SHORT ANSWER
Define a social data query with its scope, period, metrics and comparison. Use 12 examples across summary, account and post levels to find records worth reviewing. Learn how to distinguish cumulative counts from period changes, check engagement-rate calculations, and verify sources, sync times and AI interpretations before choosing a next step.
Article
Give an AI social media query four things: the accounts or posts to analyze, the reporting period, the metrics, and the comparison. Ask the answer to include its data sources, latest sync time and supporting records. A specific question produces a result that is easier to check and follow up.
“Which account is best?” has several possible answers. The account with the most views might not gain the most followers. A post with a high engagement rate might have very few views. Natural-language queries help turn a business question into a defined set of records, then identify the accounts and content worth examining.
A clear comparison makes query results easier to use in daily work. This guide provides a four-part question format, 12 examples and a practical way to check the answer.
1. What four details make a social data question answerable?
Specify the scope, time, metrics and comparison. Replace “Check our US accounts” with “Compare new views for the US TikTok brand-account group across the last two complete weeks. Show the five accounts with the largest changes and the original values for both periods.”
| Detail | What to specify | Example |
|---|---|---|
| Scope | Platform, accounts or posts, plus a group, tag or target market | US TikTok brand-account group, excluding test accounts |
| Time | Start and end dates, reporting time zone and period completeness | October 12–18 against October 5–11, in Beijing time |
| Metrics | Field, cumulative or period value, and any calculation | New views; show the numerator and denominator for engagement rate |
| Comparison | Baseline, comparison group and ranking rule | Against the previous complete week, ranked by change in views |
“US market” needs a definition too. It could mean the account's country, a target-market tag maintained by the team, or viewers located in the US. Those are different attributes. Use account groups and tags when the question is about the team's intended markets.
Add: “Show the available records and identify missing metrics.” For medians or weighted rates, start with comparable records. Export the details and use a consistent calculation method when the query does not provide the calculation directly.
2. How do summary, account and post queries differ?
A summary reveals a broader change. An account query identifies where the change is concentrated. A post query brings the analysis back to specific content. Follow-up questions can move from the overall picture to accounts and then individual posts, keeping the relevant scope and metrics at each step.
| Level | Question it helps answer | What to view alongside it |
|---|---|---|
| Summary | Which platform or group contributed most to this week's change? | Platform or group, account count, dates and metric definitions |
| Account | Which accounts should we review first? | Account identifiers, group, both period values, difference and data coverage |
| Post | Which pieces of content attracted views or interactions? | Post links, publication times, observation lengths and available metrics |
A useful sequence is: “Which platform changed most?” Then ask which accounts contributed, followed by which posts on those accounts deserve attention. Keeping the initial summary separate by platform makes later comparisons easier to interpret.
For the reporting foundation, see How to Unify Cross-Platform Social Media Data. Similar field names do not automatically make metrics interchangeable.
3. What are 12 useful natural-language query examples?
Replace the groups, dates and metrics below with your own. Start with the overall picture, then examine accounts and posts. Ask the answer to list any missing metrics.
The examples use October 12–18, 2026, with October 5–11 as the previous period, both in Beijing time. Use “new” only when period data is available. If only a current cumulative count exists, ask about cumulative performance instead.
Start with the overall picture: four summary queries
| No. | Example question | What to request in the answer |
|---|---|---|
| 01 | List this week's new views and last week's values separately for TikTok, YouTube, Instagram and Facebook. Which platforms changed most? | Each platform's current value, prior value, difference and included account count |
| 02 | Within TikTok, compare new views for the US and German market groups. Include the number of accounts in each group. | A same-platform comparison that retains differences in group size |
| 03 | How many TikTok posts carry the tutorial and product-demo tags? Sum their views, likes, comments and shares separately. | Sample counts, original metrics and the tags used |
| 04 | Show daily changes in views and shares this week, separately by platform. Mark dates with missing data. | Platform-specific trends, with missing dates kept separate from zero values |
Find the accounts: four account-level queries
| No. | Example question | What to request in the answer |
|---|---|---|
| 05 | Which five accounts in the US TikTok brand group had the largest changes in new views? Show both weeks' values. | Account identifiers, both periods and absolute changes alongside the ranking |
| 06 | Which accounts are above their group median for both new views and view-based engagement rate? Define the rate and list the included accounts first. | The sample, formula, medians and qualifying records |
| 07 | Separate newly connected accounts from accounts with complete data for both weeks. Which cannot yet support a week-on-week comparison? | Connection or coverage dates and a reason for each exclusion |
| 08 | Which accounts had relatively high views but small net follower changes? Include their post counts without assuming a cause. | Views, follower changes and publishing volume shown together |
Question 06 helps identify accounts worth reviewing. Assess their performance against account size and business goals. When calculating medians separately, export comparable records and retain the account list and original values.
Examine the content: four post-level queries
| No. | Example question | What to request in the answer |
|---|---|---|
| 09 | Among TikTok posts published by this group this week, which ten currently have the most cumulative views? Include links, publication times and latest sync times. | Post links and data update times, distinguishing cumulative views from weekly gains |
| 10 | Within one account, compare engagement rates for tutorial and product-demo posts. List views and interaction counts for every post, then assess whether the sample is sufficient. | Records, sample size, formula and low-view samples |
| 11 | Compare views, likes, comments and shares for these two post links. Are their publication dates and observed ages similar? | Available metrics and differences in observation length |
| 12 | For authorized TikTok posts with the relevant fields, compare completion rate and average watch time for similar-length videos. List posts with missing fields separately. | Eligible samples, video lengths, returned fields and missing values |
A newly published post and a month-old post have had different opportunities to accumulate views. Compare the first seven days after publication when that data exists. If it does not, show each post's age rather than interpreting unequal exposure time as a difference in content quality.
