Knowledge base maintenance: who owns each article, when to review it, what to retire
Knowledge base maintenance: one owner per article, review dates set by risk, light approval and a plan for leavers, with what 56 public bases show readers.
Knowledge base maintenance keeps articles accurate through one accountable owner per article, a review date set by the harm a wrong answer would do, and a way for any reader to flag or fix an error. Approval controls what gets published, archiving removes what no longer applies without destroying the record, and a departing owner's articles pass to a named successor before the account closes. No trusted source gives evidence for any review interval, so each team chooses one and writes down why.
Many public knowledge bases do not even show a date or an owner. In October 2026 we checked 56 public knowledge bases and help centres, 30 in English and 26 in French. On 30 of them (54%), neither of the two articles we opened showed a date or an owner. Internal knowledge bases cannot be checked from outside, so anything this page says about them is an inference from public ones.
What 56 public knowledge bases show their readers
The sample was drawn by a written rule across six sectors, from public bodies to charities. For each base we opened three pages and recorded what a reader sees before clicking anything.
| What the reader sees (on at least one of two articles) | Bases | Share |
|---|---|---|
| A search box (on any of the three pages) | 44 | 79% |
| A date: published, updated or reviewed | 25 | 45% |
| A feedback path on the article, such as "Was this helpful?" | 19 | 34% |
| An owner: a named author, team or department | 8 | 14% |
| A label for the type of content or its audience | 6 | 11% |
| Version or validity information | 4 | 7% |
| Neither a date nor an owner | 30 | 54% |
With 56 bases, the margin is wide: the 95% confidence interval for the 54% figure runs from 41% to 66%. The figure is also a lower bound, because a base counts as showing a date or an owner when only one of its two articles does. More English bases than French ones showed neither (19 of 30, against 11 of 26), but with samples this small the gap may be chance.
A reader is more likely to find a search box than to learn who wrote a page or when anyone checked it. Owners are far rarer than dates. Where one appears, it is short: "From: Department for Education" on a government help centre, a line on a local council page naming the team responsible, an author's name on a software help centre. No retail, banking, utility, transport or charity base named an owner.
Dates come from the software more than from editorial choice. Of the 18 bases built on dedicated help-centre software, 13 showed a date, because their templates print one. Of the 38 built on general websites, 12 did. A "last updated" stamp printed by a template tells the reader that someone edited the page, not that anyone checked it. Missing dates are a bigger gap than old ones: 31 bases showed no date at all, while on five, the most recent date was more than two years old.
The most complete pages in the sample were two patient leaflets from a National Health Service (NHS) hospital trust in England. Each printed the department responsible, a production date, a review date and a leaflet code with a version number. One leaflet was produced in September 2023 with a review date of September 2026. When we opened it in October 2026, that date had passed. A printed review date is a promise that only a process can keep.
Content ownership: one accountable owner, shared upkeep
Every trusted source we read says that someone must be named. The guidance on the knowledge clause of ISO/IEC 20000-1, the service management standard, lists the roles to fill, and one of them is to identify each source of knowledge and give it an owner. The NHS website makes a clinical lead accountable for the evidence behind its content and a content director accountable for editorial standards. APQC (American Productivity & Quality Center), a non-profit benchmarking body, says that someone must be responsible for validating captured knowledge, for keeping or disposing of it, and for correcting it when it is out of date.
KCS (Knowledge-Centered Service, renamed Knowledge-Centered Success in April 2026) warns against the opposite failure. Collective ownership is one of its core concepts, and its Practices Guide names individual ownership of articles as a cause of duplicates: people hesitate to edit an article that belongs to someone else.
The two positions fit together. One person is accountable for each article's accuracy, and anyone who uses the article can flag a problem or fix it. KCS calls the second part "flag it or fix it": fix what you are trained and confident to fix, flag the rest, and make sure every flag reaches someone who can act on it. KCS sends flags to a subject matter expert; in the model on this page, that person is the article's owner.
A person or a team?
GOV.UK gives pages to organisations rather than to people: the organisation tagged to a page can edit it. That survives staff changes, but a team cannot answer a question. In our customers' practice, a team owns a space or a subject, and one named person in that team is accountable for each article. When that person leaves the role, the team names the next one. KCS gives a similar job to Knowledge Domain Experts, who keep a whole collection healthy.
When an owner leaves
The introduction of ISO 30401, the knowledge management standard, names staff turnover and organisational change as causes of knowledge loss, and its clause 4.4.2 asks for activities that retain current knowledge, such as succession plans. In a study of an Australian government department from 2008 to 2013, published in 2018, losing knowledge hit daily operations hardest: lower productivity, capability gaps, slower work. APQC says a knowledge transfer plan should state who will own the resulting content and when it is reviewed or retired.
