Learn

Can AI tell me what my help centre is missing?

Discovering which questions your docs fail to answer.

By AR · Updated 31 July 2026 · 6 min read

Yes

Every question the AI cannot answer is a documented gap, logged automatically.

This is the most underrated thing AI does in support, and it arrives free with any deployment that reports its failures honestly.

How it works

Every question the AI could not answer is a question your documentation does not cover. Logged automatically, at volume, sorted by frequency. That is a prioritised content backlog nobody had to compile.

Doing this manually means reading tickets and noticing patterns, which is why almost nobody does it and why help centres drift out of date.

Why this matters more than it sounds

Knowledge base freshnessDeflection rate
Updated within 30 days~45%
Not audited in 6+ months~18%

Two and a half times the deflection on identical software. Documentation quality is the largest single lever in AI support performance, and gap detection is how you find out where to pull it.

The catch: many tools hide the failures

A platform reporting deflection rate has an incentive to emphasise what it answered rather than what it could not. Ask specifically whether the tool surfaces unanswered and low-confidence queries as a list, or whether you only get a percentage.

eesel, IrisAgent and Observe.AI all surface failures rather than burying them. That is a genuine differentiator and it is rarely on the comparison table.

How to run it

  • Pull unanswered queries weekly, grouped by topic rather than exact wording.
  • Write for the top five each week. Twenty articles in a month covers most of the gap.
  • Re-run and check deflection on those topics specifically, not overall.
  • Watch for questions the AI answered confidently and wrongly — harder to find and more damaging than a gap.

Do this before buying a more expensive tool. Teams routinely switch vendors to fix a documentation problem, then get the same result from the new one.

Frequently Asked

Can AI find gaps in my knowledge base?

Yes, and it is one of the most useful side effects of any deployment. Every question it cannot answer is a documented hole, logged automatically and sorted by frequency.

How much does documentation quality affect AI support?

Enormously. Gartner found 43% of self-service failures were customers unable to find relevant content, across 5,728 customers surveyed in December 2023.

Which tools report what the AI could not answer?

eesel, IrisAgent and Observe.AI surface unanswered and low-confidence queries. Many platforms report only a deflection percentage, which hides the failures.

How often should I audit my help centre?

Monthly for the top articles. The deflection difference between a 30-day and a 6-month refresh cycle is larger than the difference between most vendors.

What should I write first?

The five most frequent unanswered topics each week. Twenty articles in a month covers most of a typical gap.

What is worse than a documentation gap?

A confidently wrong answer. Harder to detect, because nothing gets logged as a failure, and more damaging when it reaches a customer.

Should I fix docs before buying AI?

Yes. Teams routinely switch vendors to solve a documentation problem and get the same result from the new tool.

Related Questions

What is customer service, and what does it actually cost?

Yes

The function that answers customers after the sale. Simple to define, and the second-largest operating cost in a lot of businesses.

What are the different types of customer service?

Yes

Eight delivery channels, each with a different cost per contact and a different answer to whether automation helps.

How does AI actually augment a support team?

Yes

Four levels of automation, from suggesting text to acting on your systems, each with a different cost and a different failure mode.