recruiter-prompt-bank · 20 vendors
What AI screeners say they check: 14, 9, 7, 4, nothing.
The note
Twenty vendor cards in a five by four grid. Four print their lines, 14, 9, 7 and 4. Four others are stamped none or not disclosed.
Those four are the AI screening and sourcing vendors, of the 20 whose public pages I read on 28 September, that gave no concrete criterion that I could find. Three had no public example of a criterion, Fetcher, Hiretual and HiredScore, and Paradox's record says its exact fields are not disclosed.
The most specific answer among the rest was Workable's: the bank's record, a paraphrase of its help page, has its screening agent evaluating candidates on 14 matching criteria categories. Zoho Recruit's Zia listed 9 signals. Teamtailor's Co-pilot named 7 inputs it reads. Greenhouse Talent Matching named 4 fields: skills, experience, job titles, industry. For the rest, the lists run from 1 to 9 items.
The check a buyer can run from outside passes for most of the field: there is a page, and it names something. What most of them name is a list of what gets read; two records in the bank, Workable's and Manatal's, say required criteria weigh more than preferred ones, and only Workable's gives a ratio. Eightfold's six factors come without weights, and the scores sit on several scales: Eightfold 0 to 5, Gem in percentages, Bullhorn 0 to 100, Lever a percent match.
Limits. The counts hold for what the pages said on 28 September, and they can change tomorrow. 33 of the 43 rubric records are paraphrased from the pages, not quoted, and the lists in the bank mix inputs with criteria as they were merged during collection. Lever's help page failed to load, so Lever rests on its marketing page. I did not test any product.
Would a list of what a screener reads, with no weights attached, pass your procurement review as a description of how it decides?
Where the numbers live
Pages change; no product tested. 33 of 43 rubric records are paraphrases. Eightfold: six factors, 0 to 5 score.