Independent audit · October 2026 · Solo
For research and learning only. This is an independent study I did on my own to understand how a large help center is organized. I don't work for LinkedIn, I'm not affiliated with or endorsed by LinkedIn, and I had no access to its internal data. All findings come from the public Help center as I saw it, signed in to my own account, in October 2026. LinkedIn and related names belong to their owners.
LinkedIn's Help center has the answers people need, but people can't find them easily. The page mixes different kinds of links, uses different names for the same topics, and hides the way to talk to a person. I checked the page and every page it links to, found 16 problems, and planned a simpler way to organize the help center.
I had two reactions as soon as I opened the page.
It is called "LinkedIn shortcuts." On a help page, I expect a link to explain something first. Half of these links skip that and take me straight to my settings to change something.
Help, settings, rules, and contact options are all mixed on one page. It is hard to tell where to go.
When I clicked through every link, both reactions turned out to be right. "Cancel LinkedIn Premium subscription" opens a helpful article. But "Close your account," a much bigger step, opens a settings page with no explanation. The first time I clicked it, I got an error: "Hmm, this page doesn't exist."
These problems are not just about wording. They come from how the help center is built: what kinds of pages it has, how topics are named, and who looks after the content.
The biggest problems: people can't easily reach a person, can't find the Premium and product help pages, and see messy text in search results. I found 16 problems in five groups.
Contact us has "Chat with support," marked "Online now." The home page only links to it from the footer. Past support cases are hidden in the profile menu.
The product dropdown only changes what search looks in. The page itself doesn't change. Billing and Learning help homes exist, but only the profile menu leads to them.
There are 60 Premium articles, but no tile, topic, or dropdown option leads to them.
It says "Other ways we can help," but it holds policy pages and a status page. None of them help you contact anyone.
Shortcuts, Recommended topics, and Suggested articles all show "popular" content. "Cancel Premium" shows up twice.
Three open your settings in a new tab, with no sign that they will. Three open help articles.
"Close your account" opens settings, not an article, and never says what happens to your data or connections.
"View all" opens "LinkedIn Consumer Solutions Topics," which even asks "Was this answer helpful?"
One article ends with "Upgrade to Premium." The Premium cancel article says the same two rules twice.
The home page says "Connections"; the full list says "Build Your Network." "Basics" is on the home page but not in the full list. Neither includes Premium.
One article is tagged both "Job Search" and "Search and Apply for Jobs." Another has no tag at all.
"Basics" puts member help next to Recruiter product details. The Premium tag includes Sales Navigator webinars.
Snippets include code words like "ORDERED" and "FEATURE_LAUNCHER," and even a raw web address.
For "change password," the top result is a reset flow for signed-out people. The "Reset your password" shortcut opens that same flow, even when you're signed in.
Billing search shows almost the same results as LinkedIn search. A privacy article suggests "Deceased LinkedIn member" as related reading.
Capital letters are used differently on the same page. One shortcut article was last updated a year ago. A fake-invoice warning shows only in Billing search.
Premium is LinkedIn's paid plan for members. The help center already has 60 articles about it, but there is no Premium section to find them. The only way in is a small tag link at the bottom of an article. That tag page is one long list with no order:
So a member who wants to compare plans, learn a feature, or cancel has to scroll through 60 articles or guess the right search words.
I was signed in with a Premium plan, so the "Cancel Premium" shortcut may show only for Premium members. I couldn't check what a free member sees. Either way, a free member thinking about upgrading needs this information too.
In the new plan, each section has one clear job, in the order people need it: search, browse, then contact someone.
A new layout only lasts if the help center underneath follows clear rules. The first rule: every page is one type, and each type has one job. Then a link's label can say exactly what it opens.
Same layout for every product: search, quick actions, topics, popular articles, and contact at the top.
LinkedIn, Billing, Learning, and a new Premium homeA real page, not an article. Its name comes from one shared list.
"View all" opens "All help topics"Steps first, a "Go to settings" button at the top, no ads.
Change your passwordWhat will happen and what the limits are, then a link to the how-to.
What happens when you close your account"What happened" explains the cause; "What to do" gives the steps.
Bounced or unconfirmed email addressTables, no steps.
