How the BJP won West Bengal in 2026: what the polls said and what happened

The BJP won 207 seats on counting day. It became 208 after a repoll in Falta on 21 May. Suvendu Adhikari, a former Trinamool leader who moved to the BJP, was sworn in as Chief Minister on 9 May, the first BJP Chief Minister of the state. Mamata Banerjee, who had led the state for 15 years, lost the Bhabanipur seat to him by 15,105 votes.

Three things stand out. The BJP’s win was broad, not narrow: 133 of its seats came by more than 20,000 votes. Turnout jumped sharply even though the voter list had been cut. And the state’s voter-list revision (called SIR) has become the main argument about the result. We look at that argument closely, because the numbers people quote are not measuring the same thing.

The one pre-poll survey we could confirm in detail was VoteVibe, published by CNN-News18 on 23 March 2026. It gave the Trinamool 184 to 194 seats and the BJP 98 to 108, with vote shares of 41.9% and 34.9%. In the same poll 48.5% of people named Mamata Banerjee as their preferred Chief Minister. That was an opinion poll taken weeks before voting, and it is one poll, not a consensus of all polls.

The past results also pointed towards a Trinamool win. In the 2024 general election in the state, the Trinamool won 29 of 42 Lok Sabha seats with 46.2% of the vote, and the BJP won 12 with 39.1%. Assembly and Lok Sabha shares are not directly comparable, but they show the BJP had not been ahead of the Trinamool in any election in this period.

Exit polls, taken from people leaving polling booths, moved towards the BJP. But even the boldest of them fell short of the actual number. The table below is based on the Free Press Journal’s roundup.

So the exit polls did not agree with each other, and none matched the result. The BJP’s 208 was above the top of the highest range, and People’s Pulse had the wrong winner. We could not find a poll that expected the Trinamool to fall as low as 80.

Some agencies gave seat ranges only for one party, so this table shows what each published. We have not judged the quality of any poll’s method.

Note that the BJP’s gain in vote share was about 7.9 points, and the Trinamool’s loss about 7.3 points. That is a large swing by any standard, but it is not a collapse of the Trinamool’s base. It still polled about 40.7% and 2.6 crore votes.

The BJP polled about 2.94 crore votes in 2026, against 2.29 crore in 2021, which is about 65 lakh more. The Trinamool polled about 2.60 crore, against 2.90 crore, which is about 29 lakh fewer. The BJP gained more than the Trinamool lost because far more people voted: turnout rose from about 82% to about 93.7%.

Here is the odd part. The state’s electoral roll was smaller in 2026 than in 2021, by roughly 50 to 55 lakh names by our data, yet about 36 lakh more votes were cast. In other words, a much larger share of the roll turned out. Who those extra voters were, and whom they voted for, is something the results alone cannot tell us.

Comparing vote counts is arithmetic, not proof of voters switching. A voter who did not vote in 2021 and voted BJP in 2026 looks identical, in these totals, to a Trinamool voter who switched.

Three seats are left out of the table. An independent won one in 2021 and the BJP took it in 2026, a small party’s seat went to the AISF, and in Bishnupur our data has no 2021 match. Looking at the final two candidates in each seat, the BJP won with the Trinamool second in 203 seats, and the Trinamool won with the BJP second in 67.

The victories were mostly wide. The median winning margin across the state was 26,238 votes. The BJP won 133 seats by more than 20,000 and 24 by more than 50,000. Only 13 of its wins were by fewer than 5,000. The Trinamool’s 80 winners had a median margin of 27,114, and only seven of its seats were close.

Some individual seats have their own stories. Suvendu Adhikari won both Bhabanipur, from Mamata Banerjee, and Nandigram, by 9,665 votes. He gave up Nandigram and kept Bhabanipur. In Falta, the Election Commission cancelled the 29 April poll over reported tampering and ordered a repoll on 21 May, and the BJP’s Debangshu Panda won it by 109,021 votes. Trinamool minister Aroop Biswas lost Tollygunge by 6,013.

Before this election the Election Commission ran a Special Intensive Revision (SIR) of the voter list. The Chief Electoral Officer’s press note of 28 February 2026 gives the numbers.

Of the 58.2 lakh forms not received when the draft was made, 24.2 lakh were listed as deceased, 19.9 lakh as shifted, 12.2 lakh as absent, and the rest as already enrolled elsewhere or other. Names under adjudication were to be decided by judicial officers and added by a supplementary list. The roll on polling day was about 6.8 crore, in our data. Whether the process was fair is a separate debate. Here we look only at the narrower question of whether it changed the result.

Several analysts have tested this using deletion data compiled by the Sabar Institute, a Kolkata research group. They reached different numbers because they counted different things.

These are not contradictory. A count where “deletions exceed the margin” only says the deleted names could have mattered in that seat if they had voted, and voted one way. It does not say they would have. Scroll itself stops short of saying the deletions caused the result. Ray, an economist, goes further and tests how likely such an outcome is. He finds that if the deleted voters had turned out at the state’s actual 92% rate, about 95% of them would have had to vote against the BJP to cost it the majority. In his words, the state-level outcome does not turn on the deletions under any plausible assumption, though particular seats may have.

Ray’s own note says this is not an argument that the SIR did not matter. Bhabanipur is his example: by his count, about 36,500 names were in the contestable group against Mamata’s losing margin of 15,105. Scroll’s count for the seat, using all deletions, is over 51,000. The Caravan quotes political scientist Gilles Verniers saying such exercises are “at best, incomplete” and at worst “meaningless”. Ray also makes a wider point that the process itself, and how individuals were treated, is a question of principle apart from the arithmetic.

To sum up what the analyses above show: the deletions were large enough to matter in some close seats, and the one analysis that tests the state-level result (Ray’s) does not find that they explain it. That is a summary of other people’s work, not a finding of our own, and we do not have deletion data by seat to test it ourselves.

Nothing official explains why voters chose as they did. Debraj Ray, in the opening of his note, lists the factors he expects researchers to look at: what he calls considerable arrogance in the Trinamool, the R.G. Kar case and how the government responded to it, a widespread perception of corruption, and the politics of a border state with deep Hindu-Muslim divisions. He presents these as questions for later study, and so do we. We have not found a post-poll survey for this election with breakdowns by gender, age, religion or region, which is what it would take to test them.

What the numbers do say is limited: the BJP’s gain came mostly in seats the Trinamool had held in 2021, turnout rose by about 11 points, and the Trinamool kept around 40% of the vote. Anything more specific about who moved is, for now, informed opinion.

• Who voted for whom. No post-poll survey we could find covers this.
• Whether the additional voters were new voters, past non-voters, or switchers.
• How many deleted voters would have voted at all, and for whom.
• Why the exit polls were so spread out, and none matched.
• The status of any court challenges to individual seats. We have not confirmed any, so we list none.

Seats, votes, margins and seat flips are our own analysis of results published by the Election Commission of India, the same data behind the West Bengal 2026 election page. Turnout is about 93.7% by the Commission’s figure, and 93.6% in our data, so we say “about 93.7%”. The 2021 baseline (turnout about 82%, about 73.8 million electors) differs slightly between compilations. Exit-poll and survey numbers are from the outlets named, and analysts are credited by name. If you find a number that does not match the Commission’s, tell us on the contact page.