No source we read explains how to find orphaned articles at scale, or how to reassign them when two teams merge or someone leaves without a handover. A routine that covers both cases:
- Before the last day, list every article the person owns.
- Name a successor for each space or subject, rather than one person for everything.
- Ask each successor to read the articles before taking them over, and to archive what nobody needs.
- Ask the person leaving which steps they do from memory, and write those down. Our guide to knowledge transfer covers the methods.
- Each month, look for articles whose owner's account is closed. That is how you find the orphans nobody planned for.
How often to review: what the sources say, and what they do not
No trusted source gives evidence for any review interval. A rule of thumb repeated widely online says to review every article at least once a year, and every three to six months where content changes fast. We found no standard, public body or study behind it.
| Source | What it sets | Interval |
|---|---|---|
| KCS | Review happens when an article is used: "reuse is review". Articles nobody uses are not reviewed. | none |
| NHS website | Every editorial page reviewed, and new evidence and user feedback handled as they arrive | three years at most |
| US Environmental Protection Agency (EPA), guidance on procedures (2007) | Update whenever the procedure changes, and review periodically | one to two years, given as an example |
| GOV.UK | An optional review date that sends an email reminder, and a list of pages overdue | none |
| UK Department for Education | A clear process to check that content is current or should be retired | none |
| ISO 30401 and ISO 9001:2026 | Outdated knowledge handled, and knowledge kept valid | none |
Even a rule that exists has to be kept. The NHS content policy, last reviewed in October 2022, gave October 2025 as its own next review date.
Three triggers: use, change and date
On use. KCS asks whoever uses an article to check it while they work. This is cheap, but it misses articles that are rarely used and costly when wrong, such as a disaster recovery plan.
On change. The EPA asks for a procedure to be updated whenever the work changes. When a product, a rule or a system changes, every article that describes it is due at once, so tag each article with what it describes.
On date. A review date is the backstop for what use and change miss: the articles nobody opens until the day they matter. The NHS sets a maximum, and GOV.UK sends a reminder when the date set by the author arrives. Most teams need all three triggers.
Set the interval by risk
ISO 30401 asks organisations to identify and prioritise the knowledge of most value (clause 4.3), and its 2022 amendment defines critical knowledge as knowledge with a very significant effect on performance. That gives a basis for intervals: the more harm a wrong answer does, and the faster the subject changes, the shorter the interval. The table below is one way to decide, not a norm. Write down your choice and the reason, so the next owner can change it.
| Content | Main trigger | Review date |
|---|---|---|
| Safety, security and compliance procedures | change, and after every incident | short |
| Product and system how-tos | change: each release | the release cycle |
| Support fixes | use | none while people reuse them; archive when they stop |
| Policies and rules | change: a new rule | the policy's own review date |
| Rarely used but critical: recovery, emergencies | date, and a test run | short, because use will not trigger a review |
| Stable reference: contacts, terms, glossaries | use and feedback | long |
Content governance: approval, validity and archiving
What approval is for
An approval step is a check before content reaches readers. The NHS website needs two approvers from its clinical team for new content and major changes. The EPA asks for review by people with the right experience, then approval by a supervisor and the quality officer. EU rules for medicines manufacturing ask for instructions to be approved, signed and dated by authorised people, with a date from which they apply. For documents inside a quality management system, ISO 9001:2026 asks for review and approval when a document is created or updated, and for each one to be identified, for example by title, date, author or reference (clause 7.5.2).
How review is done matters. In a 2009 study of knowledge repositories, review that was transparent and aimed at helping authors improve encouraged people to contribute. The authors note that strict validation is often recommended without anyone testing its effect on contributors. KCS approves people rather than each article: staff earn the right to publish by passing an exam, and coaches check samples of their work afterwards. A sensible rule is approval before publication where a wrong answer causes harm (safety, legal, money, promises to customers), and review after publication elsewhere.
Validity dates
Some content is right until a known date: a temporary procedure, a price list, a rule that changes on 1 January. Give it an end date when you publish it, so it does not depend on someone remembering. GOV.UK lists closed schemes and news over a year old among the typical pages to retire.
Archive, do not delete
GOV.UK has two ways to retire a page. A withdrawn page stays at its address with a banner, leaves internal search and can be restored. An unpublished page is removed and can redirect to current content. GOV.UK also says that guidance pages should usually be updated rather than withdrawn. KCS archives rather than deletes, and warns against shrinking a knowledge base just to make search results more relevant.
