How to track brand sentiment without breaking your trendline

The instrument is the measurement

In a one-off study, the question is a means to an answer. In a tracker, the question is the instrument, and an instrument that changes between readings doesn't measure change — it produces two incomparable numbers.

This is survey research's oldest rule and it is broken constantly, almost always by someone acting in good faith. A new researcher joins, reads the questionnaire, notices a genuinely ambiguous question, and fixes it. The next wave's number moves. Nobody can say how much of the move was the wording.

The damage is permanent and retroactive: you now can't compare across the change in either direction.

What has to stay identical

More than the question text, which is the part people remember.

The wording, exactly — including punctuation and any preamble.

The scale — its number of points, its labels, and whether the ends are labelled. Moving from a 5-point to a 7-point scale ends the series, and so does relabelling "Somewhat agree" to "Slightly agree."

The answer order. Option order affects choice, consistently. If you randomize, keep randomizing; if you don't, don't start.

The position in the survey. A question asked third gets different answers than the same question asked fifteenth, because everything before it primes.

The sample definition and the mode. Same recruitment criteria, same screening, same device mix if you can control it.

Any one of these changing is a break, and only the first is usually documented.

Add questions freely, edit them never

The safe move is additive. New questions at the end of the survey don't affect the answers to existing ones, so you can grow a tracker over time without damaging the series you already have.

This is what makes the loop between open-text findings and tracked measurement work. A reason that recurs in your open responses becomes a new closed question in the next wave, and every wave after that measures it — while the existing tracked questions stay untouched.

Put new questions at the end rather than in the topically sensible place. The topically sensible place is in the middle, and inserting there shifts the position of everything after it.

When you genuinely must change it

Sometimes the question is broken, the product changed, or a regulator requires different wording. Two options.

Start a new series. Cleanest. The old line ends, the new one starts, and you present them as two series rather than one. Nobody is misled.

Run both for one wave. More expensive and much better if the trend matters. Field the old and new versions to the same sample in the same wave, measure the gap between them, and use that as a documented bridge. State the bridge every time you present the series across the change.

What you must not do is change it and carry on with one line. That produces a chart with an invisible discontinuity, and somebody will eventually make a decision on the apparent movement.

Cadence is part of the instrument too

Fix the cadence and hold it. Quarterly is right for most brand measures — brand perception moves slowly, and monthly waves mostly measure noise while costing four times as much.

Field in the same weeks each time. A wave that runs in mid-December and a wave that runs in early November are exposed to different worlds, and seasonal variation in sentiment is real and larger than most effects you're trying to detect.

If a wave slips, record it. A gap or a shifted window is a caveat you'll need when explaining a move.

Sample consistency is where trackers quietly rot

Wording gets attention; the sample rarely does. It's the more common cause of a mysterious shift.

Keep the recruitment source constant. Moving from one panel to another, or from a panel to your own customer list, changes the population and produces a step change that looks like a real result.

Watch your quotas and your screening. If your screener drifts — a new question added, a criterion loosened — you're measuring a different group.

Track the composition as a diagnostic. If the mix of tenure, segment, or region shifts between waves, you'll want to know before you explain a number.

Bind a number to the record, not to a deck

A tracker's value compounds only if the readings live somewhere a decision can reference them.

Define the score before you field: which question, which rule, and the minimum number of completes below which a wave posts nothing. Freeze that rule with the instrument. A scorer added to an already-fielded wave, or changed midway, has the same problem as changed wording.

Then let the number land as a metric on the timeline where your decisions are recorded, so a change you made in March sits on the same chart as the sentiment reading in June.

Before you field the next wave