Internal Sites Generated, Audited, And Reviewed
Spin up a department homepage with real employees, policies, forms, and training pulled live — across 48 section types and 5 page archetypes. Batch-create pages, run a content audit that surfaces stale and empty pages, and route everything through the review-before-publish workflow. 13 RISKY write tools sit behind explicit confirmation.
Why Internal Sites Become Stale Within A Quarter
Sites Agent attacks the four specific failures that turn a launched intranet into a graveyard within 90 days — without forcing the comms team to manage 24 pages by hand.
"Create A Department Page" Is A 3-Hour Job, Repeated 12 Times
Hero, quick links, team members lookup, news feed embed, stats, resources. Per department. Twelve departments. The math doesn't work, so most companies launch a portal with three pages and call it done. The agent's batch_create_pages turns 36 hours into one confirmation.
"Pull The Engineering Team Into A team_members Section" Means Manual Copy/Paste
Every page needs real employees, real policies, real forms, real training. Without live lookups into the org chart, Policy Hub, Forms, and Training, pages either go up with placeholder data or with copy/paste lists that drift the moment someone joins or leaves.
Stale And Empty Pages Pile Up Invisibly
A page goes up in January, nobody updates it, by August half the content is wrong and the other half is empty placeholder sections. Without an audit_site tool surfacing the rot, governance is reactive — and by the time someone notices, trust in the intranet is already gone.
AI Pages Get Published Without A Human Reviewing Them
A model-generated page about "leadership philosophy" sounds plausible but reads like LinkedIn slop. Without a review-before-publish workflow that an actual reviewer has to approve, AI-generated content lands in the company's voice without ever passing through a human filter.
Section Types Pile Up But Nobody Knows Which One To Pick
48 section types. The page author wants "list of upcoming training" — is that an event grid, a resource list, a training embed, or a custom callout? Without a guided picker that knows the org's data sources, authors default to free-text blocks and the page loses its live-data connection.
Page Edits Get Made But Nobody Sees The Diff Before Publish
A comms admin updates the "benefits enrollment" page on Monday. The change reverts a paragraph the legal team added last week — and nobody catches it because the publish flow doesn't show a side-by-side diff. The mistake gets discovered after the all-hands email goes out.
Sites Agent At A Glance
Sites AI
AI page generation · org-grounded lookups · review workflow · governance audit.
Inside Sites Agent — The Actual Capabilities
Every block below maps to real tools the agent uses against your Sites data. 50+ tools total — page CRUD, section CRUD, asset library, org lookups, navigation, review workflow, governance audit, versioning, translation. 13 risky writes (publish, delete, batch, find/replace, restore) sit behind explicit confirmation.
Page Generation Grounded In The Org Chart And App Data
Templates pre-wire the right section types for each archetype — operational_hub, executive_overview, onboarding_journey, knowledge_article, department_showcase, culture_page, initiative_launch. Lookups inject real data: lookup_employees for team_members sections, lookup_policies for policy links, lookup_forms for embedded widgets, lookup_training for resource hubs.
- create_page_from_template / create_site_from_template — generate pages and full sites from canonical templates with real content auto-populated.
- 5 visual modes — utility, balanced, editorial, executive, campaign — controlling spacing and layout density.
- 7 page archetypes — operational_hub, executive_overview, onboarding_journey, knowledge_article, department_showcase, culture_page, initiative_launch.
- Live org lookups — lookup_employees, lookup_policies, lookup_forms, lookup_training inject current data, not snapshots.
Batch Create + Review Workflow Before Publish
batch_create_pages generates department, team, or initiative pages in one call — every page lands as a DRAFT, never auto-published. The review workflow (request_page_review → approve_page_review / reject_page_review / request_page_changes) routes drafts through an assigned reviewer before publish_page takes the page live.
- batch_create_pages — multiple pages in one call from templates. RISKY · confirmation-gated. All output is DRAFT.
- 4-stage review workflow — request_page_review, approve_page_review, reject_page_review, request_page_changes. Only the assigned reviewer can act.
- publish_page / publish_site / schedule_page_publish — three publish paths, all RISKY, all confirmation-gated.
- Versioning + restore — list_page_versions, compare_page_versions (read), restore_page_version (RISKY, destructive).
Content Audit, Self-Maintaining Pages, And Find/Replace — Governance That Stays Current
audit_site checks every page for stale content, empty sections, missing heroes, thin content, and governance findings. Beyond staleness, self-maintaining content watches the live records each page references — policies, forms, people, events — and when one changes or is removed, it raises a governance finding and stages a one-click refresh the owner approves. suggest_improvements returns per-page recommendations. find_and_replace sweeps a string across the entire site (preview by default, apply on confirmation).
- audit_site — comprehensive content audit (stale pages, empty sections, missing heroes, thin content, governance findings).
- Self-maintaining content — a daily scan flags pages whose linked policy / form / person / event changed or was removed, and stages a one-click refresh; the live page is unchanged until the owner approves.
- list_stale_pages, list_expiring_pages — governance reads with configurable lookbacks/lookaheads.
