Which pollsters to trust in France
Every pollster that has polled France, rated on how close its final polls came to the result and how openly it publishes; 12 polls there in the last 30 days.
Scored the same way as our British ratings, out of 100. Round-one presidential and legislative final polls since 2002 count towards accuracy. Each pollster is ranked against France's final polls, and one with little record is pulled towards a typical one, so grades show how a pollster compares with its rivals here and can't be compared across countries.
| Pollster | Grade | Score | Elections | Polls now |
|---|---|---|---|---|
| Harris Interactive | B | 70 | 7 | 12 |
| Cluster17 | B | 67 | 3 | 7 |
| Ifop | B | 65 | 11 | 16 |
| Odoxa | B | 65 | 4 | 2 |
| Ipsos | B | 63 | 11 | 4 |
| OpinionWay | C | 59 | 7 | 8 |
| Elabe | C | 56 | 5 | 5 |
| Verian | C | 53 | 9 | 1 |
| YouGov | C | 51 | 3 | 3 |
Past pollsters
A final-poll record, but no polls in the last year| Pollster | Grade | Score | Elections | Polls now |
|---|---|---|---|---|
| Le Terrain not polling now |
B provisional | 67 | 1 | 0 |
| Louis Harris not polling now |
B provisional | 66 | 1 | 0 |
| LH2 not polling now |
B | 66 | 3 | 0 |
| CSA not polling now |
C | 48 | 6 | 0 |
| BVA not polling now |
C | 45 | 9 | 0 |
| AtlasIntel not polling now |
D provisional | 40 | 1 | 0 |
At past elections
The average miss of its final polls, by party, in points: positive means it had the party too high.
| Elections | RN | LR | LFI | PS | Ensemble | Left (NFP) | RN + allies | |
|---|---|---|---|---|---|---|---|---|
| Ifop | 11 | −0.6 | −0.8 | −1.2 | −0.9 | −0.7 | +0.2 | +2.3 |
| Ipsos | 11 | −0.3 | −0.2 | −0.7 | +0.2 | −0.1 | −0.1 | +1.9 |
| Verian | 9 | −0.2 | −0.7 | −1.9 | +0.2 | −1.4 | −0.5 | +2.4 |
| BVA | 9 | −0.6 | −0.9 | −0.5 | +1.2 | −1.7 | +0.6 | +2.9 |
| Harris Interactive | 7 | −0.4 | −0.7 | −1.2 | 0.0 | −0.6 | −0.2 | +2.6 |
| OpinionWay | 7 | −0.8 | +0.2 | −1.5 | +0.3 | −0.7 | −0.5 | +2.6 |
| CSA | 6 | +0.4 | −1.9 | +3.4 | +0.3 | – | +1.6 | +2.1 |
| Elabe | 5 | +1.0 | +1.5 | −2.3 | +0.6 | −0.9 | 0.0 | +2.5 |
| Odoxa | 4 | +1.3 | −0.3 | −3.3 | +1.1 | +0.4 | −0.5 | +3.3 |
| Cluster17 | 3 | −2.1 | −0.2 | −3.9 | – | −0.8 | +2.7 | +0.3 |
| LH2 | 3 | +0.6 | −1.3 | +3.9 | −2.2 | – | +1.3 | +0.7 |
| YouGov | 3 | −0.1 | −2.1 | −5.9 | – | −0.8 | −1.2 | +3.4 |
| Le Terrain | 1 | +0.2 | −2.5 | +2.4 | +1.6 | 0.0 | – | – |
| Louis Harris | 1 | −3.9 | +1.2 | – | +1.8 | – | – | – |
| AtlasIntel | 1 | −1.8 | – | −3.3 | – | −0.1 | – | – |
Now
Its latest poll minus our average, in points. A lean is not an error: nobody knows yet who is right.
