Single-seat forecast
Your win probability, vote-share range and a written brief for one constituency, with swing factors and next moves.
Ask about a pilot →AI election intelligence
for the next campaign
Know your chances, seat by seat. PollBytes combines past results, local news and video signals, and your own surveys into a forecast for one constituency. It goes only to you and is never published.
.POLLBYTES / WE READ EVERY SEAT
POLLYBYTES.ONLINE
Campaign war room
Forecast updates up to polling day, with your ground survey folded in as it arrives.
Ask about a pilot →Party bundle
Many seats at once for a party or alliance, ranked by how winnable each one is, to help with ticket and resource decisions.
Ask about a pilot →Pilot pricing for the 2027 cycle. Write to contact@pollbytes.online for details.
INSIDE THE SYSTEM
Built on evidence, not gut feel.
Every forecast combines four kinds of evidence. Here’s what goes in.
01 · ELECTION HISTORY
Past results
Trained on past election data
Results from previous assembly elections set each seat’s starting point and show how much it tends to swing.
02 · VARIABLES
100+
Variables for every seat
From past margins to local issues, more than 100 variables describe each constituency and feed the model.
NEWSYOUTUBESOCIAL MEDIA
03 · SOCIAL SWING
News + social
Swing factors from news and social media
Local news, YouTube and social posts are filtered for spam and off-seat noise, then scored to catch shifts in voter mood.
Party AParty BUndecided
04 · GROUND SURVEYS
Surveys
What voters say directly
Field surveys update the forecast, weighted by sample size, survey quality and how recent they are.
HOW THE MODEL WORKS
Thousands of signals in. One clear call out.
Pick a block to follow its data through the filter and into the simulation.
OVERVIEW
Four blocks, one forecast
Raw data enters on the left. Each block throws away what it can’t trust, and only clean signal reaches the model, which runs the election thousands of times.
Choose History, Media, Survey or Model above to see what each one does.
BLOCK 1 · RESULTS HISTORY
Where the seat has been
Every past result for the constituency, weighted toward recent elections. Volatile seats get a weaker starting point, so new evidence moves them more.
Filters out: broken rows, duplicates, shares that don’t add to 100%.
Source: official constituency results.
BLOCK 2 · YOUTUBE AND NEWS
What voters are saying now
AI reads video titles, transcripts, comments and local news, keeps only items about this seat, and scores how each one reflects on every party.
Filters out: spam, duplicates, paid promotion, national debate with no local content.
Uses: public content only. AI reads and scores it; it never creates campaign media.
BLOCK 3 · GROUND SURVEY
What voters tell us directly
Field surveys update the forecast, weighted by sample size, survey quality and age. Undecided voters are split in line with everything else we know.
Filters out: small or stale samples lose weight; a survey’s weight halves every 21 days.
Optional: works with your own or a commissioned survey. Without one, ranges are wider.
BLOCK 4 · BAYESIAN + MONTE CARLO
From evidence to odds
The three blocks update one Bayesian estimate. Then 20,000 simulated elections, each with extra polling error, turn it into a win probability and vote-share ranges.
Output: win odds, margin range, swing factors and next moves in a written brief.
20,000simulations per forecast
SEE IT IN ACTION
Ask what every candidate wants to know
A replay of the PollBytes assistant answering a candidate’s questions about a fictional seat. Each answer comes with its live result.
QUESTIONS IN THIS DEMO
Demo answers use a fictional seat and illustrative numbers. Real forecasts are private to each client.
Results history
Recency-weighted baseline and volatility from past results for each seat.
YouTube and news
AI drops spam, duplicates and off-seat noise, then scores local mood.
Ground surveys
Field data weighted by sample size, survey quality and age.
Bayesian simulation
Twenty thousand simulated elections turn evidence into odds.
Private by design
Forecasts go only to the client. We never publish predictions.
Scored, not claimed
No accuracy figures until blind backtests on past results are scored and published.
Built on Claude
Large language models read the media so analysts do not have to.
Briefs, not dashboards
Every forecast ends in a plain brief with swing factors and next moves.
Built for the candidate, not just the party.
Every forecast is made for one constituency and goes only to the client who asked for it.
520 SEATS,
UP + PUNJAB FIRST.
Uttar Pradesh 403Punjab 117
20,000 RUNS,
ONE CLEAR CALL.
Simulated winning margins. Bars right of the tie line are wins.
4DATA BLOCKS
520LAUNCH SEATS
20KSIMULATIONS PER FORECAST
PrivateNEVER PUBLISHED
TRY IT
Play with the model
Move the sliders and watch the Bayesian update at work. A small survey barely moves history; a big, fresh one can overturn it.
Illustrative two-candidate example, not a forecast for any real seat.
CHANCE OF WINNING
–
Expected margin · Rated
OUR ACCURACY STANDARD
We’ll show a record before we claim one.
Indian election forecasts have missed badly before, and trust is low. So PollBytes won’t quote accuracy figures until they’ve been earned and scored in the open.
Scorecard not yet published
- 01
Blind backtests
Run the model on past UP and Punjab results using only data available before polling day.
- 02
Scored, not cherry-picked
Report hit rate, Brier score and calibration: do seats called at 70% really win about 70% of the time?
- 03
Time-stamped forecasts
Lock each client forecast with a time-stamp before polling, so it can be checked later without ever being published.
TRUST AND COMPLIANCE
Confidential by default. Within the rules, always.
Private, never published
Forecasts go only to the client who commissions them. We don’t publish forecasts and never run or share exit polls (Section 126A, Representation of the People Act, 1951).
Silent before polling
No commentary or survey results about any constituency in the 48 hours before polling closes (Section 126, RP Act).
One candidate per seat
We never work for two competing candidates in the same constituency, and client names stay confidential.
Clear, dated invoices
Every engagement is invoiced with dated deliverables. Candidates should check with their election agent how fees fit their expenditure account (Section 77, RP Act).
Consent-based surveys
Surveys collect only what is needed, with informed consent, following the Digital Personal Data Protection Act, 2023.
AI that reads, never fakes
Claude filters and scores public news and video. We don’t create deepfakes or any synthetic campaign content.
Forecasts are probabilities with stated ranges, never guarantees of a result. Questions or data requests: contact@pollbytes.online
Reading every seat,
byte by byte.
Results history, local news, voter videos and field surveys, filtered until only real signal is left.
CONTESTING IN 2027?
Get a private forecast for your seat.
We’re taking a small number of pilot candidates and party teams for the Uttar Pradesh and Punjab elections. Tell us your constituency and we’ll reply.
contact@pollbytes.online