- {"suggest_improvements / find_and_replace / bulk_update_sections — per-page recommendations and sweep-style edits; sweeps are dry-run by default, apply" => "true + confirmation to commit."}
Outcomes Teams Can Measure
The agent's job is to compress page-creation cycle time, keep org-data lookups live (not copy/paste), surface stale and empty content before employees do, and make sure nothing AI-generated gets published without a human review. Measure against your pre-agent baseline.
- Page-creation cycle time — hours from "we need a department homepage" to a published page passing review, vs the manual baseline.
- Org-data freshness — share of team_members / policy / form / training sections sourced from live lookups vs static copy.
- Stale-page burndown — pages last-updated >90 days either refreshed or archived per month, via audit_site + list_stale_pages.
- Review override rate — share of AI-generated pages rejected or requested-changes in the review workflow; treats too-high as a quality signal.
- Find/replace adoption — sweep-style edits (rebrands, role-title updates, link migrations) done in one confirmation instead of page-by-page.
- AI reach — how often Ask AI answers through each page and how many employees it served, from the built-in AI citation analytics; the modern measure of whether a page earns its keep.
13 Risky Writes, Always With A Human In The Loop
Sites Agent's RISKY_TOOLS list is 13 entries — the longest in the agent fleet because Sites is a write-heavy authoring surface. Every entry requires explicit user confirmation; the destructive ones (delete_page, delete_site, restore_page_version, remove_sections, find_and_replace with apply=true) carry warnings the user has to acknowledge.
- Publish gates — publish_page, publish_site, schedule_page_publish. All RISKY, all confirmation-required.
- Destructive deletes — delete_page, delete_site, remove_section, remove_sections, remove_site_member. Cannot be undone; the confirmation prompt makes that visible.
- {"Sweep-style writes — find_and_replace, bulk_update_sections, batch_create_pages, clone_page. Dry-run preview by default; apply" => "true + confirmation required to commit."}
- Restore is destructive — restore_page_version replaces current sections with the snapshot; the confirmation prompt explicitly warns the current content will be overwritten.
WHAT TEAMS TRY INSTEAD
The four alternatives — and none of them know your org chart, your policies, or your section types
Most intranet owners try one of these four first. None stick because none can pull from Policy Hub, Forms, Training, and the org chart in a single call — and none gate publish through a review workflow that an actual reviewer signs.
ChatGPT, Claude, or Copilot draft + paste into the CMS
Generic AI writing pages a human still has to wire up
- Sites Agent inserts real team_members from the org chart, real policies from Policy Hub, real forms — not Lorem-ipsum prose
- batch_create_pages produces 12 department pages in one confirmation; copy/paste tops out at one page at a time
- audit_site surfaces stale and empty sections post-launch; generic AI has zero visibility into what's already shipped
SharePoint Copilot, Notion AI, or Confluence AI
Vendor-trapped page AI inside a single document silo
- Reaches into shifts, training, recognition, and HRIS to embed live data — vendor AI can't see across silos
- 48 archetype-aware section types (events, KB embeds, team grids, training lists) — not just rich-text blocks
- Review-before-publish workflow with a side-by-side diff; SharePoint AI publishes the moment you click apply
Custom CMS + RAG retrieval layer
A platform team building Sites and the AI on top of it
- Already shipped — section schema, permissions, review workflow, and 13 RISKY-gated writes are in the platform
- Inherits per-business confirmation thresholds, AiApiLog retention, and the existing permission model for free
- New section types and tools land via the agentic framework — no rewrite when "we need a training section" comes in
The comms team writes 24 pages by hand
The default when AI tools don't fit the section model
- Generates a department homepage with real employees, policies, and forms in one prompt — not a 3-hour Notion build
- Cuts content audit from a quarterly spreadsheet exercise to a single audit_site call that flags stale and empty pages
- find_and_replace + bulk_update_sections do platform-wide sweeps that would take comms ops a week of manual edits
PLATFORM ADVANTAGE
Sites Agent inherits everything the platform already runs
A standalone intranet builder has to plumb each of these. Sites Agent gets them for free because the platform already owns them.
Live org chart and HRIS
team_members sections render real reports-to relationships and titles from the org chart — no copy/paste, no drift when someone joins or leaves.
Policy Hub and Forms inline — and self-maintaining
Sections embed live policies and forms by reference. When a referenced source changes or is removed, the page flags itself and stages a one-click refresh for the owner — no quarterly link audit.
48 archetype-aware section types
Hero, quick links, KB embeds, training grids, event lists, stats, resources — the agent picks the right section for the data, not free-text blocks.
Review-before-publish workflow
Every AI-generated page lands as a draft. A human reviewer sees the side-by-side diff and signs off before the page goes live.
Audit trail with version restore
Every section edit logs to the audit trail; restore_page_version is itself RISKY-gated so a rollback is a deliberate act.
RubyLLM model tiering
Section drafts run on standard tier; bulk sweeps and audits run on small. Cost per page stays bounded as the org scales to hundreds of pages.
INDUSTRY FIT
Industries where an intranet builder pays back in weeks
Sites Agent shines where intranet scope is wide, content owners are scarce, and stale pages erode trust fastest.