The standards point the same way. ISO 30401 requires activities that handle outdated or invalid knowledge, for example by updating, archiving or deleting it (clause 4.4.2). ISO 9001:2026, published on 16 September 2026, describes organisational knowledge as dynamic: kept valid and relevant, and updated through use and experience (Annex A.7.1.6). Its guidance on documented information adds that obsolete information kept on file should be controlled so that nobody uses it by mistake (Annex A.7.5). An archive that removes an article from search but keeps its history fits both: readers stop finding the old version, and the record of what was true, and when, is still there.
What every article should show its reader
Four signals let a reader judge an article before acting on it, and each one is cheap to show.
- A review date, not only an edit date. Pages on the NHS website show when they were last reviewed and when the next review is due. A "last updated" stamp moves with every typo fix.
- An owner: a person or a team the reader can ask. Only eight of the 56 bases showed one.
- The type of content and its audience: a procedure, a policy, a FAQ, "for managers". Six of the 56 showed one.
- A way to react on the article itself, ideally one that lets readers say what is wrong. GOV.UK puts "Report a problem with this page" on its pages. Of the 56 bases, 19 had a feedback path on the article. Most of the others offered only a general contact link.
How to measure knowledge base accuracy
No trusted source defines a measure of how much of a knowledge base is correct. KCS and HDI (a support practitioner body) warn instead that counting activity harms quality, and APQC asks for evidence of business results rather than activity reports. The KCS guide is direct: "goals placed on activities will corrupt the knowledge base". A target of 10 articles a month produces 10 articles a month, whatever their value. In 2018, HDI added that counting articles fills the base with low-value content.
Errors do not fix themselves. In a peer-reviewed study of a large public question-and-answer site for programmers, published in 2019, only 20.5% of the answers observed to be obsolete were ever updated. That site is not a company knowledge base, so read the figure as a warning rather than a benchmark.
What a team can track honestly, knowing that most of these do not measure accuracy itself:
| Signal | What it shows |
|---|---|
| Share of articles with an active owner and a review date that has not passed | whether the routine is running |
| Share of reads on articles past their review date | how much reading relies on unchecked content |
| A monthly random sample, each article checked against its source of truth | whether articles are right |
| Flags raised by readers, and how long each waits | whether errors reach an owner |
| Searches that return nothing | what is missing |
Only the sample checks whether articles are right; the others show where to look. KCS coaches score samples against a six-question checklist: is the article unique, complete and clear, is its title good, are its links valid, is its metadata correct. Add one question of your own: is it still true?
When an AI assistant answers from the knowledge base
An assistant that answers from your knowledge base repeats what the articles say. A stale article used to mislead the few people who opened it; now it can feed every answer built on it, and the person reading the answer may never see its date or owner. APQC's 2026 priorities survey found that AI "exposes weaknesses in content quality, governance, taxonomy, and ownership". The method on this page stays the same, but skipping it costs more. Our guide to AI knowledge management covers what changes, and WYKIWYG: what you know is what you get explains the research on why an assistant is only as right as its documents.
Where the sources stop: how this works in Elium
The sources say what to do (name owners, review and retire content) but not how to keep it running across thousands of articles while people join and leave. Elium is a European knowledge management platform built around these steps. Each article has one owner, whose name appears at the top of the article. A verification date, from one month to one year ahead or on a day you choose, puts the article in its owner's task list as the date approaches, and a badge tells readers whether it is verified or expired. Administrators see which articles have expired, who owns them, and the share of reads on expired articles.
When someone leaves, an administrator transfers their articles space by space, and the successor can review each article before taking it over, or archive it. A space can make approval mandatory for new articles and updates. Any contributor who spots an error in someone else's article can correct it and send the change to the owner for approval: flag it or fix it, with the owner deciding. Archived articles leave search and can be restored.
The interval stays your decision. At the Département de l'Orne, where 25 IT professionals support 1,800 staff, every procedure has an expiry date and an annual review, so none stays active without being verified.
Where to start this month
- List every article with its owner and its last review date, and count those with neither.
- Take the 20 most read articles and the 10 that would do most harm if wrong. Give each an owner and a review date.
- Put a way to flag errors on every article, and decide who receives the flags.
- Write down your intervals and why, and your routine for leavers.
Common questions
Knowledge base maintenance is the routine that keeps articles correct: an owner for each article, reviews triggered by use, change and date, approval where errors cause harm, archiving, and a handover when people leave.
The article's owner reviews it, at an interval your team chooses by risk, because no trusted source gives evidence for any interval. The NHS website reviews health content at least every three years, the US EPA gives one to two years as an example for procedures, and KCS reviews articles when they are used.
Tie reviews to the change, not the calendar: tag each article with the product or feature it describes, review those articles with each release, and ask every reader to fix or flag errors.
