To run an A/B test experiment in Google Ads with Claude or ChatGPT, connect your Google Ads account to PaidSync.ai and ask the assistant to create an experiment on a campaign, which splits traffic between the original and a variant so you can compare them. You choose the control campaign, the split and the dates. Starting, applying or deleting an experiment needs your confirmation.
A campaign experiment tests one change against the original on a share of traffic. Three tools cover its life:
create_experiment creates an experiment from a control campaign, with a traffic split that defaults to 50 percent and a type of SEARCH_CUSTOM or DISPLAY_CUSTOM. It needs confirmation.list_experiments lists the experiments with their status, dates and traffic split.update_experiment changes an experiment's status. INITIATED starts it, GRADUATED applies the winner, and REMOVED deletes it. It needs confirmation.list_experiments only reads your account. create_experiment and update_experiment change it.
| Tool | What it does |
|---|---|
| create_experiment | Sets up the experiment |
| list_experiments | Shows experiments and their status |
| update_experiment | Starts, graduates or removes the experiment |
Test one change at a time, or you will not know what moved the result. Run it long enough to cover your conversion lag. Graduating applies the experiment arm to the original, so confirm only once the data is clear.
These steps use Claude to run an A/B test experiment. PaidSync.ai also works in ChatGPT, and the prompts are the same.
Create a free account at paidsync.ai and approve Google Ads on Google's own sign-in screen. You do not need a developer token or a Google Cloud project, only a login with access to the account where you will run an A/B test experiment.
Open Claude's connector directory, add PaidSync.ai and sign in. Ask "list my Google Ads accounts", then pick the account where you will run an A/B test experiment.
Write down the single change you want to test, such as a new bidding strategy, and the campaign that will act as control.
Ask the assistant to create the experiment with your split and dates. It returns a preview and needs your confirmation.
Ask it to start the experiment. When the data is in, compare results and either graduate the winner or remove the experiment.
List the experiments to see the status, and compare both arms on cost per conversion after the test window. Do not graduate on a few days of data.
Use these two prompts to run an A/B test experiment, in this order.
list_experiments only reads your account, so the lookup before you run an A/B test experiment changes nothing.create_experiment and update_experiment run as a preview by default, so you see the change before it is applied.create_experiment and update_experiment need your explicit confirmation, and that applies when you run an A/B test experiment.create_experiment and update_experiment run.create_experiment and update_experiment can show a preview but not apply it. Tokens are encrypted at rest.create_experiment and update_experiment are applied, the change is logged, on a best-effort basis, with a before-and-after result.Every Google Ads tool PaidSync.ai offers is on the Google Ads MCP server page, including the tools used to run an A/B test experiment.
In Google Ads, open Experiments, create a custom experiment from a campaign and set the split. With PaidSync.ai, ask Claude or ChatGPT to create and start it, and each step needs your confirmation.
It applies the winning arm to the original campaign. Confirm it only after the comparison is clear.
Yes. PaidSync.ai connects Google Ads to ChatGPT and to Claude, and the same tools run in both, so you can ask it to run an A/B test experiment the same way you would in Claude.
A connected Google Ads account, the control campaign, a name, and optionally a traffic split and dates.
PaidSync.ai is free for 15 tasks a month with no card, enough to run an A/B test experiment and see the result. Pro starts at $99 a month. See pricing.
PaidSync.ai runs 630+ tools across 14 ad and analytics platforms. Connect your accounts once and use them in the AI assistant you already work in.
Get Started Free Book a Demo