Why Copilot and modern enterprise search reward structured, owned content (not a new search tool). And how an Answer-First information architecture turns your SharePoint into something that successfully answers questions.
TL;DR
- Search has shifted from "type keywords, scan ten blue links" to "ask a question, get an answer." Copilot and AI-powered enterprise search are built around that expectation.
- Those tools can only answer well if your content is structured, tagged, and owned. As Gartner notes, poor information quality "pollutes search results with redundant, obsolete, and trivial information". AI amplifies your mess, it doesn't fix it.
- When search fails, people don't file a ticket. They ping a colleague. The Atlassian/McKinsey research shows knowledge workers lose around 9 hours a week to information hunting, and over half believe the only way to get an answer is to ask someone.
- An Answer-First IA maps every page to the question it must answer, and assigns a human owner to that answer.
- Start with a page-type audit, not a new tool. The tool is the easy part.
Your people don’t want to search anymore. They want to ask.
Question-based search (the kind Copilot and modern enterprise search now lean on) expects your SharePoint to be a library, not a loft. Most intranets are still a loft. The good news is that fixing it is mostly a content-and-structure job, which is exactly the kind of work we do when we help organisations redesign SharePoint so people can find things. Let’s unpack what needs to change.
The shift: from keywords to questions
Keyword search was a compromise. You typed ‘leave policy parental 2024’, scanned results, opened three PDFs, and eventually asked Janet from People & Culture.
Question-based search changes the contract. Your people now type ‘how much parental leave am I entitled to after two years?‘ and expect a sentence back, with a citation. Think of it as the difference between a filing cabinet and a reference librarian. The cabinet holds documents. The librarian answers the question.
The catch? The librarian is only as good as the collection. If your content is untitled, undated, and owned by "IT (legacy)", no amount of AI will save you. This is the part the market is quietly learning: Gartner's Predicts 2026 warns that embedded AI "remains siloed within individual applications", so bolting AI onto messy content tends to produce confident-sounding answers built on the wrong source. (It’s also why getting your content Copilot-ready matters far more than the licence itself.)
The real cost: hours a week, quietly
The research here is remarkably consistent. McKinsey found employees spend an average of 1.8 hours a day (9.3 hours a week) searching for and gathering information. A 2025 Atlassian study put it even more bluntly: nearly half of workers’ time goes to “busy work” like chasing information and tracking down colleagues, and over half of employees believe the only way to get what they need is to ask someone or schedule a meeting.
That last finding is the one that should keep you up at night, because it describes a hidden cost. When search fails, people stop searching. They ping a colleague, interrupt a manager, or quietly re-create a document that already exists. Your official knowledge base becomes a graveyard while a parallel, undocumented one lives in DMs.
A few warning signs worth watching for on your own intranet:
- A spike in “can someone send me the latest…” as a recurring Teams message.
- Duplicate documents with near-identical names and different dates (the “final_final_v3” problem). Usually a symptom of weak document management.
- The same question landing in #general three times a month.
- Searches for common terms returning zero useful clicks.
You don’t need a fancy analytics suite to spot these. You need to watch what people do, which, tellingly, is roughly what every honest practitioner in this space recommends.
Answer-First IA: design around the question, not the folder
Traditional information architecture asks ‘where does this document live?‘ Answer-First IA asks ‘what question does this page answer, and who owns that answer?‘ Same content. Different starting point.
You build it by listing the top 100 questions your people ask - pulled from Teams DMs, helpdesk tickets, and the ‘most searched’ report. Match each question to a page type: policy, how-to, FAQ, contact card, form. Each page type gets a template, required metadata, and a named owner.
Spoiler: your top 100 questions are usually covered by maybe 12 page types. That’s the whole game.
Metadata is the muscle; owners are the skeleton
Metadata does the heavy lifting for question-based search. Audience, topic, date of last review, jurisdiction, document type. These are what let search confidently say ‘here’s the answer for a Sydney-based casual worker, updated last month.‘ ” It’s also what powers a well-tuned SharePoint search experience, where filters and synonyms quietly do half the work.
But metadata without owners decays within a quarter.
Someone leaves. A policy updates. The tag stays. The page lies. This isn’t hypothetical. Practitioners consistently find that the root cause of “search is broken” isn’t the software, it’s the absence of organisational standards and ownership. Content gets added by whoever needs it, organised around the org chart rather than how people think, and then nobody is responsible for keeping it current.
So make ownership structural:
- Assign an owner at the page-type level, not page-by-page (otherwise it’s unmanageable).
- Build review cadence into the content type, not someone’s memory. A calendar ping every 6 or 12 months.
- Retire aggressively. An expired policy is worse than a missing one. No owner, no publish.
