Clean up entities before trusting Share of Voice

Why entities need cleanup

The platform automatically reads through every AI response in a wave and pulls out the entities it finds: brands, attributes, and other named things worth tracking, then classifies each one into a category. This saves you from having to tag anything by hand, but it is an automated read of open-ended text, and automated reads are not perfect.

The hardest part of this job is deciding where the line sits between "this belongs in your comparison set" and "this is unrelated noise." That line gets blurrier in categories the AI does not know especially well, where it may surface a name that is only loosely connected to what you are actually comparing, or miss the difference between two genuinely separate things. Left uncorrected, this noise sits inside the same numbers you use to judge performance.

For example, a wave analysing kitchen appliance brands might surface a certification mark or an industry body as its own entity: it is a real, recurring term in the responses, but it is not a competing brand, so it belongs in "other" rather than in your comparison set. Left in place, every mention of it dilutes the denominator and makes every real brand's share look smaller than it should.

Why this matters for Share of Voice

This matters most for Share of Voice, since that metric is a share of a denominator built from every entity the platform found. An unrelated entity sitting in that denominator quietly dilutes everyone else's share, making real brands look smaller than they are relative to each other. The same noise affects the analysis tab's relationship view too: an entity that keeps appearing alongside a real brand purely because it was never cleaned up can look like a meaningful pattern when it isn't one.

Cleaning up after a wave

After a wave completes, it is worth a short visit to the entity management page before you trust the numbers.

• Remove noise. Look for anything that clearly does not belong in your comparison and move it to "other," which removes it from the Share of Voice denominator without deleting the underlying data.
• Merge duplicates. Check for duplicate variants of the same entity — two spellings of the same brand name, or a shortened and a full version of it — and merge them together so they are counted as one entity instead of splitting its mentions across two rows.

Once you're done, refresh your report so the cleanup actually shows up in your export rather than sitting behind stale, cached numbers.

The same "don't trust it blindly" instinct applies to the citations behind a claim, not just the entities behind a number — see checking source quality for how to spot a confidently-stated but unsupported citation.