Ranking the available responses
Stay organized with collections
Save and categorize content based on your preferences.
Many signals help Assistant rank the available responses, including the
following main factors:
- How sure Assistant is that it understood what you asked.
- Whether a response is actually available for a particular interpretation of
your request.
- How satisfied previous users were with a particular response to similar
requests.
- How recently the response was created, to help you get a variety of fresh,
high-quality responses.
- How well a response works on the device that you are using. For example,
responses that are optimized for devices with screens are likely to be
ranked lower on speakers. On a partner device where Assistant is built-in,
if you ask for something specific to that device, such as changing the
volume or playing a movie, the device manufacturer may handle some or all of
the response, according to what the partner decides is the best user
experience.
- What else you asked for recently. For example, if you say, “Hey Google,
start a five minute timer,” and then shortly after say, "Hey Google, stop,"
Assistant may use your earlier request to understand what you mean.
- What you are currently doing on your device, such as which app you have open
when you ask Assistant for help, or what Assistant is already helping you
with. For example, if you are listening to music and you say “Hey Google,
skip,” Assistant jumps to the next song. Similarly, if you are part of the
way through making a restaurant reservation using Assistant, it prioritizes
completing the reservation above other possible responses.
- In limited circumstances, some high-quality responses may be manually
curated to rank higher, for the purpose of improving the user experience.
For example, to help users get timely information about COVID-19 and
mitigate misinformation that could risk public safety, we may curate
information from authoritative sources like the World Health Organization
and governmental health authorities.
These main factors for ranking are weighted differently based on how you’re
engaging Assistant and your personal preferences, as explained below.
Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates.
Last updated 2024-09-18 UTC.
[null,null,["Last updated 2024-09-18 UTC."],[[["\u003cp\u003eGoogle Assistant prioritizes responses based on understanding your request, response availability, user satisfaction, and response recency.\u003c/p\u003e\n"],["\u003cp\u003eDevice type, recent interactions, current activity, and occasional manual curation also influence response ranking.\u003c/p\u003e\n"],["\u003cp\u003eResponse ranking considers factors like how well Assistant understands your request and how satisfied previous users were with similar responses.\u003c/p\u003e\n"],["\u003cp\u003eAssistant uses context from previous interactions and current device activity to better understand your requests and provide relevant responses.\u003c/p\u003e\n"],["\u003cp\u003eIn some cases, high-quality responses are manually curated to ensure accuracy and user safety, especially for critical topics like public health.\u003c/p\u003e\n"]]],[],null,["# Ranking the available responses\n\nMany signals help Assistant rank the available responses, including the\nfollowing main factors:\n\n- How sure Assistant is that it understood what you asked.\n- Whether a response is actually available for a particular interpretation of your request.\n- How satisfied previous users were with a particular response to similar requests.\n- How recently the response was created, to help you get a variety of fresh, high-quality responses.\n- How well a response works on the device that you are using. For example, responses that are optimized for devices with screens are likely to be ranked lower on speakers. On a partner device where Assistant is built-in, if you ask for something specific to that device, such as changing the volume or playing a movie, the device manufacturer may handle some or all of the response, according to what the partner decides is the best user experience.\n- What else you asked for recently. For example, if you say, \"Hey Google, start a five minute timer,\" and then shortly after say, \"Hey Google, stop,\" Assistant may use your earlier request to understand what you mean.\n- What you are currently doing on your device, such as which app you have open when you ask Assistant for help, or what Assistant is already helping you with. For example, if you are listening to music and you say \"Hey Google, skip,\" Assistant jumps to the next song. Similarly, if you are part of the way through making a restaurant reservation using Assistant, it prioritizes completing the reservation above other possible responses.\n- In limited circumstances, some high-quality responses may be manually curated to rank higher, for the purpose of improving the user experience. For example, to help users get timely information about COVID-19 and mitigate misinformation that could risk public safety, we may curate information from authoritative sources like the World Health Organization and governmental health authorities.\n\nThese main factors for ranking are weighted differently based on how you're\nengaging Assistant and your personal preferences, as explained below."]]
