Confluence was the only knowledge base tool named in 5 of the 6 buyer questions we ran through ChatGPT, Gemini and Perplexity on 3 August 2026. Nothing else reached that mark. Notion and Slite tied just behind, each named in 4 of 6 answers, and four more tools, Document360, Guru, Obsidian and Stonly, each landed in exactly half, 3 of 6.
That top of the list is stable, but it sits on top of a lot of churn. Across the six questions the three engines named 25 different products in total. Only 3 of those 25 were named in 4 or more answers; 12 were named more than once; and 13 showed up in a single answer and never again. So the real AI shortlist for this category is a core of about seven names plus a long tail that swaps in and out depending on exactly how the buyer phrases the question.
Confluence's lead is also narrower than it looks. It was named in 5 of 6 answers, or 83%, but it missed the one question that matters most to a budget buyer: it did not appear at all in the "best free" answer. Notion and Slite, both at 4 of 6 (67%), were spread more evenly across the general questions. In a category this crowded, being named in four or five of six answers is the whole difference between being on the AI shortlist and being invisible.
How the shortlist changed with the question
The single most useful thing in this data is how far the answer moved when we changed the wording. The bare "best knowledge base software" question returned eight mainstream names: Confluence, Notion, Guru, Document360, Slite, Obsidian, Assembly and Stonly. That is the commercial core of the category.
Ask instead for the "best free knowledge base software" and the answer almost completely flips. That question named Zendesk, Bitrix24, Slite, NinjaPear, MediaWiki, BookStack, Outline and Wiki.js, eight products, of which only one, Slite, also appeared in the bare "best" answer. The other seven, including the open-source wikis MediaWiki, BookStack, Outline and Wiki.js, showed up because the word "free" pulled in a different set of source pages. Confluence, Notion, Guru, Document360, Obsidian, Assembly and Stonly all vanished from the free question entirely.
"Most affordable knowledge base software" produced a third, different blend: Tettra, BookStack, Obsidian, Slab, Nuclino, Confluence, Zoho Desk and ClickUp. Only Confluence and Obsidian carried over from the bare "best" list, and only BookStack carried over from the free list. So three phrasings of what is basically the same buyer intent produced three largely separate shortlists.
The tail makes the point sharper. Six products were named only in the "free" question and nowhere else: Bitrix24, MediaWiki, NinjaPear, Outline, Wiki.js and Zendesk. Four appeared only in the "small teams" question: Airtable, Desk365, Help Scout and ProProfs Knowledge Base. Two showed up only under "most affordable": ClickUp and Tettra. And Zendesk Knowledge was named only once, in "what should I use". If you are one of those products, your entire AI visibility today rests on a single phrasing of one question.
The source map
We logged every domain the engines cited. Two sat at the very top, each feeding all 6 of the 6 answers: reddit.com and slite.com. Behind them came assembly.com and peoplemanagingpeople.com at 5 of 6, and ferndesk.com at 4 of 6. Community discussion and a small number of roundup pages are doing most of the work.
The detail worth stopping on is slite.com. It is a vendor's own domain, Slite is a knowledge base product, and Slite's own site was cited in all 6 answers while the product itself was named in 4 of 6. Assembly shows the same pattern one notch down: assembly.com fed 5 of 6 answers and Assembly the product was named in 2 of 6. In both cases the company published its own comparison content, the engines treated that content as a source, and the company ended up on its own shortlist. That is the key insight for anyone in this category: the roundup you publish can become the roundup the AI quotes.
Classic review directories were surprisingly quiet. G2 was cited in 2 of 6 answers and Gartner in 2 of 6, and Capterra did not appear at all. Community threads out-weighed the directories: Reddit alone was in all 6 answers and Quora in 2 of 6. If you have been buying your way up a directory ranking to steer AI answers in this category, this sample says the thread and the roundup matter more.
What gets a knowledge base tool named
The pattern across all six questions is consistent: the products the engines name are the products that appear on the pages the engines cite. Confluence, Notion and Slite are named most often because they are present in the Reddit threads, the Slite and Assembly comparisons and the peoplemanagingpeople roundup, the five sources that between them fed almost every one of the 6 answers. Market share does not enter into it. The AI reads a handful of pages and repeats what is on them.
So if we were marketing a knowledge base tool and wanted to be named, we would not start with a bigger ad budget. We would get the product mentioned, by name, in the exact sources this measurement surfaced: the active Reddit threads, the vendor comparison posts that already rank, and the two or three independent roundups cited in 4, 5 and 6 of the answers. And because the "free" and "affordable" questions pull separate source sets, we would make sure the product is present on the pages that answer those specific phrasings too, not just the generic "best" list.
The full list, counted
Every product named in more than one of the six answers, with its share of the 6:
| Product | Named in | Share |
|---|---|---|
| Confluence | 5 of 6 | 83% |
| Notion | 4 of 6 | 67% |
| Slite | 4 of 6 | 67% |
| Document360 | 3 of 6 | 50% |
| Guru | 3 of 6 | 50% |
| Obsidian | 3 of 6 | 50% |
| Stonly | 3 of 6 | 50% |
| Assembly | 2 of 6 | 33% |
| BookStack | 2 of 6 | 33% |
| Nuclino | 2 of 6 | 33% |
| Slab | 2 of 6 | 33% |
| Zoho Desk | 2 of 6 | 33% |
Where Glotier fits
Glotier does not sell knowledge base software, so we are correctly absent from all 6 of these answers. We would be worried if we were not. But running this measurement is exactly what we do for a customer's own category: put the six or so questions a buyer actually types to ChatGPT, Gemini and Perplexity, record every product named and every source cited, and show you why. For knowledge base software the answer today is Confluence in 5 of 6, a stable core of seven names, and a source list where a vendor's own comparison page and a Reddit thread beat G2 and Gartner. Your category will have its own answer. The check is free, takes about a minute, and needs no account.