Tax rates by countryOnly in quick actions and how-to articles, always labeled "(opens settings)."
Change your email addressPremium gets its own help home, listed in the product chooser and the topic list. The 60 articles are sorted by what a member wants to do:
Sales Navigator and Recruiter articles move to their own help homes.
One topic list for everything. The home page tiles, the full topic list, article tags, and every product help home use the same list, so each topic has one name everywhere.
Labels are how people move through a help center. Once each page type has one job, each label can say exactly what it opens.
Says these links do something, not explain something.
Matches what a signed-in member wants.
Explains what happens before you decide.
The chat already exists; people just can't see it.
Fits once the box holds chat, your cases, and status.
Drops company jargon.
Uses a written summary, not raw text.
I didn't have LinkedIn's data, so these are what I would track after launch, not results.
Before launch, I would test the new topic list with a card sort, where people group topics the way they think of them, and a first-click test, where people get five common tasks and I check whether their first click is right.
I start with what a user expects, then follow each problem down to where it really starts. "LinkedIn shortcuts" looked like a wording problem, but the real fix was deciding what types of pages the help center has.
Simpplr · Knowledge management lead · Live at help.simpplr.com
Simpplr's Help center held 563 articles, and its structure no longer matched the product or the way people searched. I audited the content against the engineering structure, studied 19,743 searches, and tested a new structure before I rolled it out. The Help center now runs on six hubs. In an A/B test on one section, task success rose from 61.4% to 79.8%.
Simpplr is a platform with eight customer-facing product areas that run on shared services for sign-in, user provisioning, access control, analytics, and governance. Engineering runs 46 active scrum teams across 16 functional areas and owns 241 documented product features. The Help center had to serve all of it.
It did not. The structure came from categories inherited from earlier help desk tools, and those categories no longer matched how customers think or how engineering ships. Customers struggled to find content, authors wrote duplicate articles, article depth varied, and AI search could not ground its answers reliably.
I compared the published articles with the engineering team and feature structure. Both sources showed the same thing: the platform had outgrown the Help center.
| Source | Count |
|---|---|
| Articles in the Help center | 563 |
| Release notes (monthly and hotfix) | 125 |
| Attribute-based access control (ABAC) articles | 42 |
| App management articles | 31 |
| Provisioning, SCIM, and sync articles | 30 |
| Analytics and reports articles | 20 |
| Engineering functional areas | 16 |
| Active scrum teams | 46 |
| Documented product features | 241 |
Release notes, access control, app management, provisioning, and analytics held the largest groups of articles. The last four are platform capabilities, not single products. The old structure had no consistent home for them, so they showed up in several places or in the wrong one.
The repeated searches mattered most. They showed that people often missed what they needed on the first try. The search data and the audit pointed to the same cause from two directions: the content existed, but the structure did not connect people to it.
Tree testing showed whether people could navigate the structure. It did not set the final structure on its own. The results backed four decisions: cross-product capabilities get one canonical home, product areas share one navigation pattern, top-level categories stay stable, and lower-level branches can flex as features change.
| Measure | Existing structure | New structure |
|---|---|---|
| Users or sessions | 260 | 320 |
| Task success rate | 61.4% | 79.8% |
Task success rose 18.4 percentage points, a 30% relative gain. The test ran on one section, so the number describes that section. I applied the validated pattern to the rest of the Help center and report every other result against a measured baseline.
Each hub answers one reason a customer opens the Help center.
A sequenced path with first steps at 30, 60, and 90 days, plus training, best practices, and a glossary. Customer Success can send one link.
Eight product areas. Each one follows the same pattern: overview, capabilities, workflows, and FAQs.
Platform capabilities that cut across products, such as access control, user provisioning, and app management. Each is written once and linked from every product that uses it. This hub held more than 130 articles at audit.
Grouped by monthly release. Each release holds its notes and its hotfixes together, and each hotfix name carries the release number.
APIs, webhooks, the workflow builder, the integrations library, and HRIS and iPaaS integrations, in one entry point.
How to submit a ticket, support levels and SLAs, status and incidents, and troubleshooting guides by category.
