To write help docs that cut support load before you hire, don't build a knowledge base. Write five short pages. Pick them by counting the questions that came in three or more times in the last 30 days, title each one with the user's own words, and write every page to the same template: a one-sentence answer, three to seven numbered steps, what to check if it fails, and one way to reach you. Then put each doc where the question starts (next to the button, inside the error message, in your reply macro), and compare repeat questions for 14 days before you write a sixth. Five well-placed pages beat forty that nobody finds.
Last updated: Nov 18, 2026
This update adds Gartner's self-service numbers (only 14% of issues fully resolved in self-service, 43% of failures because people could not find the right content), a picker table for the five docs, a copy-paste doc template, a weak-vs-better table, a comparison of help center tools with prices checked on 18 November 2026, a note that Google stopped showing FAQ rich results in May 2026, a 14-day measurement sheet and a worked example. The studies, prices and docs are real and linked. The Slotwren example is made up.
Why write five help docs instead of a full knowledge base?
Most early help centers fail in one of two ways. The founder never ships docs because "we need a full KB", or they ship thirty thin pages that read like a feature list. Both keep the inbox full.
Five is enough to cover the questions that actually repeat when you have tens or low hundreds of users. It is also small enough to write, place and maintain in two weeks without pausing product work. Your changelog that brings silent users back already teaches people what shipped; help docs teach them how to get unstuck without pinging you.
The rule of thumb: if a question showed up three or more times in the last 30 days, it is a doc candidate. If it showed up once because someone hit a bug you already fixed, it is not.
Does self-service actually cut support load?
It can, but less often than help desk vendors suggest, and the reason it fails is the useful part. Three findings worth knowing before you write a word:
Most people try it, few finish there. In a Gartner survey of 5,728 customers (run in December 2023, published August 2024), 73% used self-service at some point, but only 14% of issues were fully resolved there. Even for issues customers called "very simple", only 36% were.
The top failure is findability, not writing. In the same survey the most common reason self-service failed was that people could not find content relevant to their issue, in 43% of cases. 45% said the company didn't understand what they were trying to do.
Help starts outside your site. A later Gartner survey of 5,801 customers (January and February 2025) found 51% of service journeys begin on third-party platforms such as Google, YouTube and ChatGPT, with search engines the most common starting point. 74% of Gen Z customers start with a third party.
These are surveys of consumers dealing with all kinds of companies, not small SaaS products, so don't copy the percentages into your own plan. Copy the lesson. A help doc that exists but can't be found does nothing, and a doc that answers a question nobody asked does less. That is why this playbook spends as much time on picking and placing the five as on writing them.
One more number that shapes how you write. Jakob Nielsen's analysis of browsing data, How Little Do Users Read?, found people have time to read at most 28% of the words on an average page view, and 20% is more likely. On pages of 111 words or fewer they read about half. It's an old study (2008), but anyone who has watched a user skim a help page will recognise it. Put the answer in the first sentence, because that may be the only sentence they read.
How do you find the five questions worth a doc?
Open one spreadsheet. Columns: question (one line), channel (email / Discord / X DM / in-app), count, last seen, current answer (paste your last reply), doc candidate (Y/N).
Pull the raw pile. The last 30 days of support email, Discord #help, X DMs and any chat widget history. Don't clean it yet. Dump subjects and first lines.
Cluster in 20 minutes. Merge "how do I invite a teammate", "where is the invite link" and "can I add my co-founder" into one row: Invite a teammate. Do the same for billing, export, connecting the integration people ask about, cancel, and "is my data private".
Count ruthlessly. A cluster with a count of three or more is in the running. Sort by count, highest first. The top five that are how-to or policy questions become your set. Product bugs go to a fix queue, not to docs.
Kill vanity topics. "What is our vision?" and "full API reference" can wait. You are cutting repeat load, not impressing investors.
Two extra columns make the next step easier. Add "words they used" and paste the exact phrasing from two or three messages. Those words become your titles. Then add "where it started": the screen, email or error the person was looking at when they wrote in. That column tells you where the doc link has to go.
When the same tickets keep pointing at product gaps, not missing words, use a sibling playbook: turn support tickets into your next product priorities. Docs answer confusion. The roadmap answers missing capability. Don't confuse the two.
Which five help docs should a small SaaS write first?
Write the five titles as user questions, not product nouns:
Good: "How do I invite a teammate?" Bad: "Teams."
Good: "How do I connect Slack?" Bad: "Integrations overview."
