chatgpt ads roi
What Can Advertisers Do About Low-ROI AI Chat Ads?
What is the direct answer?
Check the match before you buy the click. A study by Christian Schulze and Maximilian Kaiser found that paid ChatGPT visits return far less than paid search visits, but it does not say why. One thing you control is whether your ad uses the words your buyers type. Prism tests that for free, in its own engine.
Does the study explain why paid ChatGPT clicks earn less?
No. The paid-ad part is descriptive: advertisers chose to buy, so the comparison with search is not a controlled experiment. In their discussion the authors offer one reading, that ads sit apart from the answer, which protects the product but may limit what a click is worth. The study measures the gap, it does not diagnose it. The numbers are in do ChatGPT ads pay off for advertisers?
Which parts of the result can an advertiser influence?
Four: the price you bid, the budget you cap, the tracking you set up, and how well your ad fits the questions people ask. The first three are settings in the ad platform. The fourth is wording, and it is the one you can test before spending a dollar.
What does a Prism test show?
Prism compares the words in a question with the words in your ad's title, description and keywords and gives a score from 0 to 1. A labeled card shows only when the score is 0.65 or higher. Below that, the answer shows alone and no other ad fills the space. A report lists the score for each question you tried. A miss in a test is a valid result: it shows which questions your ad does not speak to. See how ad matching works for the scoring rules and the sample match reports for what a report looks like.
What can a Prism test not tell you?
| Question | Can a Prism test answer it? |
|---|---|
| Does my ad fit this question in Prism's engine? | Yes, with a score |
| Which words in the question does my ad lack? | Yes |
| Will ChatGPT show my ad on this question? | No |
| Will the clicks turn into sales? | No |
Prism has no click, sales or conversion data from ChatGPT, and a score here does not predict what ChatGPT will do with the same ad. Prism does not write or rewrite ad copy either: you bring your own ad. A match score is a relevance check, not a return forecast.
How do you run a useful check before spending?
- Write 10 to 20 questions your buyers really ask.
- Run each one against your ad on Prism.
- Read the report for words in those questions that your ad never uses.
- Edit your own wording, then run the same questions again.
- Only then set a capped test on the platform you chose, with tagged links and a sales goal to judge it by.
For the longer version of this approach, see test ads against live questions before you spend.
Frequently asked questions
Does a high Prism score mean my ChatGPT ad will earn more?
No. The score describes Prism's own engine only. It shows your ad uses your buyers' words. It says nothing about conversion, price or how ChatGPT treats the ad.
Is testing an ad on Prism free?
Yes, testing an ad against questions on Prism is free at the time of writing.
Where do these facts come from?
- Kaiser and Schulze, Can Advertising Fund Free AI? Evidence from ChatGPT's International Advertising Rollout, working paper v1, October 2026
- ChatGPT ad visits earn shops $0.89 against $2.45 for Google search ads, PPC Land, November 2, 2026
- How ad matching works, Prism Publication
Figures checked November 3, 2026. The paper is a working paper and may be revised.
AI-written, human-reviewed.
Find out if your ad fits before you pay
Prism shows a labeled ad in an AI chat only when it fits the question. Test your own ad against real buyer questions, free, and see the score for each one.