Release notes
├── Release process and schedule
├── Latest release
├── Early product update notification
├── Deprecations and breaking changes
└── 2026 releases
├── 26.06
│ ├── Release notes 26.06
│ └── Hotfixes for 26.06
└── 26.05
├── Release notes 26.05
└── Hotfixes for 26.05 (example: Hotfix 26.05.A.1.1)A reader who checks one release sees everything that shipped for it in one place.
Every engineering functional area now has a predictable customer-facing destination. Article ownership is clear, and a new feature has a known home.
| Engineering functional area | Customer-facing destination |
|---|---|
| Platform services and admin interface | Admin and setup |
| Enterprise search | Products > Enterprise search |
| Site and content | Products > Core intranet |
| AI solutions | Products > AI platform |
| Mobile | Products > Mobile and Frontline |
| Employee newsletter | Products > Employee newsletter |
| Employee listening | Products > Employee listening |
| Recognition | Products > Recognition and rewards |
| Data engineering | Admin and setup > Analytics |
| Migration tooling | Getting started > Migration |
| DevOps, architecture, and design system | No customer-facing destination |
A structure only lasts with rules and owners. These are the parts I set up and run.
Voice and tone were already close to AI-ready: conservative and consistent. The real problems were structural. Several articles were stale enough to retire, and others had steps out of task order. One profile-creation article, for example, skipped the sign-up step that has to come first.
DevRev suggested tagging articles with the same Product area field that classifies support issues. I kept the two apart. My team designed the Product area field to sort issues down to what end users face, and to train SIMA and Computer, the two AI agents that resolve employee-facing support requests. Articles are a product artifact that ships alongside the feature, so they follow the Help center structure. Support issues are a record that exists after launch. Merging the two taxonomies would weaken both.
I set no outcome targets that I have not measured. Every later phase reports against a measured baseline, hub by hub: documentation-related ticket volume, search-to-click rate, median time to answer per topic, AI thumbs-down rate per topic, duplicate articles per topic, and average article age per hub.
The six-hub structure is live on help.simpplr.com, and all 563 articles sit inside it.
Simpplr · Support bot · Knowledge for AI
SIMA, the Support Intelligence Mentor and Assistant, is Simpplr's DevRev support bot. A bot answers only from the knowledge behind it, so I fixed the knowledge first. Within one month of publishing the new articles and setting guardrails, deflection of question-type tickets rose from 31% to 67%.
Question-type tickets made up the largest share of support volume at Simpplr, and they landed in P0 priority. Most were small, everyday questions, yet each one still cost support time and money.
Simpplr ships fast, so the knowledge base often had gaps or outdated articles. Customers searched the portal first, and a missing or stale answer became a ticket. The same issue often arrived as four to six tickets, worded differently by different reporters. Each one still needed a support engineer to read it, spot the duplicate, and answer it again.
Each ticket then moved through support, expert or product input, and the knowledge manager. Customers waiting through that chain often saw their SLA breach. Implementation teams sometimes hit the same wait during demos. A quick knowledge base update could not fix this, because product input was a dependency and documentation bandwidth was limited.
The goal was to deflect P0 question tickets with SIMA, without waiting for a full knowledge base rewrite. DevRev provides a bot with Simpplr's license, but a bot needs training before it gives the result you want. I worked with support, CX, and support operations to learn what the bot could do, and with DevRev to confirm what the license covered. We named the bot SIMA and started training it.
The plan rested on one idea: SIMA can only answer as well as the content behind it. So I fixed the content for the highest-volume ticket group first, then limited SIMA to answer from that content.
Each pass narrowed the work until the remaining questions were small enough to review in full.
| Stage | Count | What I did |
|---|---|---|
| Export | 2,330 tickets | Pulled all question-type tickets from the last 12 months, in every status, with DevRev's AI tool, Computer. |
| Group | 1,198 questions | Triaged the tickets by feature and merged similar or identical tickets into single questions. |
| Filter | 573 questions | Mapped each question to its docs. Removed questions already covered by updated articles, and questions tied to features or flows that no longer exist. |
| Structure | 87 sections | Organized the questions by feature, then by flow within each feature. |
| Consolidate | 33 articles | Covered all 87 sections in 33 articles, with answers taken from Rovo, where questions and answers carried recent update dates. |