Good: "How does billing and cancellation work?" Bad: "Plans and pricing FAQ."
Your spreadsheet decides the set, but most small products end up with something close to this. Use the table to check you haven't missed an obvious one, not as a list to copy:
Doc | Write it if you hear | Link it from | Skip it if |
|---|---|---|---|
Getting started: first win in under 10 minutes | "I signed up, now what?", "Is it working?" | Welcome email, empty state of the main screen | Your onboarding already gets most people to the first result |
Invite or share access | "How do I add my co-founder?", "Where's the invite link?" | Invite dialog, team settings | Your product is single-user only |
Connect the one integration people ask about | "Does it work with Slack / Stripe / Zapier?", "Connection failed" | Integration settings, the connection error | Nobody asks about integrations yet |
Billing, invoices and cancel | "Can I get an invoice?", "How do I cancel?", "Why was I charged?" | Billing page, receipt emails, pricing page | Never skip it once you take money |
Data, privacy or export | "Can I export my data?", "Who can see this?", "Do you train AI on it?" | Export button, settings, privacy page | Your product holds no user data worth worrying about |
If one row doesn't apply, the sixth-highest cluster in your sheet takes its place. Then freeze the list for 14 days. No sixth page mid-sprint. If a new question spikes, park it in the spreadsheet for the next round.
What should every help doc include?
Use the same skeleton for all five. Users learn the shape, and you write faster. Zendesk's own knowledge base content guide recommends the same thing: a template so authors include the right information and readers know what to expect, and articles short enough to scan.
Title as the question. Use the exact words from the cluster when you can.
One-sentence answer. Lead with the outcome. "Yes, you can invite teammates from Settings > Team. Here's how."
Steps (3 to 7, numbered). One action per step. Screenshots only where the UI is easy to miss. No essay between clicks.
If this fails. Two bullets for the usual failure modes (wrong plan, missing permission, an invite that went to spam). Tell people what to check before they write to you.
Still stuck? One line with a single reply channel and what to include (account email, a screenshot, what they already tried). Don't offer three support paths.
Here is the template as a fill-in table. Copy the middle column into a blank page and replace each line:
Section | What to write | Example (invite doc) |
|---|---|---|
Title | The question, in the user's words | How do I invite a teammate? |
Answer | One sentence: yes or no, where, and who can do it | Any admin can invite teammates from Settings > Team; they get an email with a link that works for 7 days. |
Before you start | Plan, role or setup needed (leave out if none) | You need the Admin role. Invites count toward your plan's seat limit. |
Steps | 3 to 7 numbered actions, each starting with a verb and using the exact button label | 1. Open Settings. 2. Select Team. 3. Click Invite. 4. Enter their email and pick a role. 5. Click Send invite. |
What you should see | The visible sign it worked | Their email appears under Pending with the date you sent it. |
If this fails | The two most common causes and the fix for each | No email after 5 minutes: ask them to check spam, then click Resend. "Seat limit reached": remove an old member or change plan. |
Still stuck? | One channel, what to include | Reply to any email from us with your account email and a screenshot of the Team page. |
Last checked | The date you last followed the steps yourself | Last checked: 18 Nov 2026 |
Aim for 250 to 450 words per doc. If you need more, you are documenting three jobs; split them later, not now. Zendesk describes doing exactly that on its own help center: customers couldn't work out how to change their account owner because the answer was buried in one long article about users, and breaking it into shorter topics fixed it.
Write in the voice you already use in DMs: short and specific, and slightly imperfect English is fine. Intercom's guide on how to write great help articles makes a point worth stealing: write about the job the customer wants done, not the feature they are trying to use. Draft tip: paste your best earlier reply into the template and delete the chatty parts. You already wrote the answer once. Reuse it.
What does a weak help doc look like next to a better one?
Most weak docs aren't wrong. They are written from the builder's side of the screen. Check your five drafts against this table:
Weak | Better |
|---|---|
Title: "Team management" | Title: "How do I invite a teammate?" |
Opens with "Our powerful collaboration features let you..." | Opens with the answer: "Admins can invite teammates from Settings > Team." |
"Navigate to the appropriate section" | "Open Settings, then select Team" (exact labels, in order) |
One step that hides three clicks | One action per numbered step |
Ten screenshots, one per click | One or two screenshots, only where the button is easy to miss |
Ends after the last step | Says what you should see when it worked |
No mention of failure | "If this fails" with the two causes you see most |
"Contact us via email, chat, Discord or X" | One channel and what to include |
Explains every setting on the page | Covers the one job, links to a second doc for the rest |
Teaches a workaround for a bug | Says it's a known bug, gives the workaround, and gets updated when the fix ships |
No date anywhere | "Last checked" date at the bottom |
How do you write steps people can actually follow?
