Polling after the 2026 midterms: Why the polls missed and how AI could change the game.

Fox - 13 News
14 followers
Nov 15, 2026
Well, retiring pollster Matt Towery is about
to reveal some of the secrets to his success.
He'll explain what works and what doesn't work,
and we'll also show you how the polls missed in
2026, why we knew they were off,
how they underestimated the Democratic margin,
and what we did to adjust.
Let's start with the Florida race for governor.
On average, the polls gave Byron Donalds a four to five point lead,
and our data suggested
something closer. And here's what I said before
any election results came in. On the tracking,
the polling, the modeling, the latest returns,
don't be surprised if this race winds up being
47 point something percent to 48 point something percent,
with Donalds being the victor.
Now, this is what I was seeing behind the scenes when I said
that our A.I. runs were giving us this as the most likely scenario,
a Donald's win by one point one percent.
That was built on our own turnout and independent voter calculations,
pre-election polling on party loyalty and crossover and other data.
And I'll compare that to the actual results.
Donald's won by 1.3%,
so our model missed by just two-tenths of a percentage point.
Now, our scenario did have slightly more voters splitting off
to minor party and fringe candidates
than actually did. But see, we also re-ran it,
forcing more of the splits into one candidate
or the other. And now look at this.
This time it projected a 1.5-point Donald's victory,
Again, missing by just two-tenths of a point.
And the projected vote totals were almost identical
to the actual results, 50.2 to 48.7%
compared with the actual 50.0 to 48.7%.
That is how close our AI modeling came to the actual
results in 2026. And this technology is only getting started.
Once the polling world decides
to get its hands around, how do you harness AI?
And have it represent the way people think,
rather than having to ask what people think.
You can probably create some pretty predictable models
that will be pretty much on target.
In 2026, the top-performing polls in the Florida races for Governor
and Senate were UNF and St. Pete polls.
Polling still matters, even as AI will change the game.
But there are some important points to remember.
The first point, don't scoff at all polls just
because no one has ever polled you.
Remember, Florida has nearly 13.7 million registered voters.
A typical statewide poll interviews may be up to 1,000.
So, your odds of being selected for that poll are about 1 in 13,700.
You're nearly 14 times more likely
to win Florida's pick three with a $1 straight play
than to be selected for a particular statewide poll.
But that's the point of scientific sampling.
Pollsters don't need to interview everybody.
They just need to interview a carefully selected representative sample.
Think about a large container filled with 13.7 million marbles.
Let's say 55% are red and 45% are blue.
Well, if you randomly draw 1,000 marbles,
you get a sample that's likely
to be reasonably close to that 55-45 split.
You don't have to count every marble to estimate the proportions.
And the same principle applies to voters.
But selecting people randomly isn't enough. Pollsters
also have to determine which respondents are actually likely to vote.
And so they look at factors such as their voting history,
their interest in this upcoming election,
their stated intention to vote in it.
Demographic characteristics,
their level of political engagement, and more.
And based on those factors, pollsters decide
which respondents to include in their likely voter sample.
And they may have to contact 10,000 people
to get the 1,000 responses they need.
Now, the second point, don't just look at who's ahead in the race.
Look at where the numbers are headed.
Now, for example, let's say one candidate polls at 54%
and the other 43% two weeks before the election.
In this case, which candidate would you rather be?
Okay, but now let's reveal how
those numbers moved through the prior two weeks.
Now, which candidate would you rather be?
See, a poll's just a snapshot in time,
and like a real horse race, the momentum can shift.
And in a close race, the candidate gaining momentum heading
into Election Day will most likely win.
Here's the third point. It's getting harder
to accurately measure the electorate.
Landlines are disappearing and the people
who still use them do not represent a cross-section
of today's voters.
You're primarily calling primarily senior voters and your universe
is shrinking every day. Meanwhile,
people rarely answer cell phone calls from numbers they don't
recognize and people who opt in to receiving calls
from pollsters tend to be more engaged in politics.
That's no longer a random digit call.
And by law, with cell phone polling, it can't just be a robocall.
A live person has to initiate the call.
And a lot of people are hesitant to tell
a real person what they really think these days.
And pollsters see that in some states more than others.
Places like New Hampshire and Vermont.
And Iowa, some of these Midwestern states,
they really don't like talking about their politics to strangers.
And in 2016, 2020 and in 2024,
that's where surveys underestimated President Trump's support.
People didn't want to say they were voting for Trump.
They felt like somehow that they were going to be on a target list.
People come over and dox them.
Not that that's true, but that's how people think.
And that prompted pollsters to make adjustments
to ensure that Republican voters were adequately
represented in their samples. And then in 2026, they overcorrected.
They placed too much emphasis
on capturing the Republican voters they previously missed,
while Democratic voters were more
interested in the election, more politically engaged this time around,
and more likely to turn out.
Republicans were hard to get in this cycle to get them to answer.
We thought it was because
of the typical, we're afraid to talk.
In the end, we should have taken it one step further
and figured out that we don't care.
And that's really what the answer was in Texas and Georgia
and so many places.
Republicans who voted in 2024 simply did not turn back out.
Plus, exit polls show some voters
who supported Trump in 2024 shifted toward the Democrats.
Now, polls captured some of that movement, but underestimated
its magnitude in several states like Georgia and Texas.
A lot of us got the fact that this was a Democratic wave.
The question is whether we got that it was a tsunami.
And it all suggests the polling industry still has a challenge,
accurately measuring the composition
and the intensity of the electorate.
Now it's easy to poll, easier at least
for us to poll presidential elections,
but we've done very, very well.
It's because you have a massive turnout on both sides. But
when you get into things like trying
to measure intensity among various groups in a midterm,
I think the whole industry is going to have to advance itself.
Well, the industry is already adapting using
a mix of methods to reach voters.
More pollsters, for example, are relying on panels,
although those come with some of the same problems.
Panels are created by large companies that find individuals
who are willing to take polls.
And these people are made aware of the fact
that they will be asked to take surveys.
In most of these instances, they are compensated,
and that can lead to a hyperpolitical, as
I've said before, group that knows more about politics
and therefore may give you an answer
that's a little more informed than the average bear.
That isn't always the best way to do things.
And that's why pollsters are also relying more on text messages.
Last year, Matt told us that's been a secret to his recent success.
There's a certain type of text that people will respond to,
but it has to be the exact type.
And most pollsters don't know what that is,
and I'm not going to tell anybody.
Well, guess what? Again, Matt is now retiring,
and he's finally willing to share.
And people are very,
very happy to respond to a text poll as long as they're not hitting
a link, because once they hit that link,
they don't know if it's, you know, what foreign
country or somebody trying to get into their bank account.
So it provides a question in the text
and the viewer responds either Jolly or Donald's?
Yeah, the whole thing is just scrollable and you just choose.
And then that's it.
You took us behind the curtain and
just revealed the secret, didn't you?
Well, there's another secret to it.
And that is that a lot, most pollsters,
if you look at these polls that come in, they've got 60 questions.
Now, anyone is going to take a 60-question poll.
And so now the closing points,
don't assume that 2026's
polling myths will repeat.
Pollsters are going to be watching younger voters
and Hispanic voters and other parts of Trump's coalition
to see whether their shift toward the Democrats lasts.
And the next race for the White House
is already generating early polls.
But until voters have a reason to pay attention, those numbers will
not tell us much about how the election is going to turn out.
They can tell us where a race starts, not where it will finish.