Sites, newsletter, and forms accounted for 452 of the 573 remaining questions.
Adding 573 questions to the docs in one pass would have created too much work, so I grouped them again. Inside each feature I sorted the questions by flow, which gave 87 sections. For the answers, I built an Atlassian Rovo agent that pulls from Jira, Confluence, Slack, and GitHub, so answers would not go stale the way a static article does. Every question and answer carried a recent update date.
The 87 sections became 33 articles, and I published them.
After I published the 33 articles, I worked with DevRev to set guardrails on SIMA. The guardrails make SIMA answer this type of question in a fixed pattern, and only from those 33 articles. Its answers stay tied to content that the documentation team reviewed and published.
A separate DevRev agent maps each incoming ticket to the customer's organization and subscription plan. It resolves eligible P0 tickets on the ticket itself, before they reach a support engineer or customer success manager. The content and the routing logic have to match exactly, or an eligible ticket sits unresolved.
I also built a resolution-first support agent with a 3-step resolution flow inside the conversation, for the highest-volume question type.
Deflection of question-type tickets rose from 31% to 67% within one month of publishing the 33 articles and setting the guardrails. That is a 36 percentage point increase, so SIMA deflected about two in three of these tickets. A deflected ticket skips the chain that caused the original delays: support effort, expert or product input, and routing to the knowledge manager, where SLA breaches began.
SIMA is in use for ticket deflection on this ticket group today, with the FAQ content and guardrails live and the auto-resolve path running in production.
The in-app help panel gives SIMA the same content in context. Read how the panel works.
Simpplr · In-product help · Shipped
A question mark and a gift box in the top bar open two in-product flows: a Help center and a release digest. I designed and prototyped both end to end for Simpplr's intranet. The Help center's real job sits upstream of the panel. It shows the structure and content that SIMA, Simpplr's support bot, is trained and matched against.
Sending someone out of the product to a separate help site breaks their flow, especially when the question is about the screen they are already looking at. I used the Learn panel in Adobe Premiere Pro as a reference point and designed the panel to match help content to the screen a person is on. The fix sits next to the problem instead of behind a search in another tab.
The panel's own structure was never the goal. The goal was to close a gap between the product and its knowledge, so people get an answer without leaving the product or opening a ticket. I wrote the PRD on the DevRev PLuG SDK and the Docebo catalog API, then owned the build end to end, including the matching logic that decides which help content shows on which screen. I aligned the direction with product, design, and CX leadership before it shipped.
Both experiences sit in the global top bar, one click from any page. The question mark opens the Help center. The gift box opens What's new, with a dot when there are updates the person has not seen.