Steps are where most help docs lose people. A few habits fix nearly all of it:
Use the exact label on screen. If the button says "Send invite", write "Send invite", not "submit". Bold UI labels so a skimmer can find them.
Start each step with a verb. Open, select, click, enter, copy. One verb, one action.
Say where before what. "In the top right, click your avatar" is easier to follow than "Click your avatar in the top right".
Write the expected result. After the step where things usually go wrong, add a short line: "You should see a green Connected badge."
Put the failure under the step it belongs to. If step 3 fails because the browser blocked a pop-up, say so under step 3, not in a section at the end.
Use screenshots sparingly. Crop to the area that matters, blur any customer data, add short alt text that says what the image shows, and plan to replace them when the UI changes. Every screenshot is a thing that can go stale.
Test on a fresh account. Follow your own steps from a new account with the lowest plan and role the doc is for. Your admin account with demo data hides half the problems.
If a step needs a paragraph of explanation, that is usually a product problem. Note it in your fix queue. The best help doc for a confusing screen is sometimes a clearer screen.
Where should help docs live so users find them?
A doc nobody can find is worse than no doc: you still answer the DM, and you feel guilty about the unused page. Remember Gartner's 43%. Place all five in three spots on day one:
Public help URL.
/helpor/docswith a simple list of the five titles. You don't need a search widget at this size. Link it from the site footer on every page.In-app links. Put the link near the UI that triggers the question. Invite dialog: "How invites work." Billing page: the billing doc. Empty state for the main job: the getting-started doc. One contextual link beats a buried Help icon.
Reply macro. Your default first reply becomes two sentences, the doc link, and "reply if that doesn't cover it." Save it as a snippet in your email client, chat tool or Discord.
Two more places are worth the ten minutes. The first is your error messages: when a connection fails or a card is declined, the message can link straight to the doc that explains what to do, which is the idea behind writing error messages that keep users moving. The second is your transactional email. Receipts, renewal reminders and failed-payment emails create a lot of billing questions, and a link to the billing doc in each one helps; the IndieHunt guide to billing emails that cut support tickets covers what each of those emails should say.
Optional: one line in your weekly changelog or onboarding email pointing to /help. Don't build a chatbot that guesses which of five pages to open.
Which help center tool should a solo maker use?
For five docs, the honest answer is whatever you will actually update. A plain page in your own app or marketing site is fine and costs nothing. If you want a hosted help center, these are the common options, with plans and prices as shown on each vendor's own page on 18 November 2026. Prices change, so check before you sign up.
Option | Free route | Paid starts at | Good fit when | Watch out for |
|---|---|---|---|---|
A | Free | Nothing extra | You have five docs and a deploy you already trust | No search or feedback buttons unless you build them |
Help Scout Docs | Free plan: 1 Docs site, up to 10 articles, 5 users, replies to 100 contacts a month | Standard at $25 per user per month (unlimited articles) | You answer support by email and want docs and inbox in one place | The 10-article cap is fine for this playbook, tight soon after |
Intercom | 14-day trial only | Essential at $29 per seat per month ($19 for new customers paying annually), public help center included; Fin AI agent $0.99 per outcome | You already use in-app chat and want the help center next to it | AI outcomes are billed on top of seats |
Crisp | Free plan has chat but no knowledge base | Essentials at $95 per month per workspace includes the knowledge base | You use Crisp chat and want docs inside the same widget | The knowledge base starts on the third tier, not Mini ($45) |
GitBook | Free plan for individuals | Essential at $65 per site per month billed annually ($79 monthly), adds custom domain | You want a docs site with Git sync and a clean editor | Custom domain needs a paid site plan |
Mintlify | Free plan with 5 editors and a custom domain | Starter at $150 per month billed annually ($180 monthly) | Developer-facing product, docs kept in Git next to the code | Built for developer docs; heavy for five how-to pages |
Notion Sites | Publish pages to the web on a free Notion subdomain, with search engine indexing | Paid Notion plan for SEO settings and themes; your own domain is a paid add-on | You already write everything in Notion | Easy to grow into the wiki nobody opens |
Docusaurus | Open source (MIT licence), free | Only your hosting | You like Markdown and static sites | You maintain the build and the theme |
For a solo maker starting from zero, the cheapest sensible path is a /help page you already host, or Help Scout's free plan if you also want a shared inbox. Its 10-article limit happens to fit five docs with room for the next round. Move to something bigger when the sheet says you need more than ten pages, not before.
