LINE0:09 · your ad
“Groceries cost more every single week.”
Suburban moms+14
Independents+9
Base Republicans+11
Works across the aisle. Lead with it.
The Republican message lab
Every voter has a tell.
Tell predicts how voters will react to your ad, second by second, before it airs. Then it scores that reaction for every voter on the file.
Built for Republican campaigns, committees and consultants.
Built it
from nothing.
from nothing.
Paid for by Mock PAC
MOCK AD · 0:00.0 / 0:30
Tell ScoreRead · live panelMiddle of the electorate
READ+0.0
TELL SCORE+0.0
grocery line
mugshot montage
See it by audience
The file tells you who to talk to. Tell tells you what to show them.
Campaigns score every voter for support and turnout. Nobody scores how they react to the ad. Tell does, for every spot, on the file you already use.
Every dot is a voter. Lit: the voters this spot moves.
What Tell reads
It reads the spot itself.
Not a script summary or a prompt. Every second of the video: what's on screen, what's said, how it sounds. Then it lines that up against how real voters reacted, second by second, across ten years of Republican testing. The result is a score for any spot, by audience, and for every voter on the file.
PictureWho's on screen, faces, b-roll, on-screen text and cuts.
WordsWhat's said, line by line.
SoundVoice, music, pace and silence.
PICTURE
Candidate to cameraGrocery aisleMugshotsFamily b-roll
WORDS
BioGrocery pricesCrime attackContrastClose
SOUND
Voice onlyTension musicWarm music
REACTION
Example: a mock spot, the middle of the electorate, second by second.
Moments
Find the moment that works.
Every Tell Score pins the reaction to the second. You see which line, image or sound moved which voters, in your spot and in theirs.
IMAGE0:18 · your ad
Mugshot montage
Base Republicans+8
Independents−7
Suburban moms−12
Plays to the base, costs you the suburbs.
IMAGE0:12 · their ad
Capitol riot footage
Independents+7
Suburban moms+9
Base Republicans−21
Their attack lands in the middle. Answer it there.
LINE0:21 · your ad
“I stood up to my own party.”
Independents+6
Swing voters+5
Base Republicans−4
A cheap win with persuadables. Watch the base.
IMAGE0:04 · your ad
Candidate at the kitchen table with family
Suburban moms+12
Seniors 65++6
Men without degrees+2
Open here, not on the podium shot.
SOUND0:24 · your ad
Music turns warm under the close
Seniors 65++9
Independents+5
Base Republicans+3
The close recovers what the attack spent.
Illustration. Made-up moments and numbers, net reaction by audience, to show what a Tell Score surfaces.
How Tell learns
Live panel first. Then the modeled dial on top.
Real voters dial an ad second by second. That’s the Read. Tell predicts the same curve before the panel sees it. Where the two disagree, Tell learns.
Tell Score · predictedRead · live panelA voter turning away
A live panel. Each dot is a voter pressing up or down, second by second.
Illustration of how Tell learns. Every curve here is made up. The test below is real panel data.
From a live test
A real spot, read second by second.
An attack spot from a 2026 battleground Senate test, 92 voters on the dial. The spot and the race stay private while the race is live.
All votersThe other side's voters
Cost-of-living line
“Families are holding” lineFamilies line
War framing
0:000:100:200:30
+19from the other side's voters at the families line. Cost of living landed across the aisle.
−34%in Tell's error on each side's voters once it knows whose side a spot is on.
6.7points a second a 90-voter panel disagrees with itself. Tell waves are built for 300.
A Read: real panel data, scored with G2's v1 method, net percent of voters reacting, per second.
Audiences
Scored for the universes you actually target.
Every live wave is built the same way: one third base Republicans, one third soft Republicans, one third swing voters. Each group is big enough to read on its own, so you see what fires up the base, what brings back the voters who stay home, and what moves the persuadables.
100Base Republicans
100Soft Republicans
100Swing voters
Swing votersSoft partisans and ticket splitters. The persuasion universe.
IndependentsNo party on the file, or a history of voting both ways.
Suburban momsSuburban women with kids at home, the swing bloc in most battlegrounds.
Men without degreesThe working-class men who moved right.
Hispanic votersA bloc that has been moving in recent cycles.
Low-propensity RepublicansRight-leaning voters who skip elections. The turnout universe.
Base RepublicansReliable GOP voters. Primaries, mobilization, money.
Seniors 65+Highest turnout and the heaviest TV viewers.
Under 35Streaming-first and hardest to reach on linear TV.
Rural votersTurnout and margin in small-town counties.
Built from the fields already on the file: party, vote history, age, gender, geography. Bring your own universes. If you can pull it from the file, Tell is built to score it.
Products
Know the spot. Know the voter.
Tell Score: the spot
A modeled M2M curve for any spot, before it airs.
- See the second each universe tunes in or drops off.
