← Politicians and parties

Why the manifestos are photographs, and how we made them searchable

Of the 31 manifestos filed for the 2022 general election, only 7 contain text: the rest are photographs of printed pages, which a screen reader cannot read and which nothing can search. We had a machine read them, so from here 30 of the 31 can be searched.

Search for a word in the manifestos →

Can I search for a word inside an electoral manifesto?

From here yes; inside the ministry’s PDF, almost never. The search box on the manifestos page searches 30 of the 31 manifestos filed for the 2022 general election, and every result says which page of the original document it sits on.

In the ministry’s file there is usually nothing to search: 24 of the 31 are images — a photograph or scan of a printed sheet — and the words you can see do not exist as words to a computer, any more than they do in a photograph of a road sign. This is not a limitation of your PDF reader; those words are not in there.

Source: The filed manifestos, on the ministry’s site

How did you make them searchable?

We had optical character recognition read the scans, page by page, and we keep the text together with the page each word came from.

Seven manifestos already contained text and are read exactly as filed; 23 are scans read by machine, and it says so wherever they appear. The ministry’s PDF remains the only authentic version: we do not republish it, we do not summarise, and every passage points at the exact page of the original.

Can the machine-read text be trusted?

Up to a point, which is why the site always says where a sentence came from.

Character recognition makes mistakes, especially on a skewed or faded photocopy. A passage in which even one word was read with low confidence is not shown: in its place you get the page of the ministry’s PDF, so you know the manifesto covers the subject without our putting words in a party’s mouth that may not be its own.

Which manifestos already contained text?

Seven: Italexit, Azione – Italia Viva, Alleanza Verdi e Sinistra, Movimento 5 Stelle, Italia Sovrana e Popolare, Alternativa per l’Italia and Forza Nuova.

The manifestos of Fratelli d’Italia, the Partito Democratico, the Lega and Forza Italia are all scans: their text can be read here, but only because a machine reconstructed it. The dividing line is not between large and small parties, nor between right and left — it is between filing an exported document and filing a photocopy.

The filed manifestos, from the most-voted list to the least

Each row is a manifesto filed for the 2022 general election, ordered by the result its list went on to get in the Chamber. The mark says whether the file contained any text.

The bar is the share of valid Chamber votes the list took in 2022, on a scale that runs to 100 %. It does not measure the manifesto: it is there to show that the line between text and photograph is not the line between large and small.

Fratelli d'Italia con Giorgia Meloni26%

a photograph, read by machine

Partito Democratico - Italia Democratica e Progressista19.1%

a photograph, read by machine

Movimento 5 Stelle15.4%

text inside the file

Lega per Salvini Premier8.8%

a photograph, read by machine

Forza Italia8.1%

a photograph, read by machine

Azione - Italia Viva - Calenda7.8%

text inside the file

Alleanza Verdi e Sinistra3.6%

text inside the file

+Europa2.8%

a photograph, read by machine

Italexit per l'Italia1.9%

text inside the file

Unione Popolare con De Magistris1.4%

a photograph, read by machine

Italia Sovrana e Popolare1.2%

text inside the file

Noi Moderati / Lupi - Toti - Brugnaro - UDC0.9%

a photograph, read by machine

Sud Chiama Nord0.8%

a photograph, read by machine

Vita0.7%

a photograph, read by machine

Impegno Civico Luigi di Maio - Centro Democratico0.6%

a photograph, read by machine

Südtiroler Volkspartei (SVP) - PATT0.4%

a photograph, read by machine

Mastella Noi di Centro Europeisti0.2%

a photograph, read by machine

Alternativa per l'Italia - No Green Pass0.1%

text inside the file

Partito Animalista - Ucdl - 10 Volte Meglio0.1%

a photograph, read by machine

Partito Comunista Italiano0.1%

a photograph, read by machine

Forza del Popolo0%

a photograph, read by machine

Free0%

a photograph, read by machine

Partito della Follia Creativa0%

a photograph, read by machine

share of valid Chamber votes, 2022

Filed without a national figure in 2022

These symbols filed a manifesto but do not appear with a national figure in the 2022 Chamber result. They are listed alphabetically.

  • Destre Unite—

    a photograph, read by machine

  • Forza Nuova - Apf—

    text inside the file

  • Gilet Arancioni - Unione Cattolica Italiana—

    a photograph, read by machine

  • Italia dei Diritti - De Pierro—

    a photograph, read by machine

  • L'Italia del Meridione—

    a photograph, read by machine

  • Partito Comunista dei Lavoratori—

    no text, not even here

  • Pensiero e Azione Ppa—

    a photograph, read by machine

  • Referendum e Democrazia—

    a photograph, read by machine

The dividing line is not between large and small parties, nor between right and left: it is between exporting a document and printing sheets and running them through a scanner. The two largest lists by vote are photographs; the third is not; the fourth and fifth are.

The bar says how many votes the list took, not how long or how good the manifesto is. It is there for the order of magnitude alone.

Source — all the sources this site uses: Ministero dell’Interno — Elezioni trasparenti (dait.interno.gov.it) — opens in a new tab · Retrieved ; Ministero dell’Interno — Eligendo (elezionistorico.interno.gov.it) — opens in a new tab · Retrieved

Is this a deliberate attempt to make manifestos harder to read?

Almost certainly not. It is how the file was produced, not a decision about how it should be read.

The files say so themselves. Among the scanned manifestos, the producing software is recorded as a Xerox AltaLink, a RICOH IM, a KONICA MINOLTA bizhub, a Develop ineo. These are office photocopiers. Somebody printed the document, ran it through the scanner and filed the result.

Why does it matter, if the PDF looks perfectly fine?

Because someone who cannot see the screen cannot read it at all.

A screen reader speaks the text it finds in the file. In a page that is an image it finds nothing, and a blind voter is left with silence — not a worse version of the manifesto, but none of it. The same applies, less severely, to anyone who wants to enlarge the text without it turning to mush, or copy a sentence out of it.

What we have done does not repair the filed document. The PDF on the ministry’s site is still a photograph, and anyone opening it there meets the same silence: the machine-read text exists on this site, not in the official file.

Is there a manifesto that cannot be searched even here?

One: that of the Partito Comunista dei Lavoratori.

It was not filed as a PDF, and the ministry’s server no longer serves that file to anyone: it answers 403 even to a browser. Since we hold no text for it, we say so beside the search, so that a missing result is not read as that party having written nothing on the subject.

Is filing a manifesto compulsory?

Yes. Anyone submitting a symbol must file the electoral manifesto alongside it.

The law requires the filing and the ministry publishes the files, but no rule says what form they must take: a searchable document and a photograph of a sheet of paper satisfy the obligation equally.

Source: Art. 14-bis(3), presidential decree 361/1957

How did you check this?

We downloaded the 31 files from the ministry’s site and counted the words of text in each.

To be sure the problem was the documents and not our own software, we repeated the measurement with an independent tool (poppler’s pdftotext): on the same files it also returns zero words. One case deserved care — the Referendum e Democrazia manifesto contains 112 words of text, but they are the notary’s certification on page 5, while the manifesto pages themselves are scans. A naive automated check would have counted it among the readable ones.

What about 2027?

We will repeat the same measurement on the manifestos filed for 2027, by the same method and identically for every list, and we will read the scans by machine again.

The comparison by subject, which on the 7 readable documents alone would have been a table with the cell blank beside the four most-voted parties, now exists: holding the text of 30 manifestos out of 31, we can say what each of them says about health, tax or pensions. Where no passage clears the checks on the reading, the page points at the place in the PDF rather than saying nothing.

Last updated