The employee
Opens an article to solve a problem. Scans for the answer in under thirty seconds. Gives up and opens a ticket if the article is poorly structured, incomplete, or off-topic.
Content standards
ServiceNow Articles
ServiceNow articles serve two readers at once — the employee who opens them directly and Ask Hubert, who retrieves passages from across the knowledge base. These standards make every article work for both, without compromise. The fix is content quality, not better AI.
Two readers, one article
A ServiceNow article is read by a person and a machine in parallel. The standards on this page exist so the same article works for both, because writing two versions of the same knowledge is worse than writing none.
Ask Hubert
Retrieves passages — not full articles — and combines them to answer questions. Surfaces confusing fragments when sections aren’t self-contained or scope is vague. Trust erodes when the agent answers from broken context.
When an article fails one reader, it usually fails the other. A wall of unstructured text loses the employee scanning for an answer and gives Hubert nothing to retrieve. Vague scope confuses the reader and misdirects the agent. The fix is the same in both cases: better content.
Two layers of quality
Every article is judged on two layers. Both have to be right. A well-structured article in broken English is unusable. A beautifully written article with no headings, no metadata, and a vague title will never be found.
Content design
How the article is built.
Determines: Whether Hubert can retrieve the right passage and whether a human can scan to the answer.
- Structure, headings, metadata
- Chunking behavior and scope
- Links and cross-references
Editorial quality
How the article reads.
Determines: Whether the answer, once found, is trustworthy and usable.
- Voice, tone, plain language
- Completeness and accuracy
- Inclusive and accessible writing
Five article types
Every article belongs to exactly one of five types. The type determines the template, the required sections, and the expected scope. If an article seems to fit two types, it’s probably two articles — split it.
How-to
The reader needs to complete a specific task.
Sections
- Before you begin
- Steps
- Confirm it worked
- If something goes wrong
- Related articles
Reference
The reader needs to look up a fact, value, policy detail, or definition.
Sections
- At a glance
- Details
- Related articles
Troubleshooting
The reader has a symptom and needs to diagnose and resolve it.
Sections
- Symptom
- Likely causes
- Resolution steps
- If the issue persists
- Related articles
Policy
The reader needs to understand a rule, expectation, or governance requirement.
Sections
- What the policy says
- Who it applies to
- Why it exists
- How to comply
- Related articles
FAQ
A cluster of short, related questions has the same audience and topic.
Sections
- Question 1 …
- Question 2 …
- Related articles
The non-negotiables
Five rules carry most of the retrieval weight. Get these right and an article works for both readers, even before the editorial pass.
Scope statement — non-negotiable
Every article opens with a one-sentence scope statement directly under the title: “This article applies to [audience] who need to [task or outcome].” It is the single most important sentence for retrieval — Hubert uses it to decide whether the article is relevant to a question. Missing or vague scope is the most common reason a well-written article never surfaces.
Self-contained H2 sections
Hubert chunks articles by heading. A section that begins with “as mentioned above” retrieves as a broken fragment. Test by asking: if this section were the only thing the reader saw, would it still make sense? If no, rewrite or merge.
Semantic headings, in order
H1 is the article title (ServiceNow adds it). H2 marks major sections. H3 marks sub-steps. Don’t skip levels. Headings should be answerable as a search query — “Reset your password” beats “Procedure.”
Metadata is content
Keywords are the most-skipped field and the highest-leverage one. Include synonyms employees actually use — if the official term is “authentication failure,” also list “locked out.” Owner, audience, last reviewed, and review cycle are mandatory: articles without an owner drift.
One topic per article. Cross-link, don’t duplicate.
If an article fits two types, it is probably two articles. Duplication creates drift — two versions diverge, and Hubert retrieves whichever it indexed first, sometimes the stale one. Pick the canonical article and link from the others.
Editorial guardrails
These extend the rules in Voice & Tone. Voice stays the same — JM Family writing to a competent, busy reader. Tone shifts with the article’s job.
| Article type | Tone |
|---|---|
| How-to | Confident and instructive. Imperative verbs ("Open Workday. Select Time Off."). |
| Reference | Neutral and factual. No persuasion, no hedging. |
| Troubleshooting | Calm and reassuring. The reader is already frustrated — don’t add friction. |
| Policy | Clear and respectful. State the rule, then the rationale if it helps. Never lecture. |
| FAQ | Conversational and concise. Match the phrasing of the question. |
Three SN-specific extensions
- Acronyms get spelled out on first use, every time. PTO, OOO, SETF, RGM, ITNow — spell out, parenthesize, then use the acronym. Hubert handles this better when the expansion lives on the page.
- Replace internal jargon and code names unless the code name is the actual product name employees see. “ITNow” is fine; the project code name for the team that built it is not.
- Don’t assume universal context. Not every reader is in the corporate office, in the US, or on a managed laptop. If a step depends on a context, name the context.
Pre-publish quality checklist
Every article walks the checklist before it ships, and walks it again at each review cycle. The standards work only if the gate holds. If any unchecked item names a missing fact or wrong scope, send the article back; if it’s a formatting issue, the reviewer fixes it during approval.
Article type and scope
4 itemsOne topic per article. Scope statement under the title.
Title
4 itemsSentence case, eight words or fewer, no acronyms unless more recognized than the expansion.
Structure
6 itemsEvery H2 is self-contained. The article ends with a clear next step, not in mid-air.
Metadata
7 itemsKeywords include synonyms employees actually use. Owner is a named person, not a mailbox.
Content design
9 itemsNo section depends on earlier context. Critical info is not locked inside screenshots.
Editorial
11 itemsActive voice, plain words, acronyms spelled out on first use, errors don’t blame the user.
Accessibility
4 itemsColor isn’t the only signal. Text doesn’t live only inside images.
Hubert readiness
3 itemsSections answer real questions on their own. User-phrase synonyms appear in keywords or body.
Final review
4 itemsSME other than the author has reviewed for accuracy.
One anchor example
A single comparison that captures the difference these standards make for retrieval.
Won’t retrieve
Password reset information
(No scope statement.)
Title is a noun phrase, not a question. Scope is implicit. Hubert has no signal for when to surface this article.
Retrieves cleanly
Reset your corporate password
This article applies to JM Family employees who need to reset their corporate Windows password.
Title leads with the action. Scope names the audience and the task. Keywords like “locked out” sit in the metadata so Hubert matches on how users actually ask.
First surface, more coming
ServiceNow is the first content surface with domain-specific standards. The same pattern (a standards file, a quality checklist, and templates by article type) is repeatable for Viva Engage, SharePoint, corporate email, and other internal communications channels. The framing — content quality is the floor of AI quality — applies anywhere Hubert or another agent will read content downstream.
The reasoning behind extending the system this way is captured in Decision Log ADR-008.