Measurement · Conversation
Comment sentiment in influencer campaigns
Comments can reveal enthusiasm, confusion, objections and questions that a view count cannot explain. Read them with a consistent method, and keep the conclusion tied to the comments you actually reviewed.
Explore the guideStart with this
Separate the comment's overall tone from what it says about the brand or product.
Show the sample size, selection method and uncertain comments beside any percentage.
Treat buying interest as a useful signal. It does not establish that a sale happened.
What does comment sentiment mean?
Sentiment describes the attitude expressed in text, commonly grouped as positive, negative or neutral. It can be assessed by people or estimated using a text-analysis tool. A comment may also be mixed or too unclear to classify. Do not force every response into a confident positive or negative label.
The target matters. A person might praise the creator's filming while criticising the product's packaging. An overall-tone label and a brand-specific label answer different questions.
Define the sample before reading for a result
Write down the posts, capture date, comment ordering and inclusion rules. A review of the first 100 newest comments is different from a review of the 100 comments ranked most relevant by a platform. Replies, creator comments, duplicate messages and promotional spam can also change the picture. Decide how to handle them before calculating percentages.
Commenters are a self-selected group of people who chose to respond. They do not stand for everyone who watched, everyone who follows the creator or the market as a whole. A bigger sample may provide more material to review without fixing how it was selected.
- State the number of comments reviewed, not only the post's displayed comment count.
- Say whether replies and the creator's own comments are included.
- Keep an unclassified category for unclear language, context or meaning.
- Use the same rules when comparing posts or repeat the comparison on a common basis.
Use a small, repeatable set of labels
Start with separate fields for tone, subject and useful action signals. A positive comment can also contain a question. A negative comment can reveal an obstacle that the next campaign should address. Making these separate fields avoids losing useful detail in one sentiment score.
Before reviewing a large batch, have a second reader classify a small selection using the same definitions. Discuss the disagreements, refine the rules and retain an uncertain category. If you use automated labels, review a sample manually and describe the result as an estimate rather than treating each label as a fact about a person.
| Field | What to record | Why keep it separate? |
|---|---|---|
| Overall tone | Positive, negative, neutral, mixed or unclear | The tone may concern the video or creator rather than the product |
| Brand or product reaction | Relevant positive/negative reaction; no clear reaction; unclear | General praise should not automatically become brand approval |
| Topic | A short defined category, such as fit, setup, price or delivery | Topics reveal what the audience needs explained |
| Question or buying-interest signal | The stated question or explicit interest, with the rule used | Interest and uncertainty can appear together |
Illustrative comments · written for this guide
Read the target, not just the tone
| Comment | Possible reading | Useful next action |
|---|---|---|
| Love this video. Does the bottle fit a bike cage? | Positive about the video; product-fit question | Show the fit and relevant dimensions |
| I want one, but is there a smaller size? | Buying-interest signal with a size condition | Clarify the available sizes |
| Amazing, another charger that needs its own charger. | Possible sarcasm; needs context | Review the context before assigning sentiment |
Put the count beside the percentage
A percentage should say what was divided by what. Positive comments as a share of all reviewed comments differs from positive comments as a share of classifiable comments. Either can be reported if it is clearly named, but changing the denominator between campaigns makes the comparison misleading.
Report uncertainty as part of the result. A large unclassified group means your summary covers less of the reviewed conversation. Do not silently drop it or describe the classified subset as everyone who saw the post.
Illustrative worked sample
The same sample can produce 60% or 75%
Suppose you review 100 comments: 60 positive, 10 negative, 10 neutral and 20 unclassified. Positive comments are 60% of all 100 reviewed comments. Among the 80 classified comments, they are 75%.
A clear report would say: 60 positive comments out of 100 reviewed; 20 could not be classified. If you also show the 75% figure, label it positive share of classified comments. Neither figure means that 75% of viewers like the product.
Keep buying interest separate from sales
A question about ordering, product availability or a required feature can identify a useful buying-interest signal. It remains text expressed in a comment. The person may never buy, may already be a customer or may need an answer before making a decision.
Use explicit classification rules and retain the raw count. If your team excludes irrelevant or unclear comments to define a meaningful-comment subset, state that definition and the resulting denominator. Avoid converting a comment-intent percentage into an estimated number of orders without separate evidence.
Illustrative worked example
Useful intent, limited conclusion
A review identifies 12 explicit buying-interest comments among 80 relevant, interpretable comments. That is 15% of that defined subset. It supports investigating which product questions encouraged interest; it does not support reporting 12 customers, 12 sales or a 15% campaign conversion rate.
Use the conversation to improve the next brief
Group repeated questions and objections, then translate them into content requirements. A setup question may call for a slower demonstration. Repeated uncertainty about what is included may call for an unboxing shot. Praise for an explanation may justify testing the same format with a different product question.
Keep the original campaign goal in view. Comment volume, positive tone and useful product discussion are separate outcomes. Your final recommendation should identify the observed pattern, the limitation in the evidence and one change worth testing. Share aggregate findings where possible rather than building unnecessary profiles of individual commenters.