Don’t skip the content quality step
Here’s the part people love to dodge: your content must be readable. Question-based search extracts answers from pages. If your parental leave policy is a 14-page PDF with a scanned signature on page 11, nothing is extracting a clean answer from that. And unsearchable scanned PDFs are one of the most common failure modes in enterprise search.
Rewrite your top-traffic pages as plain-language Q&A where it makes sense. Short paragraphs. Clear headings. The policy PDF can still live somewhere as the source of truth, but the answerable page sits on top of it. Think of it as the menu versus the recipe book. Both have their place; only one belongs in the customer’s hands.
Mini-template: Answer-First IA page-type audit
Run this across your top 20 intranet pages or page types. You’ll spot the gaps within an afternoon.
| Page type | Question it answers | Owner | Metadata tags needed |
| Policy page | What are the rules for X? | Policy owner (named) | Audience, jurisdiction, effective date, review date |
| How-to / procedure | How do I do X? | Process owner in the business | Task, role, system, last verified |
| FAQ | Quick answer to a common question | Comms or subject lead | Topic, audience, source page |
| Contact card | Who do I ask about X? | Team lead | Team, capability, location |
| Form / request | How do I request X? | Service owner | Request type, SLA, system |
| News / update | What changed and when? | Comms | Audience, publish date, expires |
| Reference / data | What is the current number for X? | Data owner | Source, refreshed date, confidence |
The mistake most organisations make
They buy the search tool before fixing the content.
Modern search and Copilot are genuinely wonderful, but they don’t invent answers. They surface whatever you’ve fed them. Messy content in, confident-sounding nonsense out.
The order matters. Clean a small, high-traffic slice first. Prove the uplift. Then scale. It’s less glamorous than a platform launch, and it works. (If you’d rather not rebuild your information architecture from a blank page, a structured intranet managed platform like Injio bakes consistent page types, governance and ownership in from day one.)
Where to start this week
You don’t need a budget or a project sponsor to begin. Pick your ten most-searched terms from this quarter and walk through them as a brand-new starter would. If the answer isn’t obvious within one click, you’ve found your starting point.
Then ask the three questions that reliably separate an intranet that answers from one that just stores:
- Can your people get a one-sentence answer to “how much parental leave am I entitled to?” without opening a PDF?
- Does every page type on your intranet have a named human owner and a review date?
- Have you looked at what staff type into search this month, and compared it to what they get back?
If any of those gave you pause, the good news is that the fix is structural and repeatable.
Get the content layer ready to answer, and whatever search tool you layer on top (plain SharePoint search today, Copilot tomorrow) suddenly starts earning its keep.
That sequencing (foundations first, tooling second) is exactly the work we spend our days on at WebVine, and it’s the reason “good structure is portable” is less a slogan and more a survival strategy.
FAQs
Do we need Copilot for this to matter?
No. An Answer-First IA improves plain SharePoint search, Copilot, and any future search tool you bolt on. Good structure is portable. It pays off regardless of which AI layer you eventually adopt.
How long does an IA redesign take?
A first pass on page types and top-100 questions takes about four to six weeks. A full rollout with owners and governance is a few months. Start with the top 20 pages and prove the uplift before scaling.
What if our content is scattered across Confluence, shared drives, and SharePoint?
That’s normal. The IA work tells you what to consolidate, what to leave in place, and what to index so search can reach it. Migration isn’t always the answer, findability often is.
Who should own this? IT, Comms, or HR?
All three, plus someone senior enough to unblock. A cross-functional group with a clear decision-maker moves faster than handing it to one team.
How do we measure success?
Search abandonment rate, top-searched terms returning zero clicks, and a drop in ‘does anyone have…‘ messages in Teams. Baseline now, re-measure in 90 days.
Why does AI search sometimes make findability worse?
Because AI amplifies whatever it retrieves. Gartner notes that poor information quality “pollutes search results with redundant, obsolete, and trivial information.” Without a governed, well-structured knowledge layer underneath, more AI simply means more confident noise.
Sources
- McKinsey Global Institute — time spent searching for information (1.8 hrs/day, 9.3 hrs/week), summarised via Cottrill Research.
- Atlassian workplace study (2025), via Workplace Wellbeing.
About the Author – Rachel Harnott
Rachel Harnott is the Head of Modern Work at WebVine, where she helps enterprise teams turn sprawling SharePoint estates into intranets that actually answer questions. She works at the intersection of content design, metadata governance, and AI-ready search. Pairing practical, in-the-business experience with a clear-eyed view of where Copilot and enterprise search are genuinely heading. Rachel is a firm believer that findability is a content problem dressed up as a technology problem, and she’s spent years proving it one page type at a time.