Direct answer
61 items. That is not a table printed in the standard: it is what you get when every clause in chapters 4-9 of GB/T 47746—2026 (31 clauses in all) is parsed and translated into rows a company can tick off — 48 shall + 4 should + 9 may, of which 5 are deal-breakers (failing any one of them means failing the whole self-check). Clause count and item count are not one to one: clause 5.2.2.6 (“automatic handover in specific scenarios”) becomes five separate rows for scenarios a)-e). The figures 53 and 59 that circulate online come from an earlier coarse keyword pass; they were corrected on 2026-09-19 and should no longer be cited. The current, single figure for this site is 61.
Where 61 comes from: count the clauses, then split the items
Our figure is built in three steps, and each step is documented so it can be checked:
- List every clause in chapters 4-9: 31 clauses in total.
- Grade each clause by its wording — shall (应), should (宜), or may (可) — and record whether it is verifiable.
- Translate each clause into the actions a company can actually tick off. One clause often means several actions: clause 5.2.2.6 (“automatic handover in specific scenarios”) splits its five scenarios a)-e) into five rows. After splitting, there are 35 checkable nodes and 61 self-check items.
So 61 is not “the number of clauses” and not a number the standard prints. It is the size of an operational checklist derived from the text — which is what makes it usable, and also what makes the derivation rule worth writing down.
Correction note (2026-09-19)
Earlier versions of this checklist were built from a coarse keyword match over clause nodes and were reported as 59 items, or as 53 mandatory items. Both were wrong, in two specific ways:
- Five items were graded too strictly. Clauses 5.2.3.1, 7.1.1 (two items) and 7.1.2 (two items) are worded as may (可), not shall. They are permissible options, not obligations.
- Two clauses were not covered at all. Clauses 5.2.3.2 and 5.2.3.3 — the two human-to-AI handover cases — were missing from the checklist and have been added.
The corrected and only figure this site uses is 61 items = 48 shall + 4 should + 9 may, including 5 deal-breakers. Pages and materials that still say 59 or 53 are pre-correction text.
Why the shall / should / may split matters
Grade follows the wording of the clause, not the chapter it sits in: shall is an obligation (48 items), should is a recommendation (4 items), may is a permitted option (9 items). With limited resources, clear the 48 shall items first and then use should and may to improve experience — do not invert that order.
The same chapter can contain all three grades, which is why a per-chapter count is a poor proxy for a per-grade one.
The five deal-breakers
These are scenarios a)-e) of clause 5.2.2.6 — the five situations where the service must be able to hand a customer over to a human automatically. Failing any one of them fails the whole self-check, and this is the area complaints and inspections actually hit.
How to use the table
- Do not treat a clause number as an item number — the two counts do not match.
- Grade by the wording of the clause, not by the chapter heading.
- Reuse it as an internal quarterly template: update only status, evidence, owner and conclusion each round, keeping clause numbers and item numbers stable so periods can be compared. Realign against the text when the standard is revised or the system changes materially.
📄 Printable self-check checklist (free)
The 61-item checklist above is available as a single A4 page you can print and hand out:
View the one-pager ｜ Download PDF (one A4 page)
Prefer to tick the boxes yourself? Use the online self-check tool and get a gap report.
Key facts
| Total self-check items | 61 = 48 shall + 4 should + 9 may, including 5 deal-breakers |
| Clause coverage | All 31 clauses of chapters 4-9 (5.2.2.6 split by a)-e) → 35 checkable nodes) |
| Previous (retired) figures | 59 items / 53 mandatory — impression-based, corrected 2026-09-19 |
| Correction basis | 5 items re-graded to may (5.2.3.1, 7.1.1 x2, 7.1.2 x2); 2 missed clauses added (5.2.3.2, 5.2.3.3) |
| Clauses vs items | Not one to one — clause 5.2.2.6 alone becomes 5 items |
| Is 61 official? | No. The standard prints no self-check table; 61 is our derived figure |
Sources
- GB/T 47746—2026, chapters 4-9 clause-level parse (31 clauses; 35 checkable nodes)
- GB/T 47746—2026, clause 5.2.2.6 (the five scenarios behind the 5 deal-breakers)
Follow-up questions
Is 61 an official number?
No. The standard publishes no self-check table. 61 is the figure we derived by parsing it clause by clause, and the splitting rule plus the basis for every row can be checked on this page and in the open list.
Why 61 items and not 31 clauses?
31 is the number of clauses in chapters 4-9; 61 is what those clauses become once translated into executable self-check rows. One clause can carry several obligations — 5.2.2.6 alone produces five.
Where did 53 come from, and why is it still everywhere?
It came from an early coarse keyword pass over clause nodes and spread faster than the correction did. It is defensible as a rough checklist size only if you label it as derived; quoted as the standard's own figure it fails the first time someone opens the text.
Can I reuse this checklist?
Yes — that is what it is for. Keep the clause numbers and item numbers stable, update status / evidence / owner / conclusion quarterly, and realign with the text after a revision.
Published by SavantCat · source https://savantcat.cn/en/answers/ai-service-standard/53-mandatory-items.html · please keep the attribution when quoting.