A handful of numbers tell you whether your help center is working: whether people find answers, which answers need work, and whether fewer questions reach your team. Here's what to track, what each number tells you, and a monthly routine that turns the numbers into improvements.
Start with the questions you want answered
A metric is only useful if it leads to a decision. Most help center metrics answer one of four questions:
Are people using the help center?
Do they find what they're looking for?
Which articles need work?
Is it reducing the questions that reach your team?
The metrics that matter
Metric | What it tells you | A good sign |
|---|---|---|
Visitors | How many people use your help center | It grows along with your customer base. |
Searches | What people want, in their own words | The top searches match the topics you see in support. |
Failed searches | What's missing, or what readers call by a different name | They make up a smaller share of all searches over time. |
Article views | Which topics matter most to readers | Your most-viewed articles are also your most accurate. |
Satisfaction | Whether articles help, based on reader ratings | It holds steady or rises, and no popular article is rated poorly. |
Self-service rate | How many visits end with an answer instead of a message to your team | It rises, or holds steady while visitors grow. |
Contact requests | How many people still need a person | They grow more slowly than visitors. |
AI answer ratings | Whether readers found the AI answers helpful | The share of AI answers rated helpful rises. |
Look beyond your help center too. Your support tool's ticket volume and top ticket topics are the final test: as your customer base grows, repeat questions should take up less of your team's time.
Work out the rates
Raw counts grow with your traffic, so the rates tell you more:
Failed-search rate = failed searches ÷ all searches
Self-service rate = visits where someone searched or read an article and didn't contact your team ÷ all visits
Contact rate = contact requests ÷ visitors
Satisfaction = positive ratings ÷ all ratings
Read the numbers together
One number rarely tells the whole story.
High views and low ratings mean many people rely on an article that doesn't help them. Fix it first.
Repeated searches, where people search again straight away, mean the first results didn't help, even if something came up.
A popular troubleshooting article often points to a problem in the product. Share it with your product team.
More contact requests after a release often mean an article is missing or out of date.
What good looks like
Every product and audience is different, so compare your help center with itself rather than with someone else's benchmark. Month over month, look for:
Failed searches taking a smaller share of all searches.
A self-service rate that rises, or holds steady while visitors grow.
A shorter list of low-rated articles.
Fewer support tickets on topics you've documented.
Get clean numbers first
Leave out your own team. Writers and support agents open articles all day. Filter out your office's IP addresses so their visits don't count.
Leave out bots. Automated traffic inflates visitors and views.
Compare like with like: a month with the month before, not a week with a month.
Keep a short log of events. A release, a price change or a marketing campaign can move every number, and a note makes spikes easy to explain later.
Wait for enough ratings. Two or three ratings say little about an article.
Numbers that can mislead
Time on page: a long visit can mean a thorough article or a confusing one.
Bounce rate: someone who lands on the right article, reads it and leaves has succeeded.
Page views alone: more views can mean people are hunting for an answer they can't find.
A monthly routine
Compare the headline numbers with the previous month: visitors, searches, satisfaction and contact requests.
Open the failed searches. Pick the biggest few gaps, then write or fix those articles.
Review your lowest-rated articles. Read any written feedback, then fix the ones the most people read.
Read new comments, and look at the searches where readers rated the AI answer unhelpful.
Check that your ten most-read articles still match the product.
Share a short summary with your team: what you fixed, and what readers struggled with that your product team should know about.
If your statistics only keep a short history, run the routine every two weeks, so nothing drops out of view before you look.
How HelpCenter.io helps
Statistics in HelpCenter.io shows Visitors, Searches, Article views, Views / visitor and Satisfaction, each compared with the previous period, plus a What needs your attention list of things to fix. Visitor journeys follows visits from the first page to Self-served or Contacted support. Content performance lists your top and underperforming articles, and What people search for shows failed searches and content gaps. With AI Answers on, you also see how readers rated the AI answers and what people ask AI about your articles. Bots are left out unless you include them, you can filter out your own IP addresses, and how far back you can look depends on your plan. See Read your help center statistics, See what people search for and Find your best and weakest articles.
Comments
Be the first to comment.