Retail
Batch-creates a store-team page per location with the real district manager, store associates, and the local SOP set — no district ops admin writing 200 pages.
Healthcare
Department homepages link the live on-call schedule, current clinical SOPs, and the training each role must hold — no broken policy links during a Joint Commission visit.
Manufacturing
Plant-floor sites embed live safety bulletins and the current shift roster; audit_site flags empty quality and EHS sections before the corporate review.
Financial Services
Compliance landing pages auto-embed the latest policy versions with audit-trail attribution; FINRA examiners see who reviewed what when.
Hospitality
Per-property pages with the GM, the day's events, and the housekeeping SOP set — generated once, refreshed automatically as the team rotates.
Public Sector
Department portals stay inside the tenant boundary with full audit logging; no SharePoint Copilot traffic leaving the FedRAMP-eligible deployment.
WHY MANGOAPPS WINS
An embedded builder beats a generic AI, a SharePoint add-on, or a custom CMS on every axis
The argument finance, security, IT, and comms all share — and the one a horizontal AI or single-vendor assistant structurally cannot answer.
Cheaper than the alternatives
No SharePoint Copilot add-on, no Notion AI seat, no custom CMS build, no extra comms headcount to author and audit 24 department pages.
More secure
Every section read and write logs to AiApiLog; 13 RISKY-gated writes (publish, delete, batch, sweep, restore) sit behind explicit confirmation prompts.
Easier to deploy
Already deployed if Sites is enabled. Turn the agent on, point it at the org chart, and batch-generate department pages the same afternoon.
Easier to use
Lives inside Ask AI and the page editor — no separate chatbot, no copy/paste between a generic AI and a CMS, no second permissions model.
Easier to manage
Per-business confirmation thresholds, model tiers, and section-type allowlists sit in the same admin console as every other app's settings.
Easier to extend
New section types ship as tools, not rewrites. A new "training grid" section reaches the agent the same week it reaches the section catalog.
AI is actually better
A generic or vendor AI can draft a page. Only Sites Agent can also embed the live team, current policies, due training, and an upcoming-event grid — and route it through review before publish.
At our head office within HR, marketing, and finance, we use ChatGPT quite a bit, and MangoApps AI Helper is a useful tool for us. As I go around to the stores, I’ve been showing them how the AI Intranet Helper can find things and answer plain language questions about our resources, and they’re quite excited about that.
Alison van Zijl
Head of Human Resources
Full House
Customer Success
Related Customer Stories
Video Case Studies
Video Case Studies
Video Case Studies
Customer Case Studies
Customer Case Studies
Customer Case Studies
Frequently Asked Questions About Sites Agent
50+ tools spanning site CRUD, page CRUD, section CRUD (48 section types), asset library (browse + semantic search + upload), org lookups (employees, policies, forms, training), navigation, review workflow, versioning + restore, translation, content audit, suggestions, and sweep-style writes (find/replace, bulk_update_sections, batch_create_pages). 13 of those tools are in RISKY_TOOLS and all require explicit confirmation.
No. publish_page, publish_site, and schedule_page_publish are all in RISKY_TOOLS and require explicit user confirmation. Generated pages default to DRAFT status, and the review workflow (request_page_review → approve_page_review) routes through an assigned human reviewer before publish.
lookup_employees pulls from the org chart, lookup_policies from Policy Hub, lookup_forms from Forms, lookup_training from Training. Sections render against the live records, so an employee joining or leaving a department shows up in the team_members section without anyone editing the page.
Yes — self-maintaining content. Each page tracks the live records it references (policies, forms, people, events). A daily scan detects when one of those sources changes or is removed, raises a governance finding, and stages a one-click refresh. The published page is never altered until an owner approves it, so pages stay accurate without editing themselves silently.
Yes. Separate from the authoring agent, an "Ask this page" reader assistant lets an employee ask a question and get an answer grounded only in the content they're permitted to see, with the source page cited. Admins can also see how often Ask AI reaches an answer through each page via AI citation analytics in the Sites dashboard.
find_and_replace defaults to dry-run (preview only) — the agent returns the matches it would change without changing them. To commit, the admin passes apply: true, which triggers the RISKY confirmation prompt before any text is modified. bulk_update_sections follows the same dry-run-by-default pattern.
Page-creation cycle time (hours from request to published, vs manual), org-data freshness (share of sections sourced from live lookups), stale-page burndown per month via audit_site, review override rate on AI-generated pages, and find/replace adoption for sweep-style edits. Compare against your pre-agent baseline.
Let's Talk
Since 2008, we've been building the employee platform for the frontline, earning the trust of 2 million+ users and an NPS of 78.
Why Choose Us?
- AI-Ready Platform: One intelligent place for every employee and workflow.
- Top Security: HITRUST, ISO & SOC 2 certified.
- Exceptional UX: Delightful on mobile and desktop.
- Proven Results: 98% customer retention rate.
Trusted by Legendary Companies:
Prefer to explore first? Ask AI about Sites AI Agent →