Playbooks
How to Measure Brand Sentiment: A Repeatable Review Method
Define the conversation and coding rules before reporting a sentiment percentage. Preserve mixed and unclear cases, source examples and changes in collection.
Measure brand sentiment by collecting a defined set of relevant text, applying explicit coding rules and reporting what that set contains. Preserve the original examples and uncertain cases. A tidy positive percentage is useful only when readers can understand its denominator and how the labels were assigned.
Define the eligible conversation
Record the brand aliases, product context, languages, platforms and period. Exclude unrelated name matches and document how repeated or syndicated content is handled. Choose the coding unit: post, comment, sentence or aspect. A post with several replies is not automatically several unique customers, and a person can contribute multiple items.
Capture the source URL, publication date and observation date. A retrieved set may be shaped by search ranking, public access and who chose to participate. State these limits rather than describing the sample as the voice of the whole market. Broader customer attitudes may require additional research methods.
Use labels that preserve the decision-relevant detail
Start with definitions and examples for positive, negative, neutral, mixed and unclear. Separate product aspects where helpful: a comment may praise design and criticize shipping. Google Cloud’s sentiment documentation explains how opposite emotions can cancel in an aggregate score. A near-neutral average therefore deserves context, rather than a confident claim that opinion is unremarkable.
- Positive: the item expresses praise about the defined target.
- Negative: it expresses criticism about the defined target.
- Neutral: it contains relevant information without a clear positive or negative evaluation.
- Mixed: the target or different aspects receive both praise and criticism.
- Unclear: available text or context does not support a confident label.
Worksheet: retain the evidence behind the label
| Record for each eligible item | Review use | |
|---|---|---|
| Source and dates | Original URL, publication date and observation time. | Inspect context and preserve the comparison period. |
| Target/aspect | Brand, product or issue actually discussed. | Avoid assigning someone else’s criticism to your brand. |
| Code | Label, definition and reviewer note. | Keep uncertain and mixed items visible. |
| Theme | Setup, shipping, pricing or another observed issue. | Turn tone into a specific research question. |
| Correction | Original label, revision and reason. | Maintain a reproducible review trail. |
Worked example: report counts before a headline percentage
An invented sample contains 40 eligible comments: 12 positive, 10 negative, 6 neutral, 8 mixed and 4 unclear. Positive items are 12 ÷ 40 = 30% of all eligible items. If someone excludes mixed and unclear items, the denominator becomes 28 and the same 12 positives become about 42.9%. Neither calculation should be reported without its definition. The first breakdown preserves more of the sample’s ambiguity.
The team reads negative and mixed items and finds an export-setup theme. It prepares a demonstration and routes potential product issues to the appropriate owner. It does not claim that 25% of all customers are unhappy, because the 10 negative comments are part of a selected sample. The supported action is to investigate the observed issue.
Review the classifier and the reviewers
Inspect a varied set across labels, languages and borderline cases, rather than only the most-liked comments. Record disagreements and refine the codebook. Brand24’s Mentions documentation includes manual sentiment correction; the same principle applies to an AI-assisted brief. There is no universal accuracy estimate established by this article.
Compare periods with a change log
Reuse the relevant scope and label definitions, choose a review cadence your team can sustain and log collection changes. More negative items can reflect changed volume or access rather than a broader shift in opinion. Segment by source and theme before concluding. In current Superlurk workflows, tracking uses manual checks; preserve dated evidence rather than assuming a search starts continuous collection.
Use Comment Insights for available public text and the listening guide to assign the resulting actions. Keep business outcomes separate from the sentiment labels.
Sources and research notes
Reviewed on 11 October 2026 against official sentiment definitions and label-review documentation. The counts and brand scenario are invented arithmetic examples; they do not describe a live retrieval, classifier benchmark or customer survey. How we approach social research.
- Google Cloud Natural Language: API basics
Explains sentiment score and magnitude, including neutral-looking scores that contain mixed emotions.
- Brand24: Mentions tab overview
Explains source links, mention filtering and manually editable sentiment labels.
Frequently asked questions
How do I measure brand sentiment?
Define the brand, source scope, period, inclusion rules and coding unit. Review the text, preserve mixed/unclear cases and report counts with the sample limitations.
Can I compare sentiment month to month?
Only when collection and coding are sufficiently comparable. Changed access, queries, languages or label rules can create an apparent shift.
Does a sentiment percentage describe all customers?
No. It describes eligible coded items in the collected sample, unless a separate defensible sampling design supports broader inference.
Tools used in this guide
Written by
The Superlurk Team
We build Superlurk — a cited social search engine across 24 platforms. We write about social media search, insights, and marketing.
Keep reading
What Is Sentiment Analysis? Labels, Scores and Context
Sentiment analysis estimates expressed tone in text. A neutral-looking score can conceal mixed feelings, and a model label does not establish the whole audience’s opinion.
Comment Analysis: Code Questions, Objections and Context
Analyze a bounded social comment sample with a coding worksheet. Preserve thread context, distinguish observations from interpretation and review ambiguous themes.
Social Listening Guide: Build a Repeatable Research Program
A useful listening program specifies what it collects, how evidence is reviewed and who acts. Preserve sampling limits and measurement changes across recurring reviews.
How to Track Competitors on Social Media: A Manual Review Routine
Set up a repeatable competitor review with named queries, a baseline, verified changes and decision owners. Use sampled comparisons without promising automatic alerts.