Should I search for more Zhihu viewpoints or continue one question's answers?
To discover new related content, search keywords with zhihu.web.fetch_article_search_v3. To read more viewpoints on a question you already selected, continue that question’s answer list with zhihu.web.fetch_question_answers. Save the conditions of each route separately, then compare all material against the same research question.
For example, in research on daytime projector use, keyword search finds other discussions about shading, screens and situations. Once you have selected the question “Is a home projector usable during the day?”, the question’s answer list supplies other viewpoints on that question.
Which route fills the gap?
Start with keyword discovery
To find answers published within the last month:
Send the JSON to POST https://api.beatapi.io/v1/social-data/call with normal BeatAPI Bearer authentication. In zhihu.web.fetch_article_search_v3, limit and offset are strings. vertical accepts answer, article or zvideo; an empty vertical leaves the content type unrestricted. sort: upvoted_count puts higher-voted content first for reading; a high vote count does not prove an answer correct or applicable to everyone.
One search can return answers from several questions. Confirm each answer’s parent question from the actual source before grouping; similar titles do not prove the same question. Save the search terms, type, ordering, time filter and collection time. When you change any condition, start a separate observation.
Start a known question’s answer list
In zhihu.web.fetch_question_answers, limit and offset are integers, and the first cursor and session_id are empty. The example uses the default order:
For time ordering, set order to updated. zhihu.web.fetch_question_answers reads one fixed question; searching the question’s title is a different operation and does not replace the answer list.
How do you preserve continuation?
- Save the first default-order batch for question 123 as
q123-default-01.json. - For the next batch, keep the same question and order, pass the
cursor,session_idand continuation values fromq123-default-01.json, and save the response asq123-default-02.json. - To read the same question in
updatedorder, start a separate set atq123-updated-01.json, so the two sets stay distinguishable.
zhihu.web.fetch_article_search_v3 continues differently: the search has a string offset and a search_hash_id for deduplication, so continue the search with the values the search actually returned. Do not mix the answer-list session inputs into the search. If no usable continuation value is returned, keep the current scope and review the saved material; do not build a cursor from page counts.
How do you assess another page’s value?
Check whether a new batch adds new answers or repeats, and new conditions or repeated statements. Match repeats by actual answer ID. When an answer’s text or metrics change, save the later observation separately.
For example, if the first batch describes only shaded rooms, a bright-living-room experience in the next batch shows where the conclusion applies. If the next batch only adds similar accounts under the same conditions, organize the existing evidence instead of paging further. Record an empty result and a failed read separately; neither one proves that the question has no other viewpoints.
Prompt for the assistant:
Identify the gap first: search for new questions or continue this question’s answers. Preserve conditions and sources, explaining which environment or counterexample each batch adds. Separate repeats and unresolved objects before deciding whether to expand.
Budget the actual search pages or answer pages with current pricing. Saved material can be analyzed again without a new request.

