How to Measure AI Ads: A Guide to Incrementality Testing for ChatGPT and Google AI Mode

AI ads are becoming popular, but it is not always easy to know if they are bringing real results. The reports inside advertising platforms may show clicks, leads, and sales, but they do not always prove that ads caused those results. ChatGPT ads are still new, and their measurement tools are developing. Google AI Mode ads are included in regular Search, Shopping, and Performance Max reports, making them harder to track separately.

So, how can you know if these ads are actually working? One useful method is incrementality testing. It helps you find out how many extra sales or leads your ads bring. You show ads to one group and hide them from another group, then compare the results. This guide explains the process step by step, what numbers to track, and which common mistakes to avoid.

Plan your test first. Decide what to measure, how long to test, and what results matter.

Use both methods. Platform reports help improve campaigns, while incrementality tests show their real impact.

---

Key Highlights

• Incrementality measures extra results. It shows how many additional sales or leads your ads bring.
• Ad reports can be misleading. Clicks and conversions don't always prove ads caused the results.
• ChatGPT ads can be tested by region. Advertisers can now use location-based targeting to compare results.
• OpenAI is improving ad measurement. Its partners are working on better ways to test ad performance.
• Google AI Mode ads are harder to track. Their results are included in regular advertising reports.
• Plan your test first. Decide what to measure, how long to test, and what results matter.
• Use both methods. Platform reports help improve campaigns, while incrementality tests show their real impact.

---

Why AI Ads Are Hard to Measure Right Now

Traditional search ads were easier to track. A person searched for something on Google, clicked an ad, visited a website, and made a purchase. Marketers could follow this journey and see where the customer came from.

AI ads make this process more difficult. A person might chat with an AI tool, ask several questions, see an ad, and leave without buying anything. They may return three days later using another device and make a purchase. In this case, the report may give credit to a Google search or direct website visit instead of the AI ad that first introduced the product.

Sometimes, the opposite happens too. An ad platform may take credit for a sale that would have happened anyway. This makes the campaign look successful even when the ad did not bring any extra sales.

That's why marketers need a better way to measure AI ad performance, not just rely on clicks and conversion reports.

ChatGPT ads: new channel, early tools

OpenAI started testing ads in the US in February 2026. At first, advertisers could pay based on impressions (CPM). In May, it also introduced cost-per-click (CPC) bidding. Advertisers can now track clicks and use tools like the Pixel and Conversions API to measure results.

However, measuring the actual impact of these ads is still developing. OpenAI is working with partners such as Hightouch, Tealium, and LiveRamp to improve conversion tracking. It has also named Kantar and Cint for brand studies. For incrementality testing, Haus, Measured, and WorkMagic are exploring ways to run location-based experiments.

What does this mean for marketers? The basic tracking tools are available, but platform-reported ROAS (Return on Ad Spend) does not prove that ads are bringing extra sales. Marketers still need proper testing to understand their real impact.

Google AI Mode ads: blended into your reports

Existing Search, Shopping, and Performance Max campaigns can show ads in AI Mode and AI Overviews. Google has said it does not offer a segmented report for ads that show inside AI Overviews. Industry experts have said the same about AI Mode: you cannot open your account and see which ads ran there. Results flow into your standard campaign and search-term reports.

So you cannot answer "how did AI Mode do?" from the dashboard. You can answer "did this set of campaigns drive extra revenue?" with a test.

Why last-click is not enough

Last-click credit favors the final touch, usually a brand search or a direct visit. AI ads often sit earlier in the journey. If you rely on last-click alone, you may cut a channel that is quietly helping, or fund one that is only collecting credit.

---

What Is Incrementality Testing?

Incrementality testing helps you answer one important question: "How many extra sales or leads did our ads actually bring?"

To find the answer, you compare two groups of people:

• Test group: People who see your ads as usual.
• Control group (holdout): People who do not see your ads.

You run the test for a fixed period and compare the results. If the test group gets more sales or leads than the control group, the difference helps you understand the extra results generated by your ads.

In simple words, incrementality testing shows whether your ads are bringing new customers or simply reaching people who would have purchased anyway.

Total conversions vs incremental conversions

The gap between "total" and "incremental" is where wasted spend hides.

