Which Sources Shape Software Recommendations in AI Search

The Week 29 Evidence Boundary

The Week 29 public edition contains 6,837 cited pages from 3,307 source domains across 2,595 products, 80 categories, and eight software hubs. Compared with Week 28, that is an increase of 199 cited pages, 188 domains, and 120 products. Those totals describe the edition’s public evidence footprint; they are not a product score or a measure of recommendation quality.

This report goes one level deeper by examining one representative category from each hub. It uses only the cited-page totals, domain totals, and source classes displayed in the immutable Week 29 and Week 28 archive views. It does not expose named cited pages or domains.

Eight Markets, Eight Different Source Mixes

In Sales, CRM expanded from 106 cited pages and 71 domains in Week 28 to 135 pages and 80 domains in Week 29. Its largest displayed Week 29 class is Other at 67, followed by Vendor-owned at 57. The prior edition showed Vendor-owned at 43, Other at 41, and Editorial at 16; Week 29 shows Editorial at 6, Review platforms at 7, and Community at 1.

In Marketing, Marketing Automation moved from 80 cited pages and 63 domains to 83 pages and 71 domains. Other remains the largest displayed class, changing from 43 to 42, while Vendor-owned changes from 30 to 41. The Week 29 view also displays Editorial at 1 and Community at 1.

In Support, Help Desk increased from 91 cited pages to 102 while remaining at 68 domains. Vendor-owned is still the largest displayed class, rising from 50 to 62, while Other changes from 38 to 42. The Week 29 view displays Community at 2, Editorial at 1, and Review platforms at 1.

In HR, Recruiting (ATS) grew from 77 cited pages and 58 domains to 92 pages and 67 domains. Its Week 29 classes are led by Vendor-owned at 47 and Other at 44, followed by Review platforms at 2, Editorial at 2, and Community at 1.

In Finance, Accounting moved from 95 cited pages and 72 domains to 101 pages and 76 domains. Other remains the largest displayed class but changes from 65 to 54, while Vendor-owned changes from 20 to 44. Week 29 also displays Editorial at 3, Documentation at 1, and Review platforms at 1.

In Productivity, Project Management expanded from 87 cited pages and 59 domains to 99 pages and 71 domains. Other changes from 55 to 51, while Vendor-owned changes from 18 to 47. Editorial changes from 12 to 3, and the Week 29 view displays Community at 2.

In IT & Security, Identity & Access (SSO/IAM) records the largest domain increase in this eight-market sample: from 59 domains to 76, alongside a rise from 77 cited pages to 97. Other and Vendor-owned were tied at 37 in Week 28; Week 29 displays Other at 52, Vendor-owned at 45, and Review platforms at 3.

Engineering moves in the opposite direction. Observability & APM contracts from 79 cited pages and 54 domains to 60 pages and 47 domains. Vendor-owned remains the largest displayed class but changes from 52 to 40, while Other changes from 29 to 21. Community and Review platforms each appear at 1 in Week 29.

Where the Public Corpus Expanded

The eight-market sample does not move uniformly. CRM adds 29 cited pages and 9 domains; Recruiting adds 15 pages and 9 domains; Project Management adds 12 pages and 12 domains; and Identity & Access adds 20 pages and 17 domains. Together they show that edition-level growth can be distributed very differently across individual markets.

Marketing Automation adds only 3 pages while adding 8 domains. Help Desk adds 11 pages with no change in domain count. Accounting adds 6 pages and 4 domains. These are distinct public evidence patterns: more pages do not necessarily arrive with a proportional increase in the number of domains.

Where the Mix Changed More Than the Total

Marketing Automation’s total grows modestly, yet its displayed Vendor-owned count rises by 11 while Editorial falls by 8. The aggregate can establish that the class counts changed; it cannot identify which pages entered or left the corpus, or whether those changes affected any product’s position.

Accounting shows another pronounced mix change inside a smaller total expansion. Its displayed Vendor-owned count rises by 24, while Other falls by 11 and Editorial falls by 6. Because the source classes are not mutually exclusive, these changes should be read as movements in displayed counts, not as shares of a fixed whole.

Help Desk adds 11 cited pages without adding a domain. Its Vendor-owned count rises by 12, while the changes in the other displayed classes are smaller. That is evidence of a changed public class profile, but not evidence that one source class caused a recommendation or rank.

Where the Corpus Contracted

Observability & APM is the only market in this sample with fewer cited pages and fewer domains than the prior edition: 19 fewer pages and 7 fewer domains. Its displayed Vendor-owned and Other counts also fall. The contraction is reportable as an evidence-footprint change; it should not be described as a decline in product quality or category importance.

What the Aggregates Can—and Cannot—Explain

Public source aggregates answer three useful questions: how many cited pages appear in a category, how many domains those pages represent, and which broad source classes are displayed. Comparing immutable editions also shows whether those counts expanded, contracted, or changed shape from one week to the next.

They do not reveal the identity of an individual cited page or domain, and they do not establish why an assistant produced a recommendation. The displayed source classes are not mutually exclusive, so their counts can exceed the page total and must not be converted into percentages. Evidence volume and source mix remain descriptive measures, not causes of rank.

Methodology

This analysis uses the Week 29 archive as the current immutable record and the Week 28 archive as the immediately previous official comparison. Every category observation comes from the matching Overall “Sources & Mentions” view in both editions, and every conclusion stays within those public aggregates.