Why do different outlets report the same poll differently?

Because a poll produces many numbers, and which one is led on is an editorial choice — before any question of bias arises, the same data genuinely supports several accurate headlines.

What a single poll contains: headline voting intention, changes since the last poll by that company, approval ratings, issue salience, responses to specific questions, and breakdowns by demographic group. Any of these can be the story.

The legitimate sources of difference:

Which number is led on. A party up two points on last month but down five on the year supports two accurate and opposite headlines.

Which comparison is used. Against the previous poll by the same company, against a different company, or against an average — each gives a different change.

Whether changes are significant. A movement within the margin of error is not evidence of change at all, and reporting it as a shift is the single most common error in poll coverage. With a typical sample, changes of a couple of points are noise.

House effects. Different pollsters get systematically different results because of differing methods — how they weight, how they handle undecided voters, whether they apply a likelihood-to-vote filter, and how they phrase questions. Comparing one company's poll to another's produces spurious changes, which is why poll averages exist and why serious coverage compares like with like.

Question wording, which can move results substantially on the same underlying issue.

Which subgroup is highlighted. Subgroup samples are much smaller with much larger margins of error, so a dramatic finding among a demographic slice is frequently noise.

How to read poll coverage properly:

Look for the sample size, dates, method and commissioner.

Prefer trends over single polls. No individual poll means much; a consistent direction across several does.

Note who commissioned it, since commissioned polls may ask questions designed to produce a result.

Treat "shock poll" framing sceptically — outliers occur by chance, and one in twenty polls will fall outside the margin of error by definition.

Polls measure opinion at a moment, and are not forecasts.