Should your help docs be public and indexed by Google?
Usually yes. Gartner's 2025 survey says search engines are the most common place a service journey starts, and more and more people ask ChatGPT and other AI assistants before they open your site. A public page with a clear question as its title is something those tools can find and quote. A help page hidden behind a login is invisible to them. Keep three things in mind:
Make the title match the search. "How do I cancel my Slotwren subscription?" matches what people type. "Account lifecycle" matches nothing.
Don't bother with FAQ schema for rich results. Google limited FAQ rich results to well-known government and health sites in August 2023, and its Search Central documentation updates say the FAQ rich result stopped appearing in Google Search from 7 May 2026. Plain headings and a clear first sentence do the work.
Keep account-specific pages private. Anything that shows a customer's data, internal admin steps or security details stays behind the login or out of the help center.
Support docs and acquisition docs are different jobs. These five exist to stop repeat questions from people who already signed up. If you later want docs that bring in new users from search, that is its own project with its own picking rules, covered in the IndieHunt guide to getting organic signups from product docs without a blog. Don't let the SEO version hijack this sprint. Rankings are a bonus; repeat DMs are the brief.
How do you measure whether the docs cut support load?
You are not aiming for zero DMs. You are aiming for fewer repeats of the same five.
Baseline (day 0). From your spreadsheet, note the weekly count for each of the five clusters before the docs go live.
Ship day. Publish all five, plus the footer link, in-app links and reply macro. Announce once in Discord or your customer email if you have one: "Five short how-tos are live at /help."
Log for 14 days. For every inbound question that matches a cluster, mark "sent doc link" or "doc failed / needs product fix". Keep weekly counts.
Success bar. Aim for roughly half the repeat volume on those five clusters by day 14, or a clear drop in time spent per reply because you paste a link instead of rewriting steps. If volume didn't move, the placement failed, not the word count.
A simple sheet is enough. One row per doc:
Doc | Questions per week before | Week 1 after | Week 2 after | Times you sent the link | "That didn't work" replies | Decision |
|---|---|---|---|---|---|---|
Getting started | - | - | - | - | - | Keep / rewrite / fix product |
Invite a teammate | - | - | - | - | - | - |
Connect the integration | - | - | - | - | - | - |
Billing and cancel | - | - | - | - | - | - |
Data and export | - | - | - | - | - | - |
With tens of users, the counts are small and noisy. Three questions one week and one the next might be luck. Look for the direction across all five and read the "that didn't work" replies closely; they tell you exactly which step is broken. If your tool has a "Was this helpful?" button or page views, use them as hints, but the question count in your inbox is the number that matters.
Treat the exercise like other small founder ops habits: one focused sprint, then decide. Same spirit as stopping treating support tickets like interruptions: batch the pattern, don't live in the inbox.
A made-up example: Slotwren writes five docs in two weeks
Slotwren is an imaginary booking tool for private tutors, run by one founder with about 140 paying users. The names and numbers below are invented to show the method, not results from a real product.
In the last 30 days she had 61 support conversations. Clustering took 25 minutes and gave her this:
"How do I connect Google Calendar?" or "my bookings aren't showing": 14
"How do parents pay?" or "where's my payout?": 11
"How do I change my available hours?": 9
"Can I get an invoice for my subscription?" or "how do I cancel?": 7
"Can a parent reschedule without messaging me?": 6
Bugs (a time zone display bug, a broken reminder email): 8, sent to the fix queue
Everything else, under three each: 6
Her five docs were the top five clusters, titled in her users' words. She wrote them in four evenings by pasting her best old replies into the template. The calendar doc got one screenshot (the permission screen people kept missing) and an "If this fails" line about the second Google account in the same browser, which turned out to cause most of the failures.
Placement took one more evening: a footer link to /help, a link inside the "Calendar not connected" warning, a link on the payouts page, the billing doc in every receipt email, and a reply macro.
After two weeks, calendar questions dropped from about three a week to one, and payout questions from about three to two. The reschedule doc barely moved anything. Two parents replied "I don't see a reschedule button", because the button only appears 24 hours before a lesson. That wasn't a docs problem. She moved the button into the confirmation email, rewrote one step, and parked "Can I use Slotwren for group lessons?" (four mentions) as the candidate for doc number six.
