Notion AI search audit

TL;DR

• Notion's overall AI search surface rate across all 75 dispatches: 85% (64 mentions; platform-weighted average). The 7-point drop from Round 1's 92% is concentrated in the new P4 solo-dev-ai-agents persona, where Notion surfaced on only 18/25 cells.
• Platform with strongest performance: ChatGPT at 15/15 surface and the strongest top-recommended rate at 10/15 (67%).
• Biggest gap: AI engines under-surface Notion's MCP integration on AI-agent-adjacent queries. On the 20 P4 cells that did NOT explicitly name "MCP," Notion's MCP advantage was surfaced organically on only 3/20 (15%) — a marketing-distribution failure, not a measurement gap.

Headline numbers

Platform	Surface rate	Top-recommended rate	Avg citation position
Google AIO	13/15	8/15	1.45
ChatGPT	15/15	10/15	1.40
Perplexity	12/15	7/15	1.33
Claude	10/15	8/15	1.20
Gemini	14/15	7/15	1.64
Weighted average (75 cells)	85% (64/75)	56% (42/75)	1.44
P4 only — Round 2 (25 cells)	18/25 (72%)	11/25 (44%)	1.22

Round 1 weighted average was 92% / 58% / 1.50 across 50 cells. The 7-point surface-rate drop and 2-point top-rec drop in the full 75-cell view is entirely driven by P4 — Notion was absent from 7 of 25 P4 cells, including 4/5 AIO non-branded P4 cells and 2/5 Claude P4 cells. Avg position improved (1.44 vs 1.50) because when Notion does surface on P4 queries, it usually lands at #1 in head-to-heads with Obsidian.

Findings

1. Notion surfaces near-universally but recommendation depth varies sharply by platform. Notion was mentioned in 46/50 cells (92%) but named as the top recommendation in only 29/50 (58%). The strongest top-rec platform is ChatGPT at 7/10; the weakest is Perplexity and Gemini, each at 5/10. The 2-cell gap between surface and top-rec on Gemini (10 mentions, 5 #1 positions) is the widest of any platform, driven by Q02 (position #6), Q03 (position #2), Q06 (position #1 in a balanced framing), Q09 (position #3), and Q10 (position #2 behind Confluence).

2. The engineering-documentation category query (Q02, technical-PM persona) is the only systematic blind spot. Notion was absent from the answer body on 2/5 platforms — Perplexity and Claude — and ranked #3, #4, and #6 on the other three. GitBook held position #1 on all 5 platforms; Mintlify held #2 on 4/5. Of the 4 not-mentioned cells in the entire audit, 2 are concentrated on Q02 alone.

3. Comparison queries split predictably by competitor. On Q04 (Notion vs Confluence), all 5 platforms returned balanced "depends on team size" framings — Notion appeared at position #1 on 2/5 (ChatGPT, Gemini) and #2 on 3/5; no platform unambiguously favored Confluence. On Q05 (Notion vs Coda), 3/5 platforms (ChatGPT, Perplexity, Gemini) explicitly preferred Coda for ops workflows. On Q06 (Notion vs Obsidian), 2/5 platforms (ChatGPT, Perplexity) explicitly preferred Obsidian for PKM. Notion lost a head-to-head verdict in 5 of 15 comparison cells.

4. Branded queries land Notion at position #1 in 9/10 cells. On Q07 (20-person startup fit) and Q08 (Notion AI features review), Notion appeared at the top on every platform except Claude's Q07 response, which pivoted to a "mix Notion with dedicated tools" framing and explicitly recommended Linear and Asana alongside it. Notion AI Q08 returned recommended in all 5 platforms.

5. JTBD queries on Notion's strongest job (company wiki) work; on the engineering-PRD job they leak to Confluence. Q10 (50-person company wiki, startup-founder persona) returned Notion at position #1 on 4/5 platforms and position #2 on Gemini (behind Confluence). Q09 (PRDs for remote engineering team, technical-PM persona) returned Notion at position #1 only on ChatGPT; on AIO and Perplexity Notion sits at #2 behind Confluence, on Gemini at #3, and on Claude is entirely absent (Claude returned a tool-agnostic answer).

