As of 13 August 2026, AI can only partly run A/B tests for your paid ads.

What to actually do

1. Use a tool built for this

   The route this page recommends

2. Hand it to a person

   Second choice

   A person who owns the outcome does this end to end, worth it when the failure is dear.

3. Do it yourself

   The distant third

   A chat interface, power-user skill, and roughly 1 hour until you can act on the result.

   How to actually do it

   1. Open the advertising platform and record the campaign objective, audience, location, placements, conversion action, budget, dates and current tracking setup.
   2. Gather the existing ad copy, creative files, landing-page URL, brand rules and any policy restrictions, then paste the non-sensitive details into the prompt.
   3. Ask the chatbot for the test plan and two variants, and compare every factual claim, offer detail and destination URL with your current campaign and website.
   4. Duplicate or edit the campaign in the advertising platform so the two variants differ only in the variable named by the plan, with the audience, placements, conversion event and other settings held constant.
   5. Check the preview, tracking event, destination links, exclusions, consent settings and budget allocation in the platform before publishing, then approve the ads yourself.
   6. Export or record the platform results for each variant, including spend, impressions, clicks and conversions, and paste the figures back into the chatbot without asking it to invent missing data.
   7. Compare the chatbot's interpretation with the platform report and your business records, then pause, continue or adopt a variant only after checking that delivery was comparable and the decision fits your commercial judgement.

   Prompt

   Act as a paid advertising experiment planner, not as someone with access to my ad account. Create a practical A/B test plan using the information below.
   
   Business and offer: [describe the business and offer]
   Advertising platform: [platform]
   Campaign objective: [sales, leads, app installs or other]
   Target audience and location: [details]
   Landing page or conversion action: [details]
   Current ad variants: [paste the copy, images or video descriptions]
   Available budget and test period: [details]
   Current results, if available: [paste impressions, reach, clicks, spend, conversions and revenue]
   Rules or claims that must be followed: [details]
   
   Return:
   1. One testable hypothesis.
   2. The single variable to change and the variables to hold constant.
   3. Two clearly labelled ad variants.
   4. A campaign setup checklist for the platform, including audience, placement, budget split, naming and conversion tracking.
   5. The primary metric, supporting metrics and a stopping rule stated without inventing thresholds or results.
   6. A table I can use to record results.
   7. A plain-English method for comparing the variants and identifying common sources of bias.
   8. A list of checks I must complete in the advertising platform before launch and after the test.
   
   Do not claim that you ran the test, accessed my account or saw live results. Do not invent performance data, statistical certainty, legal requirements or budget figures. Flag any missing information that prevents a fair comparison.

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.