Decide every search dependency before the destination is built

A database attach to Subscription Edition restores the Search service application onto "a new search topology", and "the index is empty" until "a full crawl of the entire indexed corpus" after the last content database (upgrade service applications, checked on 6 November 2026). A copy into SharePoint Online lands in the tenant's own index, where "content is automatically crawled based on a defined crawl schedule" (make sure content can be found, checked on 6 November 2026). What differs is how much configuration comes too, and the configuration is what people use: refiners, promoted results, the home page rollup, the people finder.

Which search behaviours do people depend on, and which of them are configuration that the destination will have to rebuild rather than receive?

The decision this chapter leads to is a disposition for every search dependency (move, rebuild or retire), recorded with where it is configured, who relies on it and the test that proves it, before the destination is built.

Which farms this chapter is for

A SharePoint Server 2016, 2019 or Subscription Edition farm with a Search service application, moving to Subscription Edition by database attach from 2016 or 2019 (Chapter 23) or into SharePoint Online by copy (Chapters 8 and 9). Chapter 10 decides the service application itself and builds it last; Chapter 21 keeps the crawler's Windows sign-in on the default zone. Start from Chapter 2's inventory.

Where search is configured

Search runs as six components (crawl, content processing, analytics processing, index, query processing, administration) and four databases, and only the search administration database holds configuration: "the topology, crawl rules, query rules, and the mappings between crawled and managed properties" (search architecture, checked on 6 November 2026). Result sources and query rules also exist for every site collection and site, the schema for every site collection (result sources, query rules, search schema, all checked on 6 November 2026), and Microsoft's search disaster recovery page for 2016 says a change made in a site collection "will only be stored in the Search Service Application (SSA) administration database" (search disaster recovery, checked on 6 November 2026). No content database holds any of it. The content does carry the display templates, as files in each Master Page Gallery (display templates, checked on 6 November 2026), and the search web parts on its pages. People search also needs a content source crawling the profile store at sps3:// or sps3s:// (deploy people search, checked on 6 November 2026).

Chapter 10 records the topology; read the rest on the old farm before it goes read-only:

# Old farm, SharePoint Management Shell. Reads the Search service application; changes nothing.
$ssa   = Get-SPEnterpriseSearchServiceApplication "<SearchServiceApplicationName>"
$owner = Get-SPEnterpriseSearchOwner -Level SSA
Write-Host "Writing content sources, crawl rules and result sources of $($ssa.Name) to search-*.txt"
Get-SPEnterpriseSearchCrawlContentSource -SearchApplication $ssa | Format-List * | Out-File .\search-contentsources.txt
Get-SPEnterpriseSearchCrawlRule -SearchApplication $ssa | Format-List * | Out-File .\search-crawlrules.txt
Get-SPEnterpriseSearchResultSource -SearchApplication $ssa -Owner $owner | Format-List * | Out-File .\search-resultsources.txt

Parsed, not yet run: try it on a test farm first. Then export each site collection's and site's settings from Site Settings: each file holds only that level's changed settings, never the defaults or the sites below it (search configuration portability, checked on 6 November 2026). Then measure use: the search reports show top queries, queries with no result and "how often query rules trigger" (search diagnostics, checked on 6 November 2026). Export them, because the old analytics "cannot be fully retained" (upgrade service applications, above). We know of no Microsoft report of the pages that show search results, so ask each site owner.

What moves and what is rebuilt

Sources for the table: the pages above, search schema online, classic search administration, hybrid search, publishing sites to communication sites, cross-site publishing, people properties online and cloud hybrid search, all checked on 6 November 2026.

Microsoft's two pages on those exports disagree: the SharePoint Server page says a server's settings do not import into SharePoint in Microsoft 365 (on the server), and the SharePoint in Microsoft 365 page says a Search service application's export does (online), both checked on 6 November 2026. We would rebuild from each file, and test one import before trusting the route.

Which experience will serve each dependency

Modern pages read only part of this configuration. Subscription Edition's modern experience "only shows results from the default result source" and ignores Refinable, Sortable and entity extraction (search experiences, checked on 6 November 2026). Online, Microsoft Search ignores Sortable, Refinable and, where it is fully deployed, custom result sources, and uses bookmarks for promoted results; classic search stays on publishing sites, classic team sites and the Search Center (classic and modern search, checked on 6 November 2026). We would name, for each dependency, the experience that will serve it.

