Elections and social media, from 2014 to 2027

ReputatioLab, Saper Vedere's laboratory for crises and reputation · File updated on 13 September 2026 · reputatiolab.com →

31ReputatioLab articlestitles and excerpts reproduced as they stand, each card leads to the article

2014 → 2026years coveredfrom the first analysis to the most recent

6summary sectionsthe contents list gives them, on the left

Publication rhythm

20143articles

20154articles

20164articles

201711articles

20221article

20243articles

20252articles

20263articles

Key takeaways

What twelve years of analysis teach us

Social media counters do not tell you who is winning. Nicolas Sarkozy led almost every indicator of the right-wing primary and did not reach the second round. These figures measure mobilisation, not a ballot. What the social web says about the right-wing primary
You do not suffer your bubble, you enter it. The mapping of ten million tweets from the presidential election shows camps closed in on their own sources, with no algorithmic feed imposing it. Are there information bubbles?
A mobilised minority looks like a majority. During the speech by François Hollande and Angela Merkel at the European Parliament, one camp hogged every ranking while its community remained a minority in the graph. The FN organises on Twitter
A rumour does not need volume. Ten thousand three hundred and twenty-seven tweets from the patriosphère were enough for a candidate to respond publicly to a rumour never mentioned before. From the patriosphère to the media
Interference is measured, not assumed. Russia Today reached 145,017 different people during the campaign, but the core that interacts more than once with Russian media remains under fifteen thousand individuals. What Russian influence on the campaign?
Social networks have become networks of loyalty. The assessment of the municipal elections says it plainly: they sustain the convinced and feed the ground campaign, they no longer conquer. Is Sarah Knafo favoured by the algorithms?

Social media indicators do not tell you who is winning

This is the common thread running through twelve years of measurement, and it was always written before the results, never after. Before each election, ReputatioLab published the ranking of candidates on social media indicators alone: engagement, shares, followers, views, size of communities during the debates. The exercise was meant to avoid post-match analysis. Above all, it established that these rankings predict nothing.

2016 Right-wing primary Nicolas Sarkozy was the candidate with the most engagement and the most exclusive audience on Twitter. The analysis notes that nobody could have predicted François Fillon's victory, unless they bet everything on Facebook shares. The analysis

2017 Citizens' primary Benoît Hamon came first on almost all indicators, Manuel Valls second. For once, these indicators were more accurate than the polls. The article immediately adds the caveat: nothing makes it possible to turn this into a rule. The analysis

2017 Presidential election Had social media voted, Jean-Luc Mélenchon would have won ahead of Marine Le Pen, with Emmanuel Macron third. The author says so before the vote: these indicators are worthless as proof of a future result at the ballot box. The analysis

The social web is not a good indicator of a candidate's strength at the ballot box.

ReputatioLab, What the social web says about the right-wing primary, 2016

The counter-example that served as a demonstration

The analysis of the Facebook audiences of the two finalists of the right-wing primary gives the most telling figure of the series. Facebook estimated the number of people interested in Alain Juppé at 280,000, against 160,000 for François Fillon. It was Fillon who won, by a wide margin.

Alain Juppédeclared interest on Facebook

280,000 people

François Fillondeclared interest on Facebook

160,000 people

How these analyses are made

The Twitter corpora come from Visibrain, supplemented by Brandwatch, Talkwalker and Radarly, and by SoPrism to read audiences as Facebook sees them. Amendments and interest-group statements are tracked with Follaw.SV. The maps are built in Gephi: each point is an account, each link a mention, a share or a follow.

Bubbles we knowingly jump into

Public debate readily attributes information bubbles to algorithms. These analyses say something else. The mapping of more than ten million tweets from the presidential election, then of half a million tweets taken over one week to look at the domains shared rather than the accounts, arrives at the same picture: each camp gets its information in its own lane.

Yes, we do have information bubbles. And no, it is not an algorithm that sets them for us. We jump into them knowingly.

ReputatioLab, Are there information bubbles?, 2017

The major media are cited by everyone: what distinguishes the lanes is what gets added to them.

Activist community	The three most shared domains
Marine Le Pen	Boulevard Voltaire, FrDeSouche, Le Figaro
François Fillon	Le Figaro, Valeurs Actuelles, L’Opinion
Emmanuel Macron	Le Figaro, Le Monde, Libération
Jean-Luc Mélenchon	Le Monde, L’Huffington Post, FranceTvInfo
Benoît Hamon	Libération, Canal +, Le Monde

The asymmetry is clearly named: the bubble formed by the Fillon and Le Pen communities is the one that departs furthest from the traditional media and relays sites described as conspiracist. The PenelopeGate case confirms it, with a single fact producing five readings. It is not a matter of choice of newspaper but of framing: on the far right, information is not merely decoded differently, it is put in context by other information and other frames of thought.

