Language
How French news headlines assign La France insoumise and the Rassemblement National different political roles, 2022-2025.
By Amr Sobhy
This study asks not whether French headlines treat LFI or RN more harshly, but what kind of political actor each headline makes the party into. Two registers structure the analysis: conflict framing (attack, denunciation, directed confrontation) and strategic-game framing (alliances, polls, motions, electoral positioning). LFI is more often cast as a conflict actor; RN as a strategic-electoral competitor. The moral layer is real but organized by outlet politics, not a single national average.
28,592
Headlines audited
25
French outlets
3
LLM annotators
400
Human-audit headlines
28,592 headlines about LFI, RN, or their leaders, collected across 25 French outlets from 2022 through 2025. Each dot here represents one headline. The three colours show headlines that mention only LFI, only RN, or both parties in the same story.
One dot becomes one coded headline. Three independent open-weight LLMs — GPT-OSS-120B, Llama 3.3-70B, and Mistral Large 2 — annotate each headline with the same codebook; the final label is the majority vote. A 400-headline human audit then stress-tests the pipeline at its weakest points.
The pipeline does not ask whether a headline is simply positive or negative. It asks what political role the headline assigns: conflict actor, strategic competitor, target, aggressor, normalized actor, or delegitimized actor.
For every headline, the study asks two questions independently. First: does this headline show the party attacking or clashing with someone? Second: does it show the party playing the electoral game — positioning itself, negotiating alliances, running in polls? The striking result is that these two questions land differently depending on the party.
Active confrontation verbs directed at named actors: attacks, denounces, lambasts, targets. Excludes generic criticism and policy disagreement.
Electoral competition, parliamentary procedure, alliance arithmetic, polling, tactical positioning. The party as an ordinary institutional competitor.
LFI appears in a conflict frame in 13.8% of its headlines, compared with 8.7% for RN. RN appears in a strategic-game frame in 45.9% of its headlines, compared with 37.2% for LFI.
The point is not that one party gets "harsher" coverage. The point is that the French press treats LFI and RN as different kinds of political objects: one a behavioral antagonist, the other an institutional competitor.
A fair challenge: maybe LFI simply appears in more left-leaning outlets, which happen to use more combative language for everyone. To test this, the analysis controls for which outlet published the headline and what year it was. The pattern survives.
The chart shows each label as a dot with a confidence range. The centre line means "no difference between LFI and RN." Dots clearly to the left = more LFI; clearly to the right = more RN. Conflict and strategic-game are far from centre with tight ranges — those are the firm findings. The other labels have wider, less conclusive ranges and should be read with more caution.
Whiskers: outlet-cluster bootstrap 95% CIs (2000 iter, 9 outlet families). OR < 1 → more LFI; OR > 1 → more RN.
The conflict pattern was consistent everywhere. This one is not. When it comes to which party gets portrayed as morally suspect or unworthy of power, the outlets split along editorial lines. Left-leaning outlets do it to RN; right-leaning outlets do it to LFI. The national average looks balanced — but that balance is an illusion produced by two opposing camps cancelling each other out.
This is the paper's deepest result. Every outlet agreed that LFI is more often the attacker. But ask which party is the villain — the one that deserves blame, that is portrayed as dangerous or unfit — and the outlets split hard.
Libération, Le Monde, and HuffPost publish significantly more headlines that cast RN as the morally problematic party. JDD and Le Figaro do the same to LFI. The chart shows this: bars to the right mean "RN gets blamed more at this outlet"; bars to the left mean "LFI gets blamed more."
Here's the key twist: if you average all outlets together, blame looks evenly split between the two parties. That average is not wrong — it is hiding something. The left-leaning and right-leaning outlets are doing opposite things at roughly equal intensity, and they cancel each other out in the national total. A headline like "media blames both parties equally" would be accurate and completely misleading at the same time.
In summer 2023, Vincent Bolloré's media group consolidated editorial control of Le Journal du Dimanche, prompting a six-week newsroom strike and the departure of much of the editorial staff. Inside JDD's own headlines, the delegitimization gap between LFI and RN inverts: before the transition, JDD delegitimized RN more than LFI (raw RN-vs-LFI OR = 1.52); after, JDD delegitimizes LFI more than RN (raw OR = 0.54). LFI delegitimization at JDD rises from 12.4% to 41.3%. RN's moves only from 18.9% to 22.2%.
The reversal is striking, but the headline-level claim is narrow. With one outlet and no matched control, this is suggestive evidence of how editorial control can re-route moral accounting — not a causal verdict.
RN's strategic-game advantage peaks in 2023 (RN–LFI gap = +15 pp; OR = 2.21), holds in 2024, and collapses in 2025 (gap = +2.2 pp, OR = 1.09, not significant) as Marine Le Pen's March 2025 conviction displaces ordinary electoral-positioning language with judicial and ineligibility vocabulary. LFI's conflict excess does not move that way — it is present every year, and is widest in 2025. The behavioral labels are also the strongest reliability tier.