The simple formula

Incremental conversions = Test group conversions - Control group conversions (scaled to the same size)

Lift % = Incremental conversions ÷ Control group conversions × 100

iCPA = Ad spend ÷ Incremental conversions

iROAS = Incremental revenue ÷ Ad spend

The "i" in iCPA and iROAS stands for incremental. These are the numbers a finance team will trust.

---

What You Can and Cannot Measure Today

If you want a second opinion on your measurement setup before you spend, our team can review it. See how we help with AI strategy consulting.

---

Which Incrementality Method Should You Use?

There is more than one way to run a test. Each has a trade-off.

Geo Holdout (Best Fit for ChatGPT Ads)

ChatGPT ads support location-based targeting, which makes geo holdout testing useful. You can show ads in some areas and keep them off in other similar areas. Then, compare the sales or leads from both groups.

For example, you can run ChatGPT ads in two states while keeping ads off in two similar states. If the areas where ads ran bring more sales, the difference can help you measure the impact of your ads. Haus also uses this approach by comparing areas that see ads with areas that do not.

Conversion Lift (Useful for Google Ads)

Google Ads offers Conversion Lift studies to help marketers understand how many extra conversions their ads generate. These studies compare people who saw ads with a control group that did not.

Google has also reduced the spending requirements for some lift studies in the past. However, eligibility rules can change, so check your Google Ads account for the latest requirements.

Why Pre- and Post-Testing Can Be Risky

Comparing sales before and after an ad campaign may seem simple, but it can give misleading results. For example, if you launch ChatGPT ads during a holiday sale or a major news event, sales may increase for reasons unrelated to your ads.

You may wrongly give the ads credit for that increase. Use this method only when other testing options are not available, and remember that its results are less reliable.

---

7 Steps to Run an AI Ads Incrementality Test

1. Set a Clear Goal

Start by deciding what you want to learn from your ad test. For example, do ChatGPT ads bring more new customers or increase sales by 5%? Keep your goal specific and easy to measure. A clear question will help you understand whether your ads are delivering real business results.

2. Choose One Main Metric

Choose one main metric that shows how your business is growing. You can track sales, revenue, qualified leads, or new customers. Avoid using clicks as your main metric because more clicks do not always mean more sales. Focus on results that matter to your business.

3. Check Your Tracking Before Starting

Before launching your test, make sure your tracking tools are working properly. For ChatGPT ads, check your Pixel and Conversions API setup. For Google Ads, review enhanced conversions and offline conversion tracking. Also, make sure the data in your reports matches your actual sales and leads.

4. Select Your Test and Control Groups

Divide your audience or locations into two groups. The test group will see your ads, while the control group will not. If you are testing by location, choose areas with similar sales and customer behavior. Check at least 8–12 weeks of past data to make sure both groups had similar results before the test.

5. Decide the Test Duration and Budget

Decide how long your test should run and how much you can spend. The test period should match your customers' buying habits. For example, if customers usually take two weeks to make a purchase, your test should cover that period. Also, allow extra time for late purchases to appear in your reports.

6. Keep Other Marketing Changes Under Control

Avoid making major marketing changes while your test is running. Do not launch big discounts, change product prices, or make major changes to other ad campaigns. These changes can affect sales and make your test results harder to understand. If a change is necessary, apply it equally to both groups.

7. Compare Results and Make a Decision

After the test ends, compare the results from both groups. Check how many extra sales or leads your ads generated and how much each additional conversion cost. Also, review the extra revenue earned from your ad spend. If the results are unclear, collect more data before deciding whether to increase or reduce your budget.

---

How to Test ChatGPT Ads

Use geo holdouts as your main test

1. Pick matched regions (state, DMA, or ZIP groups).
2. Turn ChatGPT ads on in the test regions only.
3. Keep your normal search, social, and email activity the same in all regions.
4. Compare sales between test and control for the full window.

Watch out for these ChatGPT-specific issues

• Reporting is separate from other channels. Performance on ChatGPT sits apart from your Google, Microsoft, and Meta results. You may have to reconcile by hand.
• Click-only views miss people. Someone may see an ad in a chat and buy later on another device. Include view-through and delayed buys in your analysis, but do not let them replace the geo result.
• Targeting rules can change. The product is moving quickly. Re-check your options at the start of each test.
• Partner tools are early. OpenAI itself calls its incrementality work early stage, so ask partners how they handle small sample sizes and how they validate matches.

