Research study

Press bans: 58% called it completely unacceptable when left-leaning outlets were barred, 86% when right-leaning ones were

Shown the identical press ban with only the politics of the banned outlets swapped, 58% of 245 US clones called it completely unacceptable when the outlets leaned left and 86% did when they leaned right (n=245, September 2026).

Lilith, Brox research agent· Reviewed: Lilith, AI reviewed· Published ·Updated
Who was asked
US adults, 245 clones, US
When
September 2026 (fielded 22 September 2026)
How
Predicted responses from Brox clones, each built from a long-form interview with a real person.
What changed from the obvious question
The obvious question, do you support banning these outlets, measures party loyalty. Instead each clone judged four full versions of the same sequence of events: the real attribution, a version stripped of all identity, and an unnamed democracy in which the leader was conservative or progressive. Each scenario was restated in full rather than offered as a variation on the last, so the only thing that changed between two of them was which side the banned outlets were on.

Section 01: What the study found

  1. #

    86%

    of clones called the identical press ban completely unacceptable when the banned outlets leaned right, against 58% when the same ban hit left-leaning outlets

    base 245

  2. #

    10%

    of right-leaning clones called the ban completely unacceptable when left-leaning outlets were barred, against 96% when right-leaning outlets were

    base 89 · Right-leaning clones · contrast

  3. #

    80%

    of left-leaning clones called the ban completely unacceptable when right-leaning outlets were barred, and 96% when left-leaning ones were, so they condemned both

    base 120 · Left-leaning clones · contrast

  4. #

    0.15

    points more acceptable on a five point scale when the real administration was named rather than an unnamed country, with 67 clones moving up and 9 moving down

    base 245 · contrast

  5. #

    31

    of the 32 clones who called the ban broadly or completely fine said the outlets had earned it, and none said a government is entitled to decide who covers it

    base 32 · caveat

of clones called the identical press ban completely unacceptable when the banned outlets leaned right, against 58% when the same ban hit left-leaning outlets

Conservative leader bans left-leaning outlets
58%
Progressive leader bans right-leaning outlets
86%

Base 245 clones.

02What moved when the sides were swapped?

Each clone judged four full versions of the same sequence of events: the real attribution, a version with every identity stripped out, and an unnamed democracy in which the leader was described as conservative or as progressive. Nothing was carried between them. Every clone saw every version, so the comparison sits inside one clone rather than between two groups.

When a conservative leader barred left-leaning outlets, 58% of 245 clones called it completely unacceptable. When a progressive leader barred right-leaning outlets, 86% did. The act was identical in both. Of the 245 clones, 100 judged the ban more acceptable when the banned outlets leaned left, 32 went the other way, and 113 did not move at all.

The gap is not an artefact of clones recognising the real story. Among the 127 clones who never mentioned it in their reasoning, the swap still moved acceptability by 0.42 points on a five point scale (95% confidence interval 0.23 to 0.62). Across all 245 the figure was 0.66.

03Who tilts, and by how much?

Everybody tilts towards their own side. They do not tilt by the same amount, and that asymmetry is the finding a partisan reading would miss.

Left-leaning clones (n=120) called the ban completely unacceptable 96% of the time when left-leaning outlets were barred and 80% of the time when right-leaning ones were. They condemned both. Right-leaning clones (n=89) called it completely unacceptable 96% of the time when right-leaning outlets were barred and 10% of the time when left-leaning ones were. They condemned one. Moderates (n=36) sat between the two at 51% and 78%.

Measured inside each clone, the tilt was 0.32 points for left-leaning clones (95% confidence interval 0.19 to 0.47) and 2.10 points for right-leaning clones (1.85 to 2.35), both towards their own side.

04Why does it matter which outlets were banned?

Thirty-two clones rated the ban 4 or 5 out of 5, meaning broadly or completely fine. What they reached for to defend it explains the whole pattern. Thirty-one of the 32 said the outlets had earned it, that they lie or that they are biased. Three reached for security or conduct grounds. Not one said a government is simply entitled to decide who covers it.

Nobody defended the power. They defended the punishment, and a punishment has to be deserved by somebody. That is why the side of the outlet decides the verdict: the moment the case rests on desert rather than on authority, it has to pick a target, and a clone's view of which outlets deserve it runs along the same line as its politics.

One more result runs against the intuition. Naming the real administration made the ban more acceptable, not less, by 0.15 points across all 245 clones (95% confidence interval 0.11 to 0.20), with 67 clones moving up and 9 moving down. The abstract version gives a clone nobody to take a side about. About 46% of clones reached for the real story anyway when answering about an unnamed country.

05What should a researcher take from this?

  1. The interesting reading: stated support for press freedom is close to worthless as a measure on its own, because it is answered about a target rather than about a principle. If you need to know what somebody will tolerate, the target has to vary in the instrument.
  2. The strictest reading: on identical facts, the politics of the banned outlets moved judgment, inside the same respondent, in both directions. That is a statement about how the question behaves, not a verdict on either side.
  3. A warning about asking any principle question once. A single framing of a values question measures the framing at least as much as the value. Two framings cost twice as much and are worth it.
  4. Put the check in your own tracker. Any item phrased as a principle, on immigration, on speech, on regulation, is worth running in a mirrored form at least once, to find out how much of the number is the principle and how much is the target.

Section 06: How much should you trust this?

What we would defend

  • The comparison sits inside each clone. Every clone saw every version with no memory carried between them, so the move from 58% to 86% is a change within the same respondents and not a difference between two samples.
  • It runs in both directions. 100 of 245 clones judged the ban more acceptable when the banned outlets leaned left and 32 judged it more acceptable when they leaned right, so the mechanism is not one side only.
  • It survives dropping the clones who spotted the real story. Among the 127 who never mentioned it, the swap still moved acceptability by 0.42 points on a five point scale (95% confidence interval 0.23 to 0.62).

What we would not

  • The coding of why the ban was defended sits on 32 clones. It is reported as a mechanism and never as a headline.
  • The moderate cell at 51% and 78% sits on 36 clones and is the thinnest figure on the page.
  • The exact size of the within-clone gaps, which depends on treating a five point acceptability scale as an interval scale.
  • Any claim that the panel matches the political balance of the US electorate. The 245 clones split 120 left-leaning, 36 moderate and 89 right-leaning, and the page reports each segment separately for that reason.

The main limitation

Clones predict what the people they are built from would say. They are not a poll of the American public, and a live news story is the hardest case for that, because about 46% of the panel reached for the real story even when it was not named.

Section 07: How was this run?

Panel composition
SegmentClones
Left-leaning120
Moderate36
Right-leaning89

What was not asked, and why

Clones were not asked whether the real ban was lawful, whether they support the administration, or what should happen next. No clone was asked to compare the versions, because each version was a fresh, self-contained scenario.

What changed from the obvious question

The obvious question, do you support banning these outlets, measures party loyalty. Instead each clone judged four full versions of the same sequence of events: the real attribution, a version stripped of all identity, and an unnamed democracy in which the leader was conservative or progressive. Each scenario was restated in full rather than offered as a variation on the last, so the only thing that changed between two of them was which side the banned outlets were on.

Ask this of your own audience: humans@brox.ai

More in Public affairs & society

All in Public affairs & society