If there is a run-off in the coming French presidential election who will the two candidates be?

TL;DR

I forecast Marine Le Pen and Édouard Philippe at 43.2%. Marine Le Pen and Jean-Luc Mélenchon is close at 38.0% because Philippe leads when Gabriel Attal is absent, while Mélenchon usually moves ahead when both centrists run (current polling data). All other named pairs are below 8%, and I assign 7% to Any other pair, mainly for a Jordan Bardella substitution or an unexpected challenger.

Context

Marine Le Pen leads every recent tested first-round configuration, generally near 34% and far ahead of the contenders for second place (Ipsos, Ifop, Elabe, Cluster17). The forecast is conditional on a runoff. No current poll places any candidate close to a first-round majority.

The unresolved contest is behind her. Philippe, Attal and Retailleau are running separate campaigns, Mélenchon has an independent LFI candidacy, and Glucksmann must first win a social-democratic primary that does not bind LFI (centre-right reporting, LFI statement, PS primary rules).

Evidence

The historical backbone is candidate-field instability. Seven months before the 2017 election, an Elabe survey tested eight slates. It placed Alain Juppé at 26%–34%, Le Pen at 25%–28%, Macron at 14%–18% and Mélenchon at 14%–15% (September 2016 Elabe report). Juppé did not reach the election. The final top four were Macron at 24.01%, Le Pen at 21.30%, Fillon at 20.01% and Mélenchon at 19.58% (official 2017 results).

The 2022 cycle was even less stable. The table uses 31 first-round poll configurations fielded from August 22 through October 31, 2021, then compares their unweighted averages with the official April 10, 2022 result. Units are percentage points of expressed voting intentions; the polling vintage is the historical NSPPolls dataset, and the results are from the Interior Ministry.

The mean absolute error among candidates finishing above 4% was 5.60 points; the root-mean-square error was 6.85 points (underlying polling records). Mélenchon’s late consolidation and Pécresse’s collapse show why a one- or two-point lead now cannot be treated as durable. Candidate nomination and tactical voting mattered more than ordinary sampling error.

Current polling has one stable result and one large interaction. The following sample-size-weighted averages cover 50 configurations from 13 surveys with fieldwork ending between July 1 and September 3, 2026. They are presence-conditional percentages, calculated from the MieuxVoter polling file.

Across the full 168-configuration dataset, adding Attal to a Philippe slate is associated with a 5.05-point fall in Philippe’s support. Adding Philippe to an Attal slate is associated with a 6.36-point fall in Attal’s support (data and scenario definitions). These are observational effects because pollsters do not choose slates randomly. The direction is still clear: Philippe and Attal draw heavily from the same pool.

The latest direct surveys have effective voting-intention bases of roughly 680 to 1,211 respondents. Their indicative uncertainty around candidates polling near 15%–20% is about two to three points (Ipsos notice, OpinionWay notice, Cluster17 notice). The observed Philippe–Mélenchon gaps are therefore too small to settle the race.

Political behavior slightly favors eventual consolidation, but less than a simple polling average suggests. Philippe proposed a right-and-centre coordination committee. Attal responded that selection should wait until early 2027. Retailleau’s campaign said there would be a Retailleau ballot in the first round (September 9 reporting). Retailleau also has a formal party mandate: 73.8% of participating LR members backed his designation, with 60.01% turnout (LR results).

I assign 50% to Philippe facing no meaningful Attal candidacy, 36% to both remaining meaningful candidates, 5% to Attal running without Philippe, and 9% to another arrangement. I then blend current polling with moderate and high campaign volatility calibrated to the 2022 errors. Conditional on Le Pen qualifying, this gives 45.0% for Philippe as her opponent, 39.5% for Mélenchon, 8% for Glucksmann, 2% each for Attal and Retailleau, and 4% for someone else.

The calculation is:

$$P(\text{Le Pen–X})=P(\text{Le Pen qualifies})\sum_s P(s)P(X\text{ is second}\mid s),$$

where $s$ is the eventual candidate configuration. I assign Le Pen a 96.1% qualification probability. The Paris Court of Appeal left her firm ineligibility period already served, and she declared her candidacy, while the remaining exclusion scenario is described as marginal and legally disputed (court reporting, legal analysis). Most of the residual risk is a legal, health or strategic replacement by Bardella. Such an outcome normally resolves as Any other pair.

What's non-obvious

The published polling average mixes different elections. Philippe’s roughly 19% score without Attal and his roughly 15% score with Attal are often placed in one average. They should instead be treated as separate branches. The first makes Philippe the favorite for second. The second makes Mélenchon the favorite or produces a statistical tie.

Glucksmann’s primary does not create left unity. Mélenchon remains outside it, and Glucksmann trails him in every fresh joint test (Ipsos, Cluster17). Conversely, straight polling underrates Mélenchon’s upside: he gained almost 13 points from the comparable 2021 polling window to the 2022 election. I do not assume a full repeat, but it keeps Le Pen–Mélenchon close to the modal outcome.

Uncertainties

• The largest gap is the absence of a binding selection rule for Philippe and Attal. A signed withdrawal agreement, primary rules or a clear polling-based threshold would move the forecast sharply.
• No fresh survey tests Glucksmann without Mélenchon. Polling after the October social-democratic primary would show whether Glucksmann can consolidate Greens, Socialists and former Macron voters (primary timetable).
• Le Pen’s cassation proceedings create a small but asymmetric risk because a Bardella replacement sends almost every likely runoff to Any other pair. I could not verify a public full-text copy of the July 7 appellate judgment; the legal estimate relies on court reporting and published legal analysis.
• The polls use online quota samples and politically weighted likely-voter subsets. They describe current preferences, not future outcomes. Repeated identical-slate polls, candidate sponsorship data and post-primary voter-flow tables would narrow the remaining uncertainty.