4. How should a query define engagement rate and period changes?
The example below uses a view-based engagement rate: (likes + comments + shares) ÷ views × 100%. Keep the formula and reporting scope consistent when comparing results. If a report uses another calculation, check its numerator and denominator first.
These three fictional posts illustrate the calculation using the same observation window:
| Post | Views | Likes | Comments | Shares | Interactions | View-based engagement rate |
|---|---|---|---|---|---|---|
| A | 10,000 | 240 | 40 | 20 | 300 | 3.0% |
| B | 8,000 | 200 | 20 | 20 | 240 | 3.0% |
| C | 2,000 | 100 | 10 | 10 | 120 | 6.0% |
| Combined | 20,000 | 540 | 70 | 50 | 660 | 3.3% |
The combined rate is 660 ÷ 20,000 = 3.3%. Averaging the three post rates would produce 4%, which answers a different question. Interaction counts are not deduplicated people. With zero views, this ratio is undefined and should be shown as not calculable.
An absolute period change is current minus previous. A relative change is (current minus previous) ÷ previous × 100%. When the previous value is zero, show the increase from zero to the current value rather than a normal growth percentage.
A post's current cumulative views cannot reveal how many happened last week. That requires period data or comparable historical records. Platform backfills, deletions and corrections can affect differences too. Ask whether the field represents events during the period, a difference between snapshots, or a current cumulative count.
The YouTube metric definitions distinguish different viewing and interaction measures. The TikTok video fields list views, likes, comments and shares separately. Retain the original field meaning before describing a derived calculation.
5. What data can Smart BIAI Agent query?
Smart BIAI Agent supports natural-language queries about authorized account and post performance. Connect the target accounts and authorize the relevant data, then review the available platform metrics.
If a metric is missing, check account permissions, the accounts available to the member and the latest sync time. Browser access to an account does not make every metric available through an API. Analyze public competitor information separately from your own account-dashboard data.
Smart BIAI currently uses Beijing time for reporting. Post-data refresh intervals vary with the age of the content, and platforms can return data late. Check the latest sync time to see whether the results cover the complete period you want to compare.
For “How did our German accounts perform yesterday?”, specify whether “yesterday” follows the system's Beijing reporting day. A German local calendar day requires sufficient time-granular data to recalculate the period. Relabeling a Beijing daily total with another time zone does not perform that conversion.
See Official APIs, Browser Login and Account Isolation for connection differences, and Data and Security for data-use and access principles.
6. How can you check the basis of an AI answer?
A usable answer shows the records, dates and calculations behind its conclusions, as well as what remains uncertain. Check that information before acting on the recommendation.
| Part of the answer | What to check | Useful follow-up |
|---|---|---|
| Data facts | Original values, objects, periods and sources | “Show both periods' original values for these five accounts.” |
| Calculations | Formula, included records, missing values and denominator | “Is this a combined rate or the average of individual post rates?” |
| Interpretation | Possible explanations and information still needed | “Which records support this explanation, and what alternatives remain?” |
| Recommendation | Content or accounts to examine and measures to observe | “Give me the post links to review and the metrics to check next time.” |
“Tutorial posts had a higher engagement rate” can be an observation. “Tutorials caused engagement to grow because they were more useful” still needs testing. Publication times, video lengths, audiences and campaign activity may have changed together. Two reports alone cannot establish that one content change caused all of the growth.
A short check covers six items: account or post list, dates and time zone, metric definitions, latest sync time, calculation basis, and links to the supporting records. If an answer contains only rankings and advice, request those details first.
For the next stage, see the social data diagnostics scenario. Use the query result to select accounts and posts for review, then decide what evidence to collect next.
Start with one account group
Start with one platform, one account group and one complete period. Ask Smart BIAI Agent for the data, comparison basis and supporting records, then follow up on the posts worth reviewing. Expand to more markets and platforms after that query proves useful.
Questions enterprise teams ask
Can AI query TikTok, YouTube, Instagram and Facebook data?
It can query connected, authorized data that is currently available. Fields vary by platform, account type and permissions. Do not assume that the same metrics exist for every account.
Does missing account data mean zero performance?
No. Authorization, field availability, sync status or missing history can prevent a value from being returned. Distinguish a valid zero from missing data rather than filling every gap with zero.
Does “the last seven days” need a time zone?
Yes. Also say whether it includes today. An ongoing day may not be complete or fully synchronized. Two complete periods of equal length usually support a clearer comparison.
Can engagement rates from four platforms be ranked together?
Check the numerator, denominator, observation window and original platform definition first. Start with comparable accounts or content within one platform, then review cross-platform trends separately.
Can AI retrieve any competitor's private operating data?
No. Public material can support research when its source is identified. It does not grant access to private audience, conversion, transaction or account-dashboard data.
Will a query automatically create publishing tasks?
A query does not change the publishing schedule. To use the findings in a post, choose the target accounts, content and time, then create and confirm the task through the available task features.
What should I prepare for a demonstration?
Choose an account group you can access, a complete reporting period and a business objective, such as finding accounts that need review. Bring the scope, time, metrics and comparison so the demonstration can address a real operating question.