A debate moves nobody

The oldest demonstration in the file concerns Nicolas Sarkozy's return to a television studio. By dividing the evening into four moments, the mapping shows communities growing without ever mixing. The same exercise on François Hollande's address to the French produces two separate blocs. In both cases, once the programme is over, the conversations stop dead.

So in reality we have communities that talk among themselves but do not engage in dialogue.

ReputatioLab, Political discussions on Twitter: the Sarkozy case, 2014

The noise factory: minorities, fake accounts, rumours

If social media do not shift convictions, they manufacture an appearance. This is the best-documented part of the file, and the most constant from one election to the next: occupying the rankings to create an impression of a majority.

135 accountsrun by a single operator on the eve of the presidential electionReputatioLab, the first act of botting in the presidential election

75 fake accountscreated overnight to launch a false affair against a candidateReputatioLab, how a shadowy outfit created accounts to cause harm

26% of accountsrelaying the European Parliament sequence had fewer than a hundred followersReputatioLab, the FN organises on Twitter

The case of the presidential campaign's botting is the clearest. Spotted because an ordinary political tweet was gathering more than twelve hundred shares, the account was only a facade: bought followers, bought retweets, and a fleet of secondary accounts posting at the same second from a single dashboard. The aim was not to convince, but to appear in the hashtag trends and word clouds that quantitative studies pick up.

This should give journalists particular pause, and tell them not to be taken in by consulting only quantitative rankings.

ReputatioLab, The FN organises on Twitter, 2015

The mechanics of a rumour, in five steps

The file contains two complete reconstructions: the false Saudi funding affair, built on a mirror site of a major Belgian daily, and the Paris rumour about a candidate's private life. Both follow the same trajectory.

Identify a weakness. A young party has not yet compartmentalised its networks, and its only grey area is its funding. That is the one chosen, because it feeds every imaginable conspiracy.
Build a credible vehicle. A mirror site resembling that of a well-known title is enough, with confusion between the two addresses doing the rest.
Target the most emotional. The first recipients are the opponents of the targeted candidate. This is the reactive paradox: the community that wants to punish the sender ends up promoting it.
Wait for the network bridge. Two weeks go by with no effect, then an elected official picks up the link and mentions of the Belgian daily soar.
Accept the final asymmetry. The hoax is spread by one camp, the rebuttal only by the press. The two do not circulate in the same networks.

So it took only 10,327 tweets from the patriosphère to get a candidate talking about a rumour that had been circulating for a long time but never mentioned until then.

ReputatioLab, From the patriosphère to the media, 2017

It remains to establish who is spreading it. The investigation into nine entirely false rumours of the presidential campaign isolates some eight hundred accounts, which piled up a million tweets in two months. Cross-referencing with the most active accounts around the two Russian state media is unambiguous: a 75% match for those that relayed at least three rumours, and up to 92% for those that relayed five. The word most used by these accounts is not the name of their candidate, but that of their target.

Foreign interference, measured rather than feared

The file holds a consistent and uncomfortable position here: documenting operations without overstating their reach. The analysis of Russian propaganda clearly distinguishes the two outlets. Sputnik barely covers the French campaign. Russia Today makes it its editorial line and reaches 145,017 different people through shares alone.

145,017 peoplereached by Russia Today's shares during the campaignReputatioLab, Russian influence on the presidential campaign

57,337 peoplereached by Sputnik, which barely covers the presidential electionReputatioLab, Russian influence on the presidential campaign

6,006 profilesonly beyond ten interactions with the two outletsReputatioLab, Russian influence on the presidential campaign

The third figure is the one that matters. This broad audience is a transient flow: at two interactions, only 14,792 individuals remain, and ten interactions bring the core down to 6,006 profiles, spread across only three communities. The share of pseudonymous accounts there exceeds 90%.

The file also documents an operation that did not come from Russia: on the false Saudi funding affair, examination of the three initial accounts tips the hypothesis towards Iran, with Russian media themselves being targeted by those accounts.

So I believe very little in informational interference that would have an impact. There will be some, that is certain. But that it will have an impact, I very much doubt.

ReputatioLab, The key issues of the presidential election, 2026

The position held today rests on a quantified precedent: the Russian network Storm-1516 identified by Viginum created one hundred and forty fake sites, for impacts that Viginum itself describes as almost nil. The analysis also recalls that the MacronLeaks largely fizzled out. What has changed, however, is the method: rather than building audiences for relay accounts, operators go where the audience already is, that is, in the comments.

From conquest to loyalty: what 2027 changes

The file ends on a reversal. The analysis of the municipal elections concludes that elections no longer take place on social media, if indeed there was ever a real battle in those spaces.

They have become networks of loyalty, not networks of conquest.

ReputatioLab, Is Sarah Knafo favoured by the algorithms?, 2026

The case against algorithms, and what measurement says about it

One study argued that a candidate benefited on X from an exposure far greater than her organic audience, by comparing her views with those of a competitor with a comparable number of followers. The debunk does not dispute the figures, it disputes the cause.

