Ad concept testing
Pick which ad to test with real money
Before you put budget behind an ad, you usually have several headlines, offers or photos and no clear favorite. SellTo.us shows them side by side to the same AI-simulated shoppers, so you can narrow the list and start a real ad test with the strongest candidates. The simulated result is a shortlist, not a prediction of clicks or sales.
Test Your Idea for Free1,000 AI-simulated participants · no card needed · free if eligible
Choose Advertising idea when you set up the test. Each simulated person then answers whether they would respond to the ad (stop, read it and click or act on the offer), maybe, or ignore it and scroll past, with a short reason. The report judges the message itself: is it clear, is it relevant to this audience, and what would make people ignore it.
What you can compare
Headlines and copy
Up to 5 versions of the text a shopper would read.
Offers and prices
A discount, a bundle or a different price, as separate versions.
Images
Up to 5 images. Every simulated person picks the one that most makes them want to respond.
Audiences
Results by shopper group show which message works for whom.
Set up a comparison you can learn from
- Change one thing per version. If version B has a new headline and a lower price, a better result can’t tell you which change mattered. Reports point this out and suggest a follow-up that changes one thing at a time.
- Keep claims honest. The report lists claims in your copy, like “clinically proven” or “lasts 30 days”, as yours to verify, and it never suggests adding new ones. Ad platforms and consumer law expect you to be able to back them up.
- Don’t read too much into small gaps. Each test is one simulated run; treat a few points of difference as a tie and test both for real.
Example: a clean set of three versions
Say you sell a refillable water bottle and want to know which angle to lead with. A set you can learn from keeps the photo, price and product the same and changes only the main message:
| Version | What changes | Question it answers |
|---|---|---|
| A | “Keeps drinks cold for 24 hours” | Does a practical benefit pull people in? |
| B | “Replace 300 plastic bottles a year” | Does an environmental reason work better, and for whom? |
| C | “Fits every car cup holder” | Does a specific everyday detail beat a big promise? |
Because only the message differs, the report can compare them directly, and the results by shopper group show whether one angle wins overall or only with one audience. Both numbers in A and B are claims you’d need to be able to prove before running the real ad. Once you’ve picked a message, a second test can compare photos with that message fixed.
The ad targeting guide
10,000 and 100,000-participant tests include a guide for Meta (Instagram and Facebook), TikTok and Google: suggested audiences with ages, interests and placements, the message and photo to try with each, how to split the budget and search keywords. Everything in it is a hypothesis for a real test. It uses one example budget throughout, which you scale to your own, and it doesn’t predict click-through or conversion rates.
From shortlist to real test
- Run the AI test with your candidate ads and note the top one or two for your most interested group.
- Launch a small paid test with those candidates to that audience, with a capped daily budget for a few days.
- Compare clicks and add-to-carts, not likes, and keep the winner.
- If you change the ad, re-test it against the old one on the same simulated shoppers before spending again.
What it can’t tell you
- No click, conversion or ROAS predictions. Real results also depend on the ad platform’s delivery, bidding, placement and timing, none of which a simulated shopper sees.
- The audience suggestions are hypotheses. Interest names and sizes differ between platforms; check them in the ad tool before you launch.
- No accuracy is claimed. Results haven’t been measured against real sales, and each test is run once, so how much the numbers would move on another run isn’t measured. More on the limits.
- Claims in your ads aren’t verified. The report flags them; you need your own evidence before running them.
Compare your ad ideas
Your first 1,000-participant test is free if eligible. No card needed. A 1,000-participant test takes about 10 to 12 minutes.