A poll is a snapshot, not a forecast
A poll measures what a sample of people say at one moment. People change their minds, some do not vote, and some decide on the day. That is why a poll is best read as a rough picture of the present, not a prediction.
The margin of error
Every poll has one, often around three percentage points. If one party is on 40 percent and another on 38, the race is too close to call, because both numbers could easily be a point or two higher or lower. Headlines that announce a 'lead' within the margin of error are overstating the evidence.
Who was asked, and how?
A poll is only as good as its sample. Look at how many people were asked, whether they were chosen to represent the population, and whether it was conducted by phone, online or in person. A poll with a small or self-selected sample, like a website vote, is not a reliable guide.
The wording of the question
Small changes in wording can shift answers a lot. 'Do you support cutting waste in government?' gets a different result from 'Do you support cutting public services?'. Good reports print the exact question.
Look at the trend, not one poll
A single poll can be an outlier. Averages and trends across many polls are far more reliable than any one result. When one poll shows a sudden dramatic swing, wait to see whether others agree.
Turnout and who actually votes
Many elections are decided by who turns up. A group that is large in polls but votes less often counts for less on the day. Remember that polling people is difficult, and that is why surprises happen.