Sarah KnafoOn XRachida Dati

17.5 million

Video views in Januaryfigures put forward by the study

1.8 million

220 000 à 240 000

Declared followerssame order of magnitude

220 000 à 240 000

The explanations put forward are editorial. One ran a campaign designed for social media, while the others broadcast reruns of programmes or extracts of rallies, and the same gap is found on the other networks, which rules out the hypothesis of a platform-specific setting. Follower numbers do not predict performance: a recent audience engages, while an older audience, part of which has left the network, no longer moves.

The trap of the borrowed indicator

The warning applies to any reading of a dashboard: the reach indicators displayed by monitoring platforms are entirely artificial, since they multiply the number of posts by the number of followers. They are not real views. Comparing two players on that basis is comparing two estimates, not two audiences.

What, on the other hand, will carry weight

The analysis of the coming campaign shifts the centre of gravity towards four areas. First, a fragmented media landscape: a right-wing ecosystem structured around a news channel, content creators large enough to offer editorial targeting by voter segment, and private messaging apps that will have to be fed. The single-strategist model is described there as finished.

Then there is door-to-door canvassing, which has become central again because the technological barrier has fallen. The analysis cites the founding comparison: one vote won for fourteen face-to-face conversations, against one for thirty-eight personalised phone calls, against one for a hundred thousand leaflets distributed.

Then there is funding, a terrain that social media do not show. While corporate donations are banned and individual donations capped at €4,600, a campaign can be launched on very few donors: in 2017, 913 of them, 1.2% of the total, accounted for 48% of the funds raised. Add to that the state of the parties' finances, which are incomparable from one camp to the other, from thirty-three million in assets for the Parti socialiste to a negative balance of eleven million for the Rassemblement national.

Finally, the data itself, whose abundance is presented as a risk as much as an asset.

For an election, data is a thermometer. An excellent thermometer, provided you know what you want to measure and use it for tracking, not for steering.

ReputatioLab, The key issues of the presidential election, 2026

What this changes for an organisation

These analyses were written for campaigns, but they describe mechanics that any exposed organisation will recognise. Four practical consequences emerge.

Monitor communities, not rankings

A hashtag ranking, a word cloud or a league table of accounts can be manufactured, and a minority but coordinated community occupies them with ease. Network analysis puts share of voice back in its real place: this is why these studies always pair quantitative measurement with a mapping. Monitoring that looks only at volumes will be caught out.

Prepare, because the window for reaction is short

The mirror-site case contains the most concrete countermeasure in the file, and it costs nothing: the targeted daily newspaper could have spotted the fraudulent address two weeks before things ran away, by monitoring the shares that cite it without pointing to its own domain. On the party side, the finding from the following campaign is of the same kind: of the 268 MPs in the majority group, only 55 had mobilised on the day the bill was presented, with no rapid-response unit, leaving the field clear for the opposition. Multi-network analysis

Responding without giving power to those who have none

Two articles converge. On the programme cancelled in autumn 2015, the manoeuvres of the two major parties to control their opponent's appearance produced the opposite of the intended effect, a perfect Streisand effect. The study of the regional elections states the rule in one sentence.

In crisis management, it is said that you should not give power to those who have none.

ReputatioLab, The regional elections: the FN flees Twitter in the second round, 2015

The corollary applies to any organisation tempted to answer noise with noise.

The astroturfing logic is therefore pointless. You create noise for nothing. You influence nothing.

ReputatioLab, Political discussions on Twitter: the Sarkozy case, 2014

Choosing the right channel for the right interlocutor

The reconstruction of the accounts followed by ministers shows a highly political and highly media-oriented ecosystem, with a strong local anchoring and three families of interest groups: agricultural, local economic and societal. Entire sectors, health, automotive, aviation and energy, are absent, and advocacy lodged on a network that ministers do not use is lost. Analysis of ministerial ecosystems The influence measurement carried out on the following government reaches the same diagnosis: an institutional executive that is largely offline. Analysis of the Lecornu government

The other route is parliamentary. An examination of the amendments co-signed after the dissolution reveals stable cross-party pairings on technical subjects: health, local taxation, environment, local authorities. It also reveals an asymmetry, with many socialist amendments co-signed by Les Républicains MPs without the reverse being observed. For anyone seeking to get a subject taken up, this is a far more precise entry map than a count of mentions. Analysis of parliamentary consensus

The analyses behind this file

Each lesson in this file points to one of these articles, published by ReputatioLab. They are sorted by sub-theme, from most recent to oldest.

Towards the 2027 presidential election: today's political scene

ReputatioLab9 April 2026

2027 Presidential Election: the 7 key issues

The municipal elections are over. Now we can look ahead to the 2027 presidential election! Conveniently, the funding period for the 2027 presidential election campaign began on 1 April 2026.

Elections and social media: from 2014 to the 2027 presidential election