A 14-day plan to ship five help docs
Days | Work | Done when |
|---|---|---|
1 to 2 | Mine 30 days of channels, cluster, count, freeze five titles | Five question titles in the user's words |
3 to 6 | Write all five with the template, one per focused block; test each on a fresh account | Every step followed once, "Last checked" dates filled in |
7 | Ship the | Each doc is linked from where its question starts |
8 to 13 | Answer with links first, log every matching question and every "that didn't work" | Two weeks of counts in the sheet |
14 | Compare counts; keep, rewrite the weakest doc, or move one parked question into a sixth page only if it reached three or more | One decision per doc |
If a friend or power user will skim one doc on day 6, take the offer. Watch where they hesitate. Every pause is a missing step or an unclear word.
What mistakes keep the inbox full?
Building a full knowledge base first. Categories, tags, search, versioning. You'll spend a week on structure and still owe the five answers. Ship five pages on a boring list first.
Docs nobody can find. Beautiful articles with no footer link, no in-app link and no reply macro. Placement is half the product.
Feature lists dressed as docs. "Overview of Automations" with every toggle named. Users ask how to finish a job. Answer the job.
Writing for SEO before writing for tickets. Start from the spreadsheet, not from keyword tools.
Documenting bugs as how-tos. If the real fix is a product change, file it and say so. A help page that teaches a painful workaround forever trains users to hate you politely.
Letting an AI bot answer from stale docs. Help desk AI agents typically answer from your help content. If the billing doc is wrong, the bot will be confidently wrong at 3 AM. Fix the five docs before you switch one on.
Three support channels in the "Still stuck?" line. People pick the one you check least. Give one.
Updating docs only when you feel guilty. When the UI changes, update the matching doc the same day you ship.
How do you keep five docs from going stale?
Stale steps recreate the inbox you just emptied, and with five docs, keeping them current is a small job if you attach it to work you already do:
Tie docs to releases. Before you ship a change to a screen, search your five docs for that screen's name. If one mentions it, update the doc in the same release.
Keep the "Last checked" date honest. Only change it when you have followed the steps again. Readers and future you can see how fresh each page is.
Do a 20-minute check once a month. Follow each doc on a fresh account, retake any screenshot that no longer matches, and run each help page through a free broken link checker so a moved settings page or dead integration link doesn't send people to a 404.
Read the "that didn't work" replies. Two people stuck on the same step means the step is wrong, even if it worked for you.
When is it time for a sixth doc, or a support hire?
After two weeks, "good" looks like this: you paste a help link more often than you rewrite steps, two of the five clusters are clearly quieter, and you know which doc is weak because people still reply "that didn't work". You haven't hired support, and you shouldn't need to until the remaining questions are genuinely new or high-stakes.
Add the sixth doc the same way you picked the first five: only from a cluster that reached three or more after the five shipped. Grow by one or two pages a month from the sheet. If you notice you are writing docs to explain screens that keep confusing people, stop and fix the screen.
A support hire starts to make sense when the questions left are not repeats: account-specific problems, refunds that need judgment, customers who need a person on a call. Until then, five well-placed pages beat a ghost-town wiki.
Help docs FAQ
How long should a help article be?
Long enough to answer one question, usually 250 to 450 words. Put the answer in the first sentence; many readers will stop there. If a doc keeps growing past that, it covers more than one job, so split it.
How many help articles does a new SaaS need?
Fewer than you think. Start with the five questions that repeat most in your inbox and add pages one at a time when a new question shows up three or more times in a month.
Do I need a help center tool, or is a page in my app enough?
For five docs, a plain /help page is enough. A tool earns its place when you want search, article feedback, an in-app widget or docs inside the same inbox you answer from.
Should I add FAQ schema to my help pages?
Not for rich results. Google says the FAQ rich result stopped appearing in Search in May 2026, after being limited to government and health sites since 2023. Clear question titles and answer-first paragraphs are what help people and search tools.
Can an AI chatbot replace help docs?
No. Help desk AI agents mostly answer from your help content, so they need good docs to work at all. Write the five first; a bot can come later if your volume justifies it.
Write the five docs your users already asked for
Your inbox has already written the brief. Count the repeats, write five short pages with one template, put each link where the question starts, and give it 14 days. If you are launching or listing a maker tool and want a home built for builders, explore MakerHunt, then go write the five docs your users already asked for.