- Compare cuts and variants before you pick one.
- Score the spot in the edit room, not the post-mortem.
For media consultants, committees and campaigns
Tell Score: the file
How each voter is likely to react to a given spot, as a column on the file.
- Scored like a support or turnout score.
- Carries what each voter liked, disliked and the message they reacted to.
- Delivered through the voter-file partners you already use, for CTV, digital and mail.
For committees, super PACs, and file and media partners
| Voter ID | Age | Party | Turnout | Support | Their reaction | Liked | Disliked | Impact | Core message | Tell Score | Show them |
|---|---|---|---|---|---|---|---|---|---|---|---|
| V-0418227 | 58 | R | 91 | 88 | Mugshot montage | “I stood up to my own party” | ▲ 0:18 | Crime and safety | A · 74 | Spot A | |
| V-0932815 | 34 | U | 42 | 51 | “Groceries cost more every single week” | Mugshot montage | ▲ 0:09 | Cost of living | B · 66 | Spot B | |
| V-1177460 | 47 | U | 63 | 49 | Kitchen-table open with family | Podium shot | ▲ 0:04 | Family and values | A · 71 | Spot A | |
| V-0270931 | 71 | R | 96 | 94 | Warm music under the close | Fast cuts in the middle | ▲ 0:24 | Character | B · 60 | Spot B | |
| V-1540288 | 29 | D | 37 | 12 | “Groceries cost more every single week” | Mugshot montage | ▼ 0:18 | Cost of living | 35 | Neither | |
| V-0861742 | 52 | U | 78 | 57 | “I stood up to my own party” | Attack voiceover | ▲ 0:21 | Independence | B · 69 | Spot B |
Illustration, made-up voters. For voters on a live panel, every press is recorded and tied to the second it happened. For everyone else on the file, Tell models the same columns.
Checked against Read.
Read is the measured curve from a live G2 panel. Every Tell Score is tested against it, out of sample, before we make a claim.
More than which spot.
Every voter carries what they liked, what they didn't, the moment that hit hardest and the message behind it. The buy gets a plan for who sees which spot, and the next cut knows what to lead with.
Built on real reactions
Trained on reactions, not opinions.
Surveys ask voters what they think. Moment-to-moment testing catches what they do, the second it happens. We've been reading it since Trump, Carson and Fiorina shared a debate stage in 2015. That's what Tell learns from. Tell is still in training, and we say so: every live panel checks its scores and makes the next one sharper.
2015moment-to-moment testing since the GOP primary debates
346political spots scored and archived
1.1Msecond-by-second voter reactions
What no one else can do
Two edges built into Tell.
Reaction to real votes
Reactions tied to who shows up.
Live panel voters are recruited straight from the voter file. After each election, the file shows who turned out. Tell ties every voter's second-by-second reaction to whether they actually voted. Predictive, not causal: which reactions go with showing up.
Every reaction checked against a real vote.
Best of many AI cuts
50 versions in, 3 winners out.
Campaigns now make dozens of ad versions with AI and can't test them all. Tell scores all 50 in a day. The live panel reads only the top three.
Tell sits inside every AI ad shop's workflow.
Who's behind it
Made by people who test these ads for a living.
Tell comes out of G2 Analytics, which has run moment-to-moment ad tests for Republican campaigns and committees for over a decade. G2 still runs the live panels; what they measure is the Read. Tell turns it into a model, the Tell Score.
Calder GageFounder, G2 Analytics
AJ RehbergFounder, G2 Analytics
The vision
Fast like a simulation. Grounded like a panel.
Today most ads air untested. The few that get tested go to a slow, costly panel, or to a simulated audience that nobody checks against real voters. Tell gives you the speed of one and the grounding of the other. Panels don't go away. They become the model's teacher: small, steady waves that keep Tell learning.
- Panels feed the modelLive G2 waves, same battery, every week.
- Tell scores every adIn a day, by audience, with no panel.
- Small panels fine-tune itScores are checked against the Read.
- It keeps learningEvery wave makes the next score sharper.
Partners
Build the 2028 map with us.
The voters we read this cycle are the voters campaigns will need to move in 2028. We're fielding a national base and a layer in each of the seven states most likely to decide 2028, with add-ons for the watch list. Partners in the first wave get first access to the Tell Scores. Pick a state to see its layer.
Back the first wave2028 battleground2028 watch listNational base
Pennsylvania
2028 battleground · Senate seat up in 2028 · 60 spots · 3,000 voters recruited from the file
Open for a founding sponsor. The sponsor gets first access to Pennsylvania's Tell Scores.
Now through Election Day
Back the first wave.
We’re fielding one wave of live research before November: this year’s ads, voters recruited from the voter file, tied to who actually turns out. We’re looking for partners to fund Wave 1 in exchange for equity in Tell.
Interested? Email calder@g2analytics.co
Fielding starts this month. The window is the next two weeks.