Designing Explicit Feedback Prompts
A useful test for explicit feedback is to ask whether the user would have seen a prompt that asked for their answer. If yes, the feedback is explicit, whatever the user chose to answer.
Why Teams Listen to User Reactions
When an AI application produces a result, the people using it usually react in some way. They might click a button, leave a rating, or simply keep using the product. These reactions are a form of user feedback, and they are one of the main ways a team can learn whether an application is working for the people it serves. This article follows one question through the different kinds of feedback: when a user gives feedback, did they know they were giving it, or was the information gathered from something they did anyway?
Asked Versus Observed Feedback
The source groups feedback into kinds based on how it is gathered. Keep the single question from the introduction in mind as you read the three kinds. If the application asked for the feedback, the answer is explicit. If the application inferred it from what the user did, the feedback is implicit. If the information came out of ordinary dialogue with an AI assistant, it is conversational feedback. The taxonomy below shows how these three kinds relate and gives an example of each.
Explicit Feedback: Asking Directly
Explicit feedback is information users provide in response to explicit requests, such as thumbs up or down, star ratings, or yes or no answers.
The defining feature of explicit feedback is that the application asks for it directly. The user is shown a prompt, and their answer is the feedback. Common forms include thumbs up or thumbs down, upvote or downvote, a star rating, or a yes or no answer to a question such as whether the application solved the user's problem.
Explicit feedback is fairly standard across applications because there are only so many ways you can ask a person whether they like something. The formats repeat from product to product. That repetition is one reason explicit feedback is better understood than other kinds: teams know how to design it, and they know how to interpret the answers.
A useful test for explicit feedback is to ask whether the user would have seen a prompt that asked for their answer. If yes, the feedback is explicit, whatever the user chose to answer.
Implicit Feedback: Reading Actions
Implicit feedback is information inferred from user actions. The user does not fill in a rating or answer a question. Instead, the team looks at what the user did and draws a conclusion from it. The source's example is a product recommendation. If someone buys a product that was recommended to them, the purchase suggests the recommendation was a good one. The user never filled in a rating. The signal was inferred from what they did.
Implicit feedback is highly application dependent. What counts as implicit feedback depends on the actions a user can take inside each particular application. A shopping application has purchases to observe, while another application may have entirely different actions. Foundation models enable many new genres of implicit feedback, but the source does not list them. The important idea is the dependence on the product's available actions. Implicit feedback can be used to infer user preferences.
| Feature | Explicit feedback | Implicit feedback |
|---|---|---|
| Is the user asked? | Yes, the application shows a prompt | No, the signal is gathered from actions |
| Typical forms | Thumbs up or down, star rating, yes or no answer | A purchase, or continuing to use the product |
| Dependence on the product | Formats repeat across applications | Depends on which actions the application offers |
| How well understood | Fairly standard, so teams know how to design and interpret it | Application dependent, so it varies more from product to product |
The two kinds differ mainly in whether the user is asked.
Tracing One Reaction Through the Classification
Classifying a single user reaction
A user receives a product recommendation from a shopping application. They do not rate it. Two days later they buy the product. Which kind of feedback does the purchase represent, and what can the team conclude from it?
- 1Identify the request
The application never showed the user a rating prompt or asked a question about the recommendation. Nothing was requested.
- 2Identify the observed action
The user's action was a purchase. The team can observe this action because the application offers purchases.
- 3Classify the feedback
Because the signal was inferred from something the user did, not from an answer to a prompt, the feedback is implicit.
- 4Draw the conclusion
The purchase suggests the recommendation was a good one. The conclusion is an inference, not a rating the user gave.
The purchase is implicit feedback. It suggests the recommendation worked, even though the user never filled in a rating.
Conversational Feedback in a Trip-Planning Chat
The source uses a travel scenario to show how feedback can be read from a conversation. Imagine you are using an AI assistant to plan a trip to Australia, and you ask it to find a hotel for three nights in Sydney. It responds with three recommendations. Suppose you reply that you would prefer a hotel in a quieter neighborhood and ask it to try again. Your reply is conversational feedback. You were not shown a rating control, and you did not press a button. You simply corrected the assistant in ordinary speech.
The conversational interface used by many AI applications makes it easier for users to give feedback. Users can encourage good behaviors and correct errors in the same way they would in everyday dialogue. The language a user uses to give directions to the AI can convey feedback about both the application's performance and the user's preferences. Because the feedback sits inside ordinary speech, the user does not have to be shown a special control to give it.
Using Conversation Feedback in Three Ways
Once feedback has been collected, the source identifies three broad uses. Evaluation means deriving metrics from the feedback, which tells a team how well the application is performing. Development means training future models or guiding their development using the feedback. Personalization means adapting the application to each user. The same feedback can support more than one of these uses.
Mistakes When Sorting Feedback Types
Common mistake Calling any user action explicit feedback
What it looks likeTreating a purchase as a rating the user gave
Why it failsExplicit feedback is information the user provides in response to an explicit request. A purchase is an action that the team observes.
Do this insteadAsk whether the application requested an answer. If not, and the signal came from behavior, it is implicit.
Common mistake Assuming implicit feedback means the same thing in every application
What it looks likeExpecting a shopping app's purchase signal to exist in every other product
Why it failsWhat counts as implicit feedback depends on the actions a user can take inside each particular application.
Do this insteadList the actions the specific application offers before deciding which implicit signals are available.
Common mistake Treating conversational feedback as a missing category
What it looks likeLabeling a correction typed into a chat as implicit because no button was pressed
Why it failsConversational feedback is extracted from conversations. It is its own kind, defined by where the information comes from.
Do this insteadClassify the correction as conversational, and note what it reveals about performance or preferences.
Common mistake Assuming an uses list is exclusive, so that feedback serves only one purpose
Classifying Feedback in Practice
What do you think happens?
A user receives a recommendation and never rates it, but later buys the product. Is this explicit or implicit feedback?
Reveal answer
Implicit feedback
The user was never asked for a rating. The signal was inferred from the purchase, which is an action the user took.
Classify each of the following as explicit, implicit, or conversational feedback, and say what each one suggests to the team: (1) a user clicks a thumbs down under a generated summary; (2) a user types 'Make it shorter and use bullet points' to an assistant; (3) a user keeps opening a feature every day without being asked anything.
Hints
- Ask first whether the application requested an answer. If it did, the feedback is explicit.
- If the information came from ordinary dialogue with the AI, it is conversational.
- If the team inferred the signal from something the user did anyway, it is implicit.
Key Takeaways
- User feedback is gathered in three kinds: explicit, implicit, and conversational, distinguished mainly by how the information was gathered.
- Explicit feedback is information users provide in response to explicit requests, such as thumbs up or down, star ratings, or yes or no answers.
- Implicit feedback is inferred from user actions, and what counts as implicit depends on the actions available in each application.
- Conversational feedback is extracted from conversations, and feedback gathered this way can be used for evaluation, development, and personalization.