Sitecore

SitecoreAI DAM Grounding: Approve the Knowledge Doc Before Agents Write

SitecoreAI DAM plus Scrunch Knowledge Studio can turn approved Content Hub assets into grounding for AI tools. The ops risk is shipping unreviewed knowledge docs. Here is the approve-before-generate gate I use.

Public AI tools invent brand facts from the open web. Knowledge Studio can ground them in your DAM. Only if a human still approves the knowledge document.

Executive Thesis

Brand answers that leave your stack still get checked by customers, partners, and regulators. When chatbots and content agents only see the public internet, they invent specs, mix old SKUs, and quote retired claims. That is not a model problem first. It is a grounding problem.

Sitecore's early-access path (Content Hub changelog, 4 Sep 2026) pairs SitecoreAI DAM with Scrunch Knowledge Studio so approved DAM sources become structured knowledge documents. Agents can then use those documents as grounding input when they generate AI-optimized content.

This Field Note is for platform engineers and content leads who already run Content Hub. I assume you can provision apps and open DAM. The work that fails in production is not the connector. It is skipping the review-and-approve gate, then wondering why the agent still sounds like a brochure from 2019.

I would rather ship grounding late than ship wrong product claims early. Early access is fine. Unreviewed knowledge docs in a live channel are not.

Basics: DAM, Knowledge Docs, and Grounding

SitecoreAI DAM (Content Hub in many orgs) holds the brand truth you already fight to keep current: product specs, FAQs, rich media, legal claims. Much of that never hits the public web. Search engines and chatbots cannot see it, so they invent.

Knowledge Studio turns selected DAM sources into a structured knowledge document. Scrunch agents can review drafts and flag sensitive content, factual conflicts, and errors. A human still approves. After approval, the document becomes grounding input for generation. When DAM updates, you refresh sources and regenerate.

Basics before you demo this to marketing: name an owner for the knowledge doc, pick a small source set (not the whole DAM), and decide which customer channels may consume grounded output.

Rename the mental model if it helps your team: Knowledge Studio output is a compiled index of brand truth, not a free-writing playground. Compiled indexes get rebuild jobs and owners. Playgrounds do not.

  • Connect SitecoreAI DAM to Scrunch with a named owner.
  • Generate from a tight approved source set first.
  • Require human approve before any agent uses the doc.
Team reviewing documents together at a desk
Knowledge docs need the same review culture as publish workflows. Photo: mohdnaeem.com blog stock (theme assets)

What Breaks Without an Approve Gate

If a draft knowledge document can ground production agents, you have created a second publish surface with weaker habits than Pages. Editors know not to push XM without review. They do not yet treat Knowledge Studio the same way.

Sensitive content flags from Scrunch agents help, but they are not a substitute for a content owner who knows which claim is still legal. Conflicting specs across two PDFs will not always resolve themselves into the newer truth.

The failure mode I care about: marketing asks the agent for a product answer, the agent uses a draft knowledge doc that still contains a discontinued SKU, support spends the afternoon unwinding quotes. Grounding without approve is just faster hallucination with your logo on it.

Write the rule on the project wiki in one line: draft knowledge docs may train internal experiments; only approved knowledge docs may ground anything a customer can read or hear.

Deep Dive: Operating Model That Survives Friday

Start with one product family or one FAQ set. Connect DAM. Generate one knowledge document. Run agent review. Have the content owner approve in writing (ticket, email, or Studio approve action). Only then enable grounding for a single channel (for example internal draft assist, not public chat).

Refresh policy matters as much as first generate. When product updates a DAM PDF, someone must regenerate and re-approve. Put that on the same calendar as content publish. Orphaned knowledge docs are worse than no grounding because they look official.

Access control: the Scrunch connection should use least privilege. Do not hand a marketing intern a write path into every DAM collection. Mirror your Hub roles.

Measure wrong answers the same way you measure bad publishes. Count them. Name the source doc version. If you cannot point to the approved document id behind a bad answer, you do not have grounding. You have hope.

  1. Pick one source collection and one owner.
  2. Generate the knowledge document from those sources only.
  3. Review Scrunch flags; fix DAM sources if the flag is real.
  4. Human approve; record date and owner.
  5. Enable grounding for one channel; measure wrong answers for a week.
  6. Schedule refresh after every material DAM update.
Notebook with checklist on a desk
Approve date and owner belong on the knowledge doc, not in chat memory. Photo: mohdnaeem.com blog stock (theme assets)

How This Fits XM Cloud and Sidekick Habits

Sidekick drafts are still not published copy. The same sentence applies here. Grounded generation can draft faster. Publish and send still need humans. If Sidekick or SitecoreAI assistants start consuming Knowledge Studio output, keep the human publish gate in Pages and in any customer-facing channel.

Content Hub remains the system of record for assets. Knowledge Studio is a derived artifact. Treat it like a search index: rebuild when sources change, monitor quality, and retire stale docs.

If your org is still on early access, say so in the runbook. Do not promise enterprise-wide grounding before the approve workflow is boring.

When leadership asks for AI everywhere, answer with the gate diagram: DAM sources → knowledge doc → approve → grounded draft → human publish. Skip any box and you are back to public-web improvisation with a Sitecore badge.

Failure Matrix

Wrong answer from grounded agent: check whether the knowledge doc is approved, whether sources are current, and whether the question sat outside the source set. Expand sources carefully; do not dump the whole DAM.

Agent ignored grounding: confirm the channel actually uses the approved document. Demo modes often skip production wiring.

Sensitive content slipped through: stop the channel, re-review the doc, fix DAM, regenerate, re-approve. Treat it like a bad publish.

Also watch for stale media rights. A DAM asset can be approved for internal use and still be wrong for public AI answers. Your knowledge-doc owner should know the difference, or pull legal in before approve.

Further Reading

Read Sitecore's Content Hub changelog note on grounding AI tools with SitecoreAI DAM and Scrunch Knowledge Studio (4 Sep 2026). Confirm early-access status with your Sitecore account team before you promise timelines.

Pair this with your existing Sidekick review checklist: draft assist is allowed, customer send is gated. Grounding changes the source of truth for facts. It does not remove the send gate.

Keep doc.sitecore.com Content Hub and SitecoreAI FAQ pages in the same bookmark folder as this Field Note. Product names and entitlements move; your approve gate should not.

If Scrunch flags conflict with what product believes is true, fix the DAM first. Do not paper over conflicts inside the knowledge doc. The next regenerate will bring the fight back.

Actionable Checklist

Print this checklist and initial each line before you enable grounding in any customer channel.

After the pilot week, either expand sources deliberately or kill the channel. Do not leave a half-grounded assistant running because the demo looked good in a meeting.

  • Name a knowledge-doc owner and backup.
  • Connect DAM with least-privilege access.
  • Generate from a small approved source set.
  • Resolve Scrunch review flags in DAM first.
  • Human approve with date recorded.
  • Enable one channel only for the pilot week.
  • Schedule regenerate-after-DAM-update.
  • Keep Sidekick or assistant send gates intact.
Laptop open on a standing desk in an office
Pilot one channel. Expand only after wrong-answer count drops. Photo: mohdnaeem.com blog stock (theme assets)

Further reading

Actionable checklist

  • Owner named
  • Sources scoped
  • Approve recorded
  • One channel piloted
  • Refresh scheduled