The behaviours that break quietly

None of these fails with an error. Each returns less, or nothing, and looks like content that is not there.

• Rollups. A Content Search web part shows content "as fresh as the latest crawl", major versions only, and nothing from sites marked not to be indexed (Content Search web part, checked on 6 November 2026). Until the index is full, a news rollup reads as no news.
• Search-driven navigation. Cross-site publishing's category pages and the Taxonomy Refinement Panel, Microsoft's "search-driven navigation experience", show only what the index holds (cross-site publishing sites, checked on 6 November 2026).
• Lookups through search. When a managed property is missing, "a custom search application or web part that uses this property no longer works" (search schema, above). Scripts and systems that query search join Chapter 22's list.
• People. The new farm finds nobody until the profile store is crawled; Microsoft recommends crawling it first and waiting about two hours before the first full crawl of the sites (deploy people search, above).
• eDiscovery. An eDiscovery Center finds "any content that's indexed by the Search service application" and nothing else (eDiscovery and in-place holds, checked on 6 November 2026), so content copied online leaves its reach; online, eDiscovery runs in Microsoft Purview (eDiscovery, checked on 6 November 2026).

Hybrid search is narrower than its pages suggest

Inbound hybrid federated search, which showed on-premises results in SharePoint Online, "has been retired as of September 2024"; outbound federation and cloud hybrid search "remain operational" (hybrid search, above). Microsoft retired the Search Content Service behind cloud hybrid search from 30 June 2025; cloud hybrid search continues only on Subscription Edition, from Version 25H1 on some Microsoft pages and the July 2025 public update on others, after onboarding again; 2016 and 2019 farms keep federation only (configure cloud hybrid search, checked on 6 November 2026; cloud hybrid search, above).

Outbound federation has upkeep too: Microsoft's server-to-server procedure ends with an "Update Hybrid federated search" script, for 2013, 2016 and 2019, that installs PowerShell modules Microsoft has deprecated (configure server-to-server authentication, Microsoft Graph PowerShell, both checked on 7 November 2026). We would test that step on a Subscription Edition farm before relying on it.

The roadmap says to allow the endpoints "marked for use with Hybrid Search", and Microsoft's endpoint data lists none. Its IP address and URL web service, read anonymously on 6 November 2026, records endpoint set 33, the "Endpoint to SearchContentService where the hybrid crawler feeds documents", as removed in version 2026043000, and its version 2026103000 lists no hybrid search endpoint. The two disagree, and we do not know which endpoints a re-onboarded crawler uses, so we would not plan on cloud hybrid search until a rehearsal shows on-premises results in the tenant.

Move, rebuild or retire

• Move happens only to Subscription Edition: the restored database's schema, result sources, query rules and result types, and the content's display templates and search pages. Prove it: after the full crawl, export each level again and compare it with the old farm's files.
• Rebuild the rest. On Subscription Edition: the topology, any content source or schedule that did not return, the profile crawl. Online: each level from its export, refiners on built-in properties aliased to the old names, classic result sources and query rules where classic pages stay, bookmarks where Microsoft Search serves, search pages on modern web parts, and a connector for each repository that has to stay findable.
• Retire what nothing uses: a query rule that never fires, a content source whose repository has gone, a search page whose owner agrees in writing. We would not retire on the reports alone: a rule used only at year end may fall outside their window.

Record each disposition before the destination is built, and run its test after the full crawl, before users search.

Out of scope

The build order is Chapter 10; the crawler's sign-in, Chapter 21; systems that call search, Chapter 22; eDiscovery cases and holds, which Chapter 6 leaves to their owners; hybrid as an architecture, Chapter 25. The Subscription Edition migration pack generates the test that a full crawl returns only what each user may open.

Before you continue

• Every search dependency is listed with where it lives: administration database, site collection, site, page or profile store
• The service application is read, and every site collection's and site's search settings exported
• The search reports are exported, and every query rule and search page has an owner or a reason to retire it
• Each dependency is moved, rebuilt or retired, with the experience, classic or modern, that will serve it
• Nothing planned online relies on cross-site publishing, a thesaurus, custom entity extraction, content enrichment or custom security trimming
• Hybrid search is in the plan only after a rehearsal showed on-premises results in the tenant
• After the full crawl, rollups, refiners, people search and every lookup through search returned what the old farm returned

Download chapter 11 (PDF)Same text, same tables, same sources as this page.