---

How to Test Google AI Mode Ads

Google AI Mode ads are not a separate campaign type. They appear through existing Search, Shopping, and Performance Max campaigns. Their performance is included in regular advertising reports, so you usually cannot measure AI Mode ads separately.

However, you can test the campaigns and features that may show your ads in AI-powered search results. Here are a few ways to do this:

• Test AI-powered features: Turn on features like AI Max for some campaigns and keep them off for others to compare results.
• Use Conversion Lift studies: Measure how many extra conversions your eligible Google campaigns generate.
• Run a Geo Holdout Test: Show ads in selected locations and keep them off in similar locations. Then compare the results.

A Practical Test Plan

1. Choose Eligible Campaigns: Select the Google Search, Shopping, or Performance Max campaigns that may show ads in AI-powered search results.
2. Use Google's Testing Tools: Run a campaign experiment or Conversion Lift study to compare campaign performance.
3. Keep Brand Search Separate: Track branded search campaigns separately because some customers may already know your brand and plan to buy.
4. Measure Extra Conversions: Compare the additional sales and leads generated by your campaigns instead of relying only on the conversions shown in Google Ads reports.
5. Check Your Own Data: Compare Google Ads results with your CRM, store sales, and GA4 data. This helps you understand whether your campaigns are bringing real customers and additional sales.

---

6 Common Mistakes to Avoid

1. Testing with Too Little Data

Running a test with very few customers or orders can give unreliable results. For example, 12 orders are not enough to clearly understand whether your ads are working. Try to collect enough data before making a decision. More data can help you get a clearer picture of your ad performance.

2. Ending the Test Too Early

Do not stop your test just because the first few results look good. Ad performance can change over time, and early results may not show the full picture. Give your test enough time to collect useful data. Always consider your customers' buying cycle before ending it.

3. Mixing Brand Search Results

Branded searches happen when people search directly for your company or product name. These customers may already know your brand and plan to purchase. If you mix these results with other campaigns, your ads may appear more successful than they really are. Keep brand search campaigns separate when comparing results.

4. Making Changes During the Test

Avoid changing prices, offers, or other marketing campaigns while your test is running. These changes can affect customer behavior and make it difficult to measure the actual impact of your ads. Try to keep other campaign settings the same throughout the test.

5. Depending on Only One Measurement Method

Do not depend only on Google Ads or ChatGPT advertising reports to measure success. Compare platform reports with geo tests, CRM records, GA4, and actual sales data. Using multiple sources helps you understand whether your ads are bringing additional customers and revenue.

6. Not Setting a Decision Rule Before Testing

Decide what results you expect before starting your test. For example, you may want at least a 5% increase in sales before increasing your ad budget. Also, decide what you will do if results are low or unclear. This prevents confusion and helps you make better decisions after the test ends.

---

Tools and Options: Should You Build or Buy?

There are different ways to measure AI ad performance. The right option depends on your advertising budget, team size, and available tools.

There is no single option that works for every business. If you run a small business with a limited advertising budget, you can start with a simple location-based test and track the results in a spreadsheet. However, if your company spends more on ads across different platforms, working with a measurement partner may be a better choice.

At RejoiceHub, we help businesses build data pipelines and automation systems that make this process easier. This includes setting up event tracking, organizing marketing data, creating clear reports, and connecting AI tools with existing business systems through AI integration. The best approach depends on your business needs, budget, and current technology.

---

Conclusion

AI ads give businesses a new way to reach people when they are searching for answers or exploring products. However, measuring their actual results is still a challenge. ChatGPT ads now support location-based targeting, which makes it easier to compare different regions. Google AI Mode ads, on the other hand, are included in regular advertising reports, making their performance harder to track separately.

The best way to measure these ads is to set up proper tracking and run a simple holdout test. Focus on extra sales, leads, and customers instead of relying only on platform reports. Start with one clear goal, compare the results, and use what you learn to improve your advertising budget. This will help you understand which AI ads are worth investing in and which are not.