New Delhi : India’s exit poll industry has a documented, data-backed history of getting it wrong – not occasionally, not by small margins, but spectacularly and in predictable directions, with some exceptions, of course. With West Bengal, Tamil Nadu, Kerala, Assam, and Puducherry results landing on May 4, here is what exit polls predicted the last time these states voted, and what that means for the numbers you saw last night.
West Bengal: The Costliest Failure In Recent Memory
The West Bengal record is the one every pollster quietly dreads being reminded of.
In 2021, the poll of polls estimated Mamata Banerjee’s Trinamool Congress at 156 seats and the challenger Bharatiya Janata Party at 121. On result day, TMC bagged 215 and BJP wrapped up at 77. Pollsters underestimated TMC by 61 seats and overestimated BJP by 49. Jan Ki Baat had BJP at 162-185. India Today-Axis My India called a narrow BJP majority. Every agency was wrong — not by a rounding error but by a landslide in the opposite direction.
This was not a one-off. In 2016, the poll of polls had TMC at 184 seats. TMC won 211. Two elections. Same state. Same directional error. Same party was underestimated both times.
In 2026, the poll of polls gives BJP 158 seats and TMC 130. If the 2021 correction factor applies even partially, those numbers look very different on counting day.
Please note: This is not an argument for a TMC win. It is an argument for humility.
Tamil Nadu: Right Winner, Wrong Everything Else
Tamil Nadu 2021 was the election where pollsters got the headline right and almost everything underneath it wrong.
Axis My India predicted the DMK alliance would win 175-195 seats. Today’s Chanakya gave them 164-186. The actual result was 159 seats for the DMK alliance, with DMK alone winning 133. The polls overshot DMK’s tally and simultaneously wrote off Edappadi K Palanisami’s AIADMK, predicting as few as 38 seats for the alliance when it actually won 75. The winner was called. The competitive texture of the race, the opposition’s resilience, and the precise magnitude of the swing were all wrong.
In 2026, Tamil Nadu has a variable that no agency has ever modelled before. Thalapathy Vijay’s TVK is being predicted to win anywhere between 13 and 120 seats depending on the poll you choose to read. That 100-seat gap is not analytical disagreement. It is every pollster possibly admitting their framework was not designed for this election. The agency that got Tamil Nadu’s direction right in 2021 — Axis My India — is also the agency predicting TVK at 98-120 seats this time. Whether that is insight or overreach is a question only results on May 4 can answer.
Kerala: The One They Got Right — And Why It Still Matters
Kerala 2021 is the exception that makes every other failure look worse by comparison.
Axis My India predicted LDF at 104-120 seats. C-Voter gave the LDF 71-77 seats and UDF 62-68 — projecting a much closer contest. The actual result was LDF 99 seats and UDF 41. Most agencies correctly called Pinarayi Vijayan’s historic second term the first time any alliance had won back-to-back in Kerala since 1977. Axis My India overestimated the LDF margin significantly, but the direction was right across the board.
Go back to 2016: Exit polls correctly predicted LDF’s return to power after a UDF term. Two consecutive correct calls in Kerala — the only state in this cycle with that record.
In 2026, the poll of polls gives UDF 76 seats, LDF 60, and BJP 4. Given Kerala’s comparatively clean polling environment and the exit poll track record here, this is the most trustworthy number on the board. If there is one state result you can read with reasonable confidence tonight, it is Kerala. It is only after the counting of votes on May 4 that Kerala’s final verdict would be read out to CM Vijayan.
Assam: The Cleanest Record Of All Four
Assam is where the exit poll industry goes to restore its confidence.
In 2021, Axis My India predicted NDA at 75-85 seats. C-Voter gave the NDA 58-71 and Congress+ 53-66 — projecting a closer race. The actual result was NDA 75 seats. Axis My India called it almost exactly. Even C-Voter, which projected a closer contest, got the winner and the broad range right. Two agencies predicted a hung assembly in 2021 and were wrong. Everyone else was in the right territory.
In 2026, the consensus is stronger and louder. Axis My India gives BJP+ 88-100 seats. P-MARQ has NDA at 82-94. Matrize predicts 85-95. Poll Diary goes as high as 86-101. Every agency, without exception, is pointing to a Himanta Biswa Sarma hattrick. Let’s wait for the real results on May 4.
Why Polls Keep Getting It Wrong
Two structural reasons, both well-documented and both directly relevant to 2026.
The first is fear. Exit polling depends on voters honestly telling strangers how they voted — often outside the booth, sometimes in full view of party workers who have spent months organising the neighbourhood. In high-pressure environments, voters frequently give the answer that feels safest. Academics call this social desirability bias.
The second is model failure. Every exit poll is calibrated on past elections. When something genuinely new appears — a first-time party, a new demographic wave, a welfare scheme that only shows up in booths and not in surveys — the model forces unfamiliar data into familiar boxes. In 2024, every major agency predicted NDA at 350-400 Lok Sabha seats. NDA won 293. BJP alone won 240. The models had been built on 2019 data and simply did not see 2024 coming.
The Hierarchy Of Trust
Not all exit polls deserve equal scepticism. Assam and Kerala have cleaner polling environments and more reliable track records. Tamil Nadu gets direction right but magnitude wrong. Bengal gets neither right with any consistency across the last two elections.
Read the 2026 numbers in that exact order of confidence. Believe Assam first. Believe Bengal last, or wait for May 4 to get the final results.
Exit polls are the best available signal before counting begins. In Assam and Kerala, they are a reasonable guide. In Bengal and Tamil Nadu, they are a starting point — not a verdict, not a forecast, and certainly not something worth forwarding before May 4.