The strategy story moves. The conflict story stays.
Delegitimization rates follow a different temporal logic than conflict or strategic-game framing. LFI delegitimization spikes sharply in 2023, during the pension reform protests, reaching 32.1% — nearly double the 2022 baseline. RN delegitimization rises more steadily, surpassing LFI in 2024 at 33.4%.
The crossover is revealing: LFI is delegitimized most when it mobilizes outside institutional channels; RN is delegitimized most as it approaches electoral power.
The press does not simply treat one party more harshly. It casts LFI and RN as fundamentally different kinds of political actors — and that difference is more revealing than any negativity score.
About 1 in 7 LFI headlines shows the party clashing or denouncing someone; for RN it is fewer than 1 in 11. The gap holds even after accounting for which outlet published the headline and what year it was.
Nearly half of all RN headlines show the party polling, negotiating alliances, or positioning for elections. For LFI it is just over a third. The press frames RN as an institutional competitor; LFI as a disruptive force.
The LFI-as-attacker pattern shows up at left-wing, right-wing, and centrist outlets alike. But which party gets portrayed as dangerous or unfit for power depends entirely on which newsroom you read.
Left-leaning outlets cast RN as the morally problematic party; right-leaning outlets do the same to LFI. Averaged together, it looks like "balanced" coverage — but that average hides two very different editorial projects cancelling each other out.
LFI delegitimization at JDD jumped from 12.4% to 41.3% after the 2023 editorial transition; RN's barely moved. A change in ownership produced a measurable change in who the outlet framed as dangerous.
RN's strategic, election-focused coverage evaporated in 2025 as headlines shifted to trials, ineligibility, and legal proceedings. LFI's conflict-register coverage was there in every year of the study, and was highest in 2025.
The conflict and strategic-game findings are confirmed by both AI annotators and human reviewers across every test. The blame and delegitimization findings are real patterns, but noisier. They are reported here as tendencies worth watching, not certainties.
The contribution is not a new scoreboard for shaming outlets. It is a vocabulary for asking better questions about how the French press constructs its two largest challenger parties.
A conflict actor? A strategic competitor? A morally suspect outsider? These are different scripts, not different tempers.
All outlets agree that LFI attacks more. They disagree on who deserves blame. That difference — universal vs. partisan — is the most important shape in the data.
JDD's 2023 reversal is the clearest within-outlet shift in the corpus. Headline data can detect editorial regime change.
The conflict and strategic-game findings are confirmed by both AI and human reviewers and survive every statistical stress test. The blame and delegitimization findings are real but softer — reported here as tendencies, not hard conclusions.
The design is observational. It documents how headlines distribute political roles between LFI and RN; it does not measure whether those roles are warranted by the parties' actual behavior, nor whether outlets intended the distributions they produced.
An accurate-coverage explanation is plausible: LFI led Article 49.3 opposition, censure motions, and street mobilization against pension reform in 2023; RN simultaneously pursued a deliberate normalization strategy. The data cannot adjudicate between accurate coverage of genuinely different political behavior and active framing construction.
"The French press is biased against (or in favour of) LFI / RN."
"A headline coded delegitimizing is morally wrong; one coded normalizing is right."
"French headlines cast LFI and RN in measurably different roles. Every outlet shares a common script about who attacks. Outlets split on who deserves blame — and that split follows editorial politics."
Three further caveats worth knowing: (i) the three AI models may share blind spots from overlapping training data — using three models catches disagreements between them, but not assumptions they all share; (ii) the corpus only includes headlines that name LFI or RN directly, which may over-represent moments when the party is the main subject rather than a passing mention; (iii) a headline and its article can frame the same story differently — this study reads the headlines only.
The useful takeaway is not which outlet is biased. It is that single negativity scores miss the main pattern: French headlines assign different roles to LFI and RN, and those roles behave differently across outlets and time.
Role framing gives the comparison a better vocabulary. LFI is more often placed in the register of confrontation; RN is more often placed in the register of institutional competition. The moral layer is real, but it belongs to outlet-level politics rather than to one stable national average.
If you use this research in your work, please cite it as:
@misc{sobhy2026conflictstrategy,
title = {Conflict or Strategy? Asymmetric Role Framing of La France insoumise and Rassemblement National in French News Headlines, 2022--2025},
author = {Amr Sobhy},
year = {2026},
note = {Working paper, version May 2026},
}Working paper, June 2026. Please contact the lab with any questions about citing or using this research.
For research collaborations, media questions, or institutional inquiries, contact Le French News Lab.
Le French News Lab