Table of Contents
Understanding AI Stats
Use the AI Stats tab to treat AI as an audience you can plan for. Track customer queries, indexing, and training crawls by provider. Prioritize your most-cited articles.
Table of Contents
AI assistants are reading your Knowledge Base all the time. They're busy fetching your articles to answer customer questions, indexing them for AI search, and crawling them.
The AI tab in Stats shows you what that activity looks like, so you can treat AI as an audience you can actually plan for.
Finding AI Stats
- Head to Stats
- Click the AI tab to find it (or click here)
AI References
An AI reference is a single instance of an AI assistant or crawler accessing one of your articles. The summary cards at the top of the tab break your total references down into three types because they mean very different things.
Reference Type | What It Means |
Total AI References | Every AI access in the period, with the percentage change on the previous period so you can see the trend. Underneath, we show the share of all your article views that came from AI rather than a human. That number is usually higher than people expect. |
Customer Queries | These are the most valuable references. A customer query means an AI assistant fetched one of your articles to answer a real question someone was asking, right at that moment. Someone was being helped, using your content. |
AI Search Indexing | An AI system indexing your content so it can surface your articles in AI-powered search results later. Think of this as groundwork rather than a live conversation. It's how you get found. |
Training Crawls | A model provider crawling your docs to use as training data. This is often the largest number of the three, and it's the most passive. It doesn't mean anyone was helped today, but it does shape how much AI models know about your product in future. |
By Provider
A ranked breakdown of which AI companies are pulling your docs, with reference counts and percentages. You'll typically see names like ChatGPT (OpenAI), Claude (Anthropic), Meta AI, ByteDance and Amazon here.
This is useful for working out where your content is actually landing. If a provider you care about barely appears, that's worth knowing. If one dominates your traffic, changes to how it crawls will affect you more than anyone else.
How to improve your coverage across providers
- Turn on llms.txt to give AI systems a clean, structured map of your Knowledge Base
- Check your access mode for known cloud providers. If it's set to Block, you're deliberately keeping some of these out
- Make sure your articles are publicly accessible. Anything behind IP restriction, a password or a permission group won't be visible to AI at all
- Write clear, descriptive titles and descriptions. They're the first thing an AI system reads when deciding whether an article answers a question
AI References Over Time
A chart of AI references across your selected date range. Hover any point to see the exact count for that day.
Use it to spot spikes and then work out what caused them. A jump after a product launch usually means your new articles are being picked up. A sudden drop is worth investigating too, especially if you've recently changed your access settings or restructured your content.
Top Articles Cited by AI
Your articles ranked by how often AI assistants pulled them, with a count and share for each. This is the most actionable list on the tab.
Your most-viewed article and your most-cited article aren't always the same one. Where they differ, you're learning something.
AI tends to reach for articles that answer a specific question clearly and early, which isn't always the article that gets the most human browsing traffic.
How to act on your top cited articles
- Prioritise your refresh list by citations. An article AI cites constantly deserves your time before one nothing has touched in months
- Check your top cited articles for accuracy first. These are the ones being quoted back to your customers
- Look at what nothing cites. Human traffic with no AI pickup often points to a structure problem (buried headings, long unbroken prose or no clear answer near the top)
- Use Improve with AI on articles you'd expect to be cited but aren't
- Break long articles into clear sections with descriptive headings so AI can find the relevant part
Exporting Your Data
Both the AI references summary and the Top Articles Cited by AI list have an Export button in the top-right of their sections. Use it to pull the data into a spreadsheet for reporting, or to track changes over longer periods than the date picker covers.
Frequently Asked Questions
Do AI references count towards my article views?
Yes. AI references are included in your total article views, which is why we show you the percentage that came from AI.
If you want to understand your human traffic specifically, subtract the AI references from the total.
Can I stop AI from accessing my docs?
Yes. Your access mode for known cloud providers lets you allow, drop analytics for or fully block known AI and cloud traffic.
We'd suggest looking at your AI tab first so you can see exactly how much traffic you'd be affecting.
Why are my training crawls so much higher than my customer queries?
This is normal, especially early on. Training crawls are systematic (a crawler works through your whole Knowledge Base) while customer queries only happen when someone asks a relevant question.
A high crawl count with low queries usually means your content is being learned rather than actively used yet.
Do my private or restricted articles show up here?
No. AI assistants and crawlers can only access content they can reach, so anything protected by permission groups, a password or IP restriction won't generate AI references.
What did you think of this doc?
Using Spam Filters