The panel opens at the side, not as a modal, so the person keeps the page they were working on in view. The order is deliberate: search first, then guided learning, then browsing, and support only after the first two have had a chance to answer the question.

A query returns a layered result instead of a flat list of articles. Selecting the "Forms" popular search fills the box and surfaces a "Navigate to feature" card first: a breadcrumb path, four numbered steps, and a "Take me there" button that jumps the person straight to the feature. Below it, an AI answer summarizes the topic and names its sources, with a matching articles list further down for anyone who wants the full procedure.

When a query returns nothing, the panel says so plainly, suggests a different keyword, and puts "Submit a ticket" directly in the empty state. No search in this panel ends at a dead stop.
Selecting a tutorial opens a player with its topic, its position in the series, and a one-line description. An "Up next" row keeps the rest of the series in view, and the series counter updates as the person moves through it, so they always know how much is left.

Submitting a ticket opens a short form inside the same panel, not a redirect to DevRev, with an intro line that still offers DevRev for complex issues. Category defaults to Question and Priority to Medium. The description placeholder asks for what happened, what the person expected, and the steps to reproduce it: the exact shape of information support needs to act on a ticket the first time they read it.

A confirmation screen closes the loop. It thanks the person, says support will follow up by email, and explains that the ticket can be tracked in DevRev, with "Back to help" and "Open in DevRev" as the two next actions. The other support path opens SIMA as a chat inside the same panel. SIMA states its job in one line and asks the person to describe the issue so it can generate the ticket itself.

What's new opens as a centered modal built like a short digest instead of a changelog dump. A left rail lists every update with its status, and the active item is marked, so the person can see how many updates there are and where they are in the list. The right side shows one update at a time: an image, a product area label, a title, and a plain-language summary. Updates that apply to one role carry a "Who's this for" line, so people can skip what does not concern them.

A Next button steps through updates in order, or the person selects any title in the rail to jump to it. The close button in the top-right returns them to the page they were on, and the modal reopens on the last update they viewed instead of resetting to the first.
| Decision | What it does |
|---|---|
| Self-service before support | The order is search, tutorials, browsing, then support. Support stays reachable from every view through the footer link. |
| Answer and action together | Search results lead with a "Navigate to feature" card and a "Take me there" button, so a person can act on the answer instead of leaving help to find the feature. |
| No dead ends | The empty state, the footer, and the "Still need help?" card all offer a way to reach support. |
| Better tickets by design | The description placeholder asks for what happened, what the person expected, and the steps to reproduce. |
| Honest AI copy | The AI answer shows its sources, and SIMA says upfront that it can make mistakes. |
| Release notes people read | Each update has a product area label, a short title, a plain summary, and an audience line. |
These results come from my in-app help and in-product text work at Simpplr.
Adobe · HelpNext program · Feb 2022 to Apr 2025 · Led a team of 4
HelpNext set out to improve how Adobe customers find and use help across a large product portfolio. I led content strategy for Creative Cloud and enterprise products, including the ProVideo help experience. I rebuilt the help around tasks, with a clear taxonomy, metadata, and single-intent articles, so each article answers one question.
A long article that answers five questions ranks for none of them, and it is hard to reuse. Splitting content by intent gives search a clear match, lets teams reuse one answer in many places, and gives AI a clean section to quote. This work shaped how I design content for AI today.
The older HelpX experience had nested topics, long scrolling pages, and navigation ordered by popularity. The redesign added contextual navigation, simplified labels and breadcrumbs, and content organized by topic.


HelpNext crossed many teams. I worked with UX design, UX research, product management, engineering, customer support, content marketing, localization, and leadership.

Adobe's error copy tended to say what went wrong without offering a path forward. I rewrote it to name the shared responsibility and point to what happens next.

Adobe wanted a trial button that got people into Premiere Pro without the friction of a purchase decision. Before I wrote a word, I worked from a persona and a mapped set of the fears that stop someone from starting a trial, so the copy could answer the real hesitation instead of guessing at it.


I placed the button and its supporting copy inside real help articles, such as one on trimming clips, so the trial offer appeared where someone was already learning the tool.

I also mapped the three paths a click could take: straight into the app, through identity verification and a plan screen, or into a tutorial. That kept the content and the engineering build in sync on what each path needed to say.