One named person per article, accountable for its accuracy, usually in the team that does the work it describes. Anyone who uses the article can still flag or fix it, which avoids the duplicates KCS links to individual ownership.
They pass to a named successor for each space or subject before the owner's account closes. The successor reads each article and keeps or archives it.
Check a random sample of articles each month against their source of truth: no trusted source defines a better measure. Alongside it, track the share of articles with an owner and a valid review date, the share of reads on overdue articles, and searches that return nothing.
Content governance is the set of rules on who owns, approves, reviews and retires each article, and who may read it. In a knowledge base, that means owners, review dates, approval where errors cause harm, and archiving rather than deleting.
Patient leaflets from an NHS hospital trust were the most complete pages in our audit of 56 public bases: each printed the department responsible, a production date, a review date and a version code.
Where to go next
Compare tools. Our knowledge management tools comparison and our guide to internal wiki software show what to check before you choose.
Use a template. The content review schedule, the knowledge audit checklist and the knowledge base style guide put this page into practice.
Do it in Elium. Our documentation explains verification dates, article ownership, article transfer and approval.
Read next. Knowledge as code, for teams that manage documentation the way engineers manage code.
Sources
- NHS, NHS website content policy, last reviewed October 2022. Checked 20 October 2026.
- Government Digital Service, Detailed guides, Manage existing GOV.UK content and Retire content. Checked 20 October 2026.
- UK Department for Education, Apply the Service Standard in DfE: iterate and improve frequently, last reviewed March 2026. Checked 20 October 2026.
- US Environmental Protection Agency, Guidance for preparing standard operating procedures, EPA QA/G-6, 2007. Checked 20 October 2026.
- European Commission, EudraLex volume 4, EU Good Manufacturing Practice, chapter 4: documentation, 2011. Checked 20 October 2026.
- Consortium for Service Innovation, Knowledge-Centered Success Practices Guide, 2026 (labelled 2027): techniques 2.1, 2.2, 5.2, 5.3, 5.4, 6.4 and 7.1; Principles and core concepts. Checked 20 October 2026.
- ISO/IEC JTC 1/SC 40, ISO/IEC 20000-1: implementing and auditing the knowledge requirement, Lynda Cooper, 2025. Checked 20 October 2026.
- ISO, ISO 30401:2018, knowledge management systems: requirements, with Amendment 1:2022 and Amendment 2:2024 (introduction, clauses 4.3 and 4.4.2, definition 3.34), and its free preview. Checked 20 October 2026.
- ISO, ISO 9001:2026, quality management systems: requirements, published 16 September 2026 (clause 7.5.2, Annex A.7.1.6 and A.7.5). Checked 20 October 2026.
- APQC, How can retiring employees' knowledge be retained without exploiting their personal information?, 2026, and 2026 KM priorities and trends, executive summary, 2026. Checked 20 October 2026.
- Rick Joslin, HDI, How to select the right knowledge management metrics, 2018. Checked 20 October 2026.
- Peter Massingham, Measuring the impact of knowledge loss: a longitudinal study, Journal of Knowledge Management, 2018. Checked 20 October 2026.
- Durcikova and Gray, How knowledge validation processes affect knowledge contribution, Journal of Management Information Systems, 2009. Checked 20 October 2026.
- Zhang, Wang, Chen, Zou and Hassan, An empirical study of obsolete answers on Stack Overflow, IEEE Transactions on Software Engineering, 2019. Checked 20 October 2026.
Data and sources
The audit covers 56 public knowledge bases and help centres, 30 in English and 26 in French, checked on 19 and 20 October 2026. The sample was drawn by a rule written in advance, with every later change recorded: for each of six sectors, web search results in the United Kingdom (English) and France (French) were read in the order shown, and the first eligible bases were kept, one per organisation. For each base we opened three pages (the home page, one top-level article and one deep article) and coded what a reader sees against written rules: a date, an owner, a type or audience label, a feedback path, version or validity information, and a search box. The evidence behind every code is kept. One person coded the sample, with no second coder. Four French sample groups, in three sectors, stayed short because the rule ran out of eligible bases, so the sample is 56, not the planned 60. Shares are given with a 95% interval; sector counts are too small for percentages. Organisations are described by sector, not named, except public bodies quoted as examples.
The rest of the page is built from public sources: standards bodies, public bodies, the owners of KCS, practitioner bodies and peer-reviewed research, each read and dated. Software vendors were left out. Statements about Elium link to our product documentation, and customer facts come from the published customer story. We build one of the products a knowledge base can run on, so the parts on practice are an informed opinion. If a source is misread, tell us and we will correct it.