| Mélenchon | Roussel | Glucksmann | Philippe | Retailleau | Dupont-Aignan | Le Pen | |
|---|---|---|---|---|---|---|---|
| Harris Interactive | +0.7 | −1.1 | −0.6 | −2.2 | −2.0 | −1.1 | −1.8 |
| Cluster17 | +1.7 | −0.1 | −0.6 | +1.3 | −1.0 | +0.9 | −2.2 |
| Ifop | +1.2 | +0.4 | −1.6 | −0.2 | −2.0 | −0.6 | +2.2 |
| Odoxa | +0.7 | −0.1 | −0.1 | −1.2 | +4.0 | −0.1 | −0.8 |
| Ipsos | −1.3 | −0.1 | +2.4 | +2.8 | +2.0 | −0.1 | +0.8 |
| OpinionWay | −2.3 | −0.1 | +0.4 | −0.2 | −2.0 | −0.1 | +1.2 |
| Elabe | −1.3 | −0.1 | +1.9 | +2.8 | 0.0 | +0.9 | +0.2 |
| YouGov | +0.7 | +0.9 | −1.6 | −3.2 | +1.0 | −0.1 | +0.2 |
Which way each pollster leans
Each of the 10 polls since the last election is set against the other pollsters' polls of the same fortnight. A pollster that keeps putting a party above everyone else has a house effect. It is not a mistake in itself: the result decides who was right. Our average does not adjust for these; it takes each pollster's latest poll with equal weight.
| Pollster | Polls | RN + allies | Left (NFP) | Ensemble and centre | LR | Other left | REC | Others | Scatter v sampling |
|---|---|---|---|---|---|---|---|---|---|
| Ifop (Ifop-Fiducial for rolling polls) Sep 2025 – Sep 2025 |
2 | +0.9 | −0.1 | −0.1 | +0.5 | −0.4 | −0.4 | −0.6 | – |
| Elabe Aug 2025 – Aug 2025 |
1 | −0.1 | −0.4 | −0.2 | −0.3 | +0.3 | +0.2 | +0.2 | – |
| Harris Interactive / Toluna Harris Interactive Sep 2025 – Sep 2025 |
1 | +0.4 | +0.4 | +0.5 | −0.5 | +0.1 | +0.2 | −0.3 | – |
| Cluster17 Sep 2025 – Sep 2025 |
1 | −0.9 | +0.1 | −0.2 | +0.5 | −0.1 | −0.1 | +0.5 | – |
Cells: points above (+) or below (−) the other pollsters of the same fortnight, shrunk toward zero for pollsters with few polls. Bold: more than two standard errors from zero. Hover for the raw figure.
Herding: too close to everyone else?
A poll of 1,000 people should land a point or two from the truth by chance alone, so honest polls scatter. The last column compares each pollster's scatter around the others (its lean removed) with what its sample sizes imply. About 1 is normal. Well below 1 means its polls sit closer to the pack than chance allows, which can mean adjusting toward other polls; well above 1 means more noise than its sample explains.
No pollster with eight or more polls is below 0.5.
Caveats: a pollster that polls rarely has few others in its fortnight; small parties' scatter is small anyway; a missing sample size counts as 1,000; online panels with heavy weighting can be steadier without copying anyone. This is a flag for a closer look, not a verdict.
How the score works
- Accuracy (50): the average miss per party of its final polls at presidential elections since 2002 and legislative elections since 2017, ranked against every final poll in this country: the more of them it beats, the more points. A typical final poll scores about 30, one in the best tenth about 46. A pollster with little record is treated as if it had two more elections at the typical level, so one lucky poll can't make a top grade.
- Lead (20): its miss on the gap between the two leaders, ranked the same way.
- Openness (30): what it publishes so its figures can be checked, out of the items we have checked so far: one not checked yet counts neither for nor against. A pollster whose publications we have not checked yet is graded on its record alone (accuracy and lead scaled to 100), marked "record only".
A 75+: Proven accurate and fully open · B 60+: A good record, and checkable · C 45+: Some evidence of accuracy, or very open · D 30+: Little evidence it gets elections right · E <30: Unproven and hard to check
Figures as of . Published by Polling Forecast.