Task-based help for new and underused features raised feature adoption by 27%. Help written to line up with a launch led 19% more users to try the color grading feature in the first 45 days.
Single-intent, reusable content, applied across the articles I owned, contributed to a 45% increase in organic traffic and a 47% improvement in search rankings for those pieces. A/B tests of the HelpNext program showed $324K more retained annual revenue. Authorship and edit history are verifiable on the public author pages.
Simpplr · AI-native product · Built from scratch
A new AI-native product needed a full knowledge portal from day one. I planned the structure, wrote the content, and set the rules for who owns what. The portal has 230+ pages for six products today, and it is built to take more.
A new product has no history to lean on. Readers needed to find setup steps, feature guides, API details, and release notes on day one, and the portal had to stay tidy as products joined it. I planned the structure for six products at once so that one pattern could absorb the next product without a redesign.
Product documentation and marketing content serve different readers, so I kept them apart and gave each a clear owner. The portal reads as a product artifact and uses product voice and tone. I also wrote separate, enforceable guidelines for product and help visuals, so marketing assets do not blend into technical documentation.
The standards behind these rules are covered in Technical writing standards.
The portal saves about $34K a year in platform costs. Targeted knowledge resources and customer education also supported 1 to 2 at-risk enterprise renewals per quarter, at roughly $232K average ARR per customer. That figure is an approximate average, not the value of any one customer.
Simpplr · Knowledge operations · Team of 6
A knowledge base stays useful only if requests, priorities, and reviews follow a clear path. I set up that path, and I lead and mentor the 6 people who run it across content, support, and enablement.
Requests used to arrive as direct messages from support engineers, customer success managers, and implementation engineers, and the order of work followed urgency or whoever asked. That worked at low volume. As Simpplr shipped faster and the Help center became a heavy investment, it caused problems with tracking, prioritizing, and auditing.
Every request is now tracked, so I can audit throughput and patterns and keep the quality of the inventory in check.
The Jira template defines six request types, each with its own SLA, bare-minimum fields, and a checklist:
An SLA starts only when four inputs are in place: a linked ticket, a scope boundary, a reproducible environment, and a named subject matter expert. This protects writers from vague requests and protects requesters from unclear timelines.
Every AI-assisted draft goes through the same gate. Automated diff testing runs first, Claude flags variances, and I test the result by hand against real product behavior before anything is published.
Simpplr · Help center, release notes, and self-serve documentation
Help content needs its own rules. I wrote a full set of standards for technical writing at Simpplr: knowledge base articles, help center pages, release notes, and other self-serve content. They are a separate body of work from my UI and UX copy standards.
Vernacular and Accordis are my content design work, for UI and UX copy such as buttons, errors, and notifications. This page covers the technical writing standards for help content. The two sets are separate documents with separate rules.
13 parts on voice, case, punctuation, terminology, structure, and accessibility, for knowledge base articles, help center pages, and self-serve content. Built from Simpplr design system standards for copy, formatting, accessibility, and terminology.
12 chapters and 3 appendices on screenshots, demo videos, and AI-generated hero images for articles and release notes, with annotated examples from real product screens. The scope is product, help, and self-serve documentation, with separate rules for marketing visuals.
Six request types, each with its own SLA. An SLA starts only when a linked ticket, a scope boundary, a reproducible environment, and a named subject matter expert are in place.
A WCAG-compliant reference and internal writing standards for the technical writing team.
Knowledge base articles, help center pages, release notes, and self-serve documentation. Covered by the writing guidelines, visual documentation standards, request template, and accessible language reference above.
Product text in the interface. Vernacular is the content design system for Simpplr and its AI-native product. Accordis checks, scores, and rewrites copy against Vernacular, with 300+ rules. Full write-ups are available on request.
Standards are how a knowledge base stays consistent after the person who built it moves on. The writing guide sets how pages read, the visual standards set how they look, and the request template sets how work enters the queue. Together they make the upkeep step of a knowledge base routine and checkable.
Simpplr and Adobe · Content strategy
Good knowledge starts before anyone writes. For every feature and release, I decide what content is needed, how deep it goes, and where it lives. I base those decisions on what users need and what the business needs.
I own this synthesis. A PM supplies feature intent, support supplies ticket risk, and I turn both into a documentation architecture. Knowledge base and technical documentation are enablement and self-serve material, not marketing collateral, so they have their own owners.
Tight. State what changed and who it is for, using a "Who's this for" line. Cut extra narrative, and use plain words that customers use instead of internal terms such as phased rollout or tenants.
Full. Keep context, benefits, best practices, and admin guidance. Cut only promotional language, duplicate summaries, and restated benefits.
A release cycle produces both, along with enablement materials, demo and voice-over scripts, and UX microcopy, so each piece needs a clear job.
I treat the Help center as a product with its own release story, not as a support leftover. It uses product voice and tone, not marketing voice. This is why I wrote separate visual guidelines for product and help content, and why I keep documentation and marketing content under different owners.
Read the full story in Adobe HelpX.