That is the short version. The longer version matters if you spend real money on leads, because the details of the study tell you how to run your own test.

What did the study find about ChatGPT ads versus Google search ads?

According to a working paper by Maximilian Kaiser and Christian Schulze, reported by Matt Southern at Search Engine Journal, a U.S. ChatGPT ad visit brought in $0.89 in revenue. A paid Google search visit brought in $2.45. ChatGPT ad visits also converted at 0.37 percent, versus 1.17 percent for paid search.

The paper is dated October 2026 and is still under review. Its paid-ad figures come from 102 advertiser sites with full sales tracking. ChatGPT's first ads launched in the U.S. on Feb. 9, 2026, so this is a young channel with a short track record.

The gap held up when the authors compared both channels inside the same websites. They adjusted for device type and week, and the conversion and revenue gaps were still there. So it is not just a case of different advertisers getting different results.

Did most advertisers get any sales from ChatGPT ads?

Mostly no. Among 87 advertisers who bought both ChatGPT ads and paid search in U.S. traffic, 84 earned less per ChatGPT visit than per search visit. Seventy-three of them had zero ChatGPT sales across 18,726 visits between them. Fourteen advertisers produced all 570 ChatGPT ad purchases.

The results were also lopsided at the top. The paper found one advertiser was responsible for 66 percent of all purchases made through paid ChatGPT across the 102 sites. Take that advertiser out and both gaps get wider.

Here is my read. An average hides a few winners and a lot of people who spent money and got nothing. Some advertisers clearly make it work. But you cannot know whether you are one of them unless you track the results separately from day one.

Does a study of online sales apply to a roofing company?

Not directly. The data comes from Google Analytics e-commerce tracking, so it measures online purchases, not phone calls or estimate requests. A roofer sells a high-ticket job after a conversation. Treat this study as a warning about intent and tracking, not as a roofing benchmark.

The authors also point out limits. The data captures the last click before each visit. It does not include prompts, ad impressions or users' subscription tiers. And because advertisers chose to buy ChatGPT ads, the comparison shows their results. It does not prove the ads caused them.

For a roofer, the last-click problem is even bigger. A homeowner might see an ad, think about it for a week, then call the number on your truck. If you only count form fills, you will undercount. If you count nothing, you are guessing.

Is the visitor researching or ready to book?

Intent is the question that matters most. Someone typing "roof leaking, need repair today" into Google wants a roofer now. Someone asking ChatGPT how long a shingle roof lasts is still learning. A search ad catches the first person. A ChatGPT ad may often meet the second.

That is my explanation, not the paper's. The study did not capture prompts, so it cannot say why ChatGPT visits converted less. But the pattern fits what you already know from the phones. A lead who found you while researching is a different lead from one who found you with water coming through the ceiling.

Before you move a dollar, ask which kind of person your ad would reach. If your best jobs come from urgent, high-intent searches, that is exactly where paid search already works. If you are spending to educate people early, you are paying a click price for a slow lead.

If you are keeping your money in Google Ads, this video walks through running campaigns on a small budget.

How do I work out what a ChatGPT ad click is worth to me?

Divide your revenue per visit from ChatGPT ads by your revenue per visit from paid search. Multiply that by your search cost per click. The result is the most you can pay per ChatGPT click and still match search on tracked revenue per dollar. The paper calls this the parity price.

Here is how to run it for a roofing business:

1. Give ChatGPT ad traffic its own tracking: separate landing page, tracking parameters and a separate call tracking number.
2. Check your ChatGPT ad sessions for ad review crawler visits. According to OpenAI's basics page, those crawlers can visit landing pages with tracking parameters, and third-party analytics tools may count them as real visits.
3. Work out revenue per visit for each channel. Use booked jobs and their value, not just clicks or form fills.
4. Divide ChatGPT revenue per visit by search revenue per visit.
5. Multiply that number by your average paid search cost per click. That is your parity price.
6. Compare it with what you are actually bidding.

The authors ran this on their own data. Among 16 accounts with at least 80 percent of search traffic from the U.S., the median paid search click cost $0.72. Scaled by the revenue gap, that gives about $0.26 per ChatGPT click. Because one advertiser dominated the revenue, resampling put the estimate anywhere from $0.05 to $0.57.

Now compare that to real prices. The largest publicly known account paid $1.72 per click, according to Opascope, which runs paid media for clients. That figure comes from one account's 15 days of spend. OpenAI's ads basics page recommends starting cost-per-click campaigns at a maximum bid of $3 to $5.

So bids are running well above the parity price in the study's data. There is one more caveat from the paper. The parity price assumes similar profit margins for both channels, and it counts only the one tracked visit.

What should a roofer do with the ad budget right now?

Leave your core search budget alone. If you want to test ChatGPT ads, cap the spend at an amount you can afford to lose, track calls and booked estimates separately, and judge the test against search once you have enough visits. Do not cut campaigns that already produce booked jobs.

OpenAI's own ads FAQ says ChatGPT Ads does not yet have performance benchmarks across advertisers, industries or campaign types. Nobody can tell you what a good roofing result looks like yet. Your own numbers are the only benchmark you have.

Meanwhile, there is a cheaper way to show up in ChatGPT: be the business it names in an answer. That depends on a clean, complete Google Business Profile, consistent listings and real reviews. You can see where you stand with the AI visibility report. If your local search basics are weak, fix those first. Our local search service covers that work.

My rule is simple. Money goes where the phone rings. Test new channels with small bets, count every call, and let the numbers decide. A new ad platform will always sound exciting. The question is whether it books roofs.

Common questions

How much does a ChatGPT ad click cost?

The largest publicly known account paid $1.72 per click over 15 days, according to Opascope, as cited by Search Engine Journal. OpenAI's ads basics page recommends starting cost-per-click campaigns at a maximum bid of $3 to $5. The study's parity price, by comparison, was about $0.26 per click.

When did ChatGPT ads launch?

The first ChatGPT ads went live in the United States on Feb. 9, 2026. By early August they had reached Australia, Canada, New Zealand, the United Kingdom, Japan, Korea, Brazil and Mexico. According to the study, ads did not run in the European Union until Aug. 24, after the data window closed.

Does OpenAI publish ad performance benchmarks?

Not yet. OpenAI's ads FAQ says ChatGPT Ads does not yet have performance benchmarks across advertisers, industries or campaign types. That means you cannot compare your results with an industry average. Track your own cost per booked job on ChatGPT and compare it against your paid search numbers.

Is the study final?

No. It is a working paper dated October 2026 and still under review. The authors also note that advertisers chose to buy ChatGPT ads, so the comparison reflects their results and does not prove the ads caused them. Treat it as a strong early signal, not a settled answer.