AI document reading in Vendarity: citations, abstention and human review

Vendarity can now read an uploaded supplier document and propose its key fields with page citations. How it works, how we tested it and what it never does.

In short

Workspaces can now switch on AI document reading. When a reviewer asks, Vendarity proposes ten fields from an uploaded document, each with a page citation or an explicit “not found”; ambiguous dates are left for a person, and a reviewer confirms every field. In a live evaluation on 100 synthetic documents, suggestions matched 472 of 475 required fields and left all 25 ambiguous or absent fields for a person.

Reading a certificate is the slow part of supplier document review. Vendarity can now do that first read for you: when a reviewer asks, AI reads an uploaded document and proposes its key details, each with the place in the document it came from. Your reviewer still confirms every field and makes every decision.

What does the AI propose?

Ten fields, the same ones a reviewer confirms by hand: document type, legal entity, certificate or policy number, issuer, standard and version, site or scope, and the issue, valid-from, expiry and review dates. For each field, the AI either proposes a value with its page and the line or snippet it came from, marks the field as not found, or marks it as ambiguous.

How are citations checked?

Where a document has embedded text, Vendarity checks that the cited lines really contain the quoted snippet and that the proposed value appears in it, and withholds any suggestion that fails. Dates must be supported by the snippet, and a numeric date whose day and month order is not explicit is never resolved for you. On scanned pages, the citation points the reviewer to the page image to check.

How did we test it?

Before switching it on, we ran a live evaluation on 1 October 2026 against 100 synthetic documents, 20 for each launch country, covering varied business documents, missing fields and ambiguous dates. Our release bar was 95% exact match on required fields.

Live evaluation, 1 October 2026 (synthetic test set)
MeasureResult
Required fields matched472 of 475 (99.4%)
All present fields matched972 of 975 (99.7%)
Ambiguous or absent fields left for a person25 of 25
Incorrect non-empty suggestionsNone in this test set

The three misses were expiry dates the AI withheld even though the document stated them, which costs a reviewer a few seconds rather than putting a wrong date on record. These results describe a synthetic test set, not your documents, which is exactly why every field still needs a person.

How do you switch it on?

  1. An Owner or Admin opens Settings → Organization, reads the processing disclosure and enables AI for the workspace.
  2. A reviewer opens an unreviewed submission and asks for suggestions. Uploading a file never sends it to AI on its own.
  3. The suggestions appear beside the document. The reviewer confirms or corrects each field, giving a reason for any correction, and then decides.

What does the AI never do?

  • Accept a document, approve a supplier or change any status.
  • Skip review: acceptance stays blocked until a person has confirmed or corrected every field.
  • Act on instructions inside a document: the text is treated as data and the model has no tools.
  • Run beyond your plan: Trial includes 100 AI pages, Core 500 and Growth 2,000 a month, with no overage charges.

What happens to the documents?

The page previews and embedded text of the requested file go to Google’s paid Gemini API, with stored interactions, search and model tools disabled. Google’s paid-service terms say it does not use prompts or responses to improve its products; it may keep limited logs to detect abuse, and processing can take place outside your country. The original suggestions, their citations and your reviewers’ corrections stay in your workspace’s evidence history. More detail is on the trust page.

Sources

  1. How Vendarity works · Vendarity
  2. Trust, data and limits · Vendarity
  3. Pricing and AI page allowances · Vendarity
  4. Gemini API additional terms of service · Google
  1. Guides5 min read

    How to use AI to read supplier documents safely

    AI can read certificates and insurance documents in seconds. How to use it without losing control: citations, abstention, ambiguous dates, human review and data terms.

  2. Checklists & templates5 min read

    What to check on a certificate of insurance before you accept it

    A certificate of insurance is a snapshot, not a policy. What to check on a COI or certificate of currency before you accept it, and what it can never prove.

  3. Standards & regulations6 min read

    How to check an ISO 9001 certificate is real and current

    A practical check for any ISO 9001 certificate a supplier sends: who issued it, whether that body is accredited, what the scope covers and whether it is still valid.

Collect and review supplier documents in one place.

14-day trial. No card needed to start.

Start your 14-day trial