The Neutrality Initiative fails in our synthetic Switzerland.
We asked 5,220 Swiss voters how they will vote on 27 September. None of them is a real person. Each one was built from official statistics and given the campaign someone like them would actually have met: what they read online and in the papers, and also what they pass in the street, since posters and the campaign paper in the letterbox reach plenty of people who never click on anything. They answered the ballot question in their own language. Among voters who have made up their minds, 35% say Yes, below every published poll and on the side the trend has been moving.
The September poll landed on our number
We locked our forecast on 15 September. At that point every published poll was from August, and all of them sat higher than our number. The following morning gfs.bern put out its second SRG poll, at 35.1% Yes among decided voters. We had also run the whole study a second time on an unrelated language model, which came out at 34.3%. That run is the short tick on our bar.
Party by party we match the polls. Ticino is the exception.
Purple bars are our forecast. The dark ticks are published polls, the grey blocks show the range where several polls disagree, and the short tick on each bar is where the second model came out. By party we end up more or less where the polls do. The place we genuinely disagree is Ticino, where the polls see a Yes majority among Italian speakers and our voters do not.
Yes share by the party voted in 2023
By language region
By education
By age
One open disagreement
In the polls, voters over 65 are the group most critical of the initiative. In our population they are the group most in favour of it. Every age group still lands on the No side, so this changes the shape of the result without changing which way it goes. The VOTO survey after the vote should tell us who was closer.
Every voter meets the real campaign, screen and street
Since December we have tracked the campaign as it happened, collecting 1,193 dated items from 428 sources in German, French and Italian. Each voter then gets what someone like them would plausibly have seen. A pensioner in Ticino reads the Corriere del Ticino. An apprentice in Lausanne scrolls TikTok and the comments under 20 minutes. Both of them also walk past the posters their canton is covered in and find the same two leaflets in the letterbox.
Where the campaign ran, channel by channel
Number of items, scaled to the largest channel: 676 articles from news sites and newspapers.
Three voters, one week each
Real headlines, newest first, as they were handed to three of our voters.
A 24-year-old in Lausanne, apprenticeship, news from social media
Pour un OUI de gauche à la neutralité - partie 3 — «CONTRE LES SANCTIONS OCCIDENTALES ILLÉGITIMES VOTONS OUI…
«Neutralité suisse en danger ? Décryptage de l'initiative populaire du 27 septembre» — À l'approche des…
Votation en Suisse : Helmut Scheben démolit ses adversaires - La neutralité suisse a des répercussions…
Article: «Votations du 27 septembre : les Suisses se prononcent sur deux initiatives populaires» (20…
Attendre que le mal soit fait avant d'agir ? Les cyberattaques et les autres nouvelles menaces ne s'arrêtent…
A 68-year-old in Ticino, newspaper reader, votes Mitte
Telegiornale — «Le tappe principali della storia della nostra neutralità». Explainer clip broadcast on 12…
Vogliamo i nostri figli e nipoti sul fronte Orientale? — «Si, Ho scelto un titolo forte, ma non sembra…
Modem — «Neutrali, ma quanto?» Con l'invasione russa in Ucraina del 2022, la Svizzera è stata confrontata…
La neutralità tra sanzioni e flessibilità
SÌ alla neutralità: a Lugano una serata trasversale, dalla Lega ai comunisti
A 41-year-old commuter from Aargau, radio and podcasts, SVP in 2023
Katholische Kirche wendet sich gegen Neutralitätsinitiative – nun protestieren Gläubige
Amnesty International: Fake News gegen die Neutralität
Die Schweizer Neutralität schützt die Bevölkerung vor Krieg, den die Politiker führen
«Kein Krieg. Ja zur Schweizer Neutralität!»: Christoph Blocher und Roger Köppel im Gespräch
NEIN = NATO-Beitritt? Nope, Lüge. #neutralitätsinitiative #svp #abstimmung #nato — description: «E witeri…
The street and the letterbox
Plenty of this campaign never reaches anyone through a screen. The Yes side declared 3.84 million francs to the Federal Audit Office and the No side 1.67 million, and the two camps spent it in very different places. Yes posters fill stations and trains across the country. No posters go up mainly where the parties opposing the initiative are strong.
So every voter in the study also walks past whatever their own canton is papered with, and finds two things in the letterbox: the Federal Council’s booklet recommending No, and the Yes campaign’s newspaper, which went to more than four million households.
One person at a time
Ask a language model how Switzerland will vote and you get the average of everything it has read, which is a plausible enough number with nobody standing behind it. We never put that question to it. Instead we hand it one person, with that person’s life and that person’s campaign, and ask what they will do with their ballot. Then we do that ten thousand times over.
A real electorate
Every voter carries the attributes that tend to decide Swiss votes: canton and language, age, education, work and income, the party they voted for in 2023, how closely they follow politics, how far they trust institutions, and where they get their news. The composition is taken from the official statistics rather than left to the model to imagine.
The question as it is printed
Each voter gets the question exactly as it appears on the ballot paper, word for word, put to them as if the vote were held today. Nothing in the wording leads them, and they never see a poll number. After the answer we ask how firm it is and why, plus a control question that catches a voter who has drifted away from the person they were built to be.
Uncertainty stays in, and the study runs twice
Nobody is a hundred percent certain about anything, so every answer is recorded as a probability. A voter who sits at seventy percent Yes counts as seven tenths of a vote. We then run the whole study again on an unrelated language model. Nationally, and for every party, the two runs agree to within a point.
Locked before the vote, scored in public afterwards
We locked and fingerprinted the forecast on 15 September, before the last polls came out, and we will score it in public against the official result, canton by canton. Where we get it wrong, that will be on the page as well.
What happens on 27 September
We will put the official result next to this page, nationally and canton by canton, so anyone can see how far off we were.
| Quantity | Forecast | Compared against |
|---|---|---|
| Yes among decided voters, national | 35.1 | Official Yes share of valid votes |
| Turnout | 47.8 | Official turnout |
| Yes by language region | 37.5 / 26.5 / 30.8 | Official cantonal results, grouped by language |
| Yes by party, education and age | as charted | The post-vote VOTO survey |
What this is for
A vote is about the hardest test a forecast can take, because it comes down to one number on one date with nothing to hide behind afterwards. The same synthetic population answers the kind of question organisations normally pay a panel for: how a message lands with women of 25 in Romandie, which claim moves rural SVP voters, what a change of campaign does to a brand or a policy. It answers that in minutes, at a fraction of what a survey pilot costs.
Forecast published 16 September 2026 and locked on 15 September. No personal data was used or produced.
Want a forecast like this for your market?
Foresenta builds synthetic populations and simulates how they will react to products, campaigns, prices, and information environments, before you commit.