Campaign strategy research · Randomized experiments

Message and ad testing

Message tests split the sample at random and show each group a different argument, contrast or ad script, then measure the change in support. You see what persuades, who it persuades and what it costs you elsewhere, before the money is spent.

When to use it

  • Before a paid media buy, to choose the message and the audience
  • When an attack lands and you need to know whether to answer it
  • Before a mail or digital program, to pick the strongest contrast
  • When the board, the donors or the consultants disagree on the message

What you get

  • The effect of each message on the ballot
  • Results among persuadable and base voters
  • Backfire and intensity checks
  • A ranked list of messages with a memo
  • Script and ad copy notes for your creative team

How the work is done

Each respondent is randomly assigned to a control group or to one message or script, and the effect is the difference in support between groups. Because assignment is random, the effect is causal, not a correlation. We report the effect overall and among the persuadable groups you care about, with the uncertainty stated.

Read the full methodology

How a project runs

  1. 1

    Agree the messages and the outcome

  2. 2

    Randomize and field

  3. 3

    Estimate effects by group

  4. 4

    Rank and recommend

Questions

What is an informed ballot test?

Voters answer the ballot, read an argument, then answer again. The change shows how much that argument moves them, compared with a group that read nothing.

Can you test video ads?

Yes. Scripts and finished video can be shown inside the survey, with each respondent seeing one version at random.

How much does message and ad testing cost?

Every project is quoted in writing on its geography, sample size, length and timeline before any work starts. Programs cost less per wave than separate studies. Send the scope and you will have a quote the same day.

Is our work kept confidential?

Yes. Client work is never shared with another client, and nothing is published unless you choose to release it, in which case the sponsor is named.

  • D +13.1Our final 2026 national poll, against a D +14.0 count
  • 2 of 42Final national polls, ranked by distance from the count
  • 7 of 8Senate winners called in our final state polls
  • 8 statesAnd the nation polled in 2026, every poll scored in public
The 2026 accuracy report