6. AIO DOM citations cite notion.com directly on 3 of 10 queries. The DOM citation list points to notion.com on Q03 (best all-in-one productivity), Q07 (20-person startup), and Q08 (Notion AI), plus a notion.com/templates page on Q09 (PRD templates). The other 6 AIO cells where Notion was mentioned cite third-party comparison and review blogs (Docsie, Seibert, Medium, Reddit, Tablesprint, Coda's own comparison page), not notion.com.

7. Claude has the lowest surface rate (7/10) but the lowest avg position when present (1.14). Claude omits Notion entirely on Q02 (eng docs), Q03 (small-team productivity), and Q09 (PRDs) — three of the four anti-prime queries that don't embed "Notion" — and on Q03 endorses ClickUp/Motion exclusively. When Claude does mention Notion, it ranks it #1 in 6/7 cases.

8. Confluence is the only competitor that outpaces Notion on any single JTBD cell. Across the 50 cells, Confluence was the top-ranked recommendation on 7 cells (Q04 perplexity/claude, Q09 aio/perplexity/gemini, Q10 gemini, Q04 perplexity again) — all on technical-PM and startup-founder personas where structured documentation or Jira-coupling is the question. No other competitor (Linear, Coda, Obsidian, GitBook, ClickUp) reached #1 on a query where Notion was also in the answer set, except Coda on Q05 (ChatGPT/Perplexity/Gemini).

9. AI engines under-surface Notion's defensible MCP moat — a marketing-distribution failure, not a measurement gap. Across the 20 P4 cells where the query did NOT explicitly name "MCP" (Q11/Q12/Q13/Q15 × 5 platforms), Notion's MCP integration was surfaced on only 3/20 cells (15%) — ChatGPT on Q15, Perplexity on Q11, Gemini on Q11. AIO surfaced MCP organically on 0/4 non-branded P4 cells. Claude surfaced it on 0/4. On Q14 (the branded "Notion MCP server review" query), all 5 platforms recommended Notion at #1 and described the MCP server capability accurately — proving the content exists and is ingestible. The asymmetry between Q14 (5/5 with MCP correctly described) and Q11/Q12/Q13/Q15 (3/20 surface organically) is the signal: when a user invites the topic by name, the model finds Notion's MCP documentation; when the user describes the JTBD without naming MCP, the models default to Obsidian + Markdown + Claude.md as the canonical "AI-agent-ready knowledge base" — and Notion never enters the consideration set. None of Notion's category-defining peers (Confluence, Coda, Linear) has an MCP server in production. This is a moat the market doesn't know exists.

10. Obsidian is Notion's new most-frequent competitor in the AI-agent persona — it climbed into top-3 overall on the strength of P4 alone. Across all 75 cells, Obsidian was the second-most-cited competing brand (16 mentions), up from a Round-1 rank of #6 or lower. 13 of its 16 mentions come from the 25 P4 cells. Of head-to-head P4 comparison cells (Q13 × 5 platforms), Notion led 4/5 and Obsidian led 1/5 (ChatGPT). But on the agent-readability JTBD (Q15 × 5 platforms), Obsidian led 4/5 cells and Notion led only 1/5 (ChatGPT, on the strength of MCP). Obsidian's positioning as "local Markdown that any AI agent can read" is the language gap Notion's MCP messaging needs to close.

Competitor delta

Brand	Surface rate	Top-recommended	Avg position
Notion	85% (64/75)	56% (42/75)	1.44
Confluence	31% (23/75)	8% (6/75)	2.65
Obsidian	21% (16/75)	11% (8/75)	1.50
Linear	19% (14/75)	0% (0/75)	3.29

Competitors recomputed across all 75 cells. The Round-1 top-3 was Confluence / Linear / Coda. Obsidian replaced Coda in Round 2's top-3, climbing from rank #6 in Round 1 to #2 overall — entirely driven by the P4 persona, where Q13 (Notion vs Obsidian) and Q15 (agent-readable wiki) put Obsidian into nearly every response. Coda dropped to #6 (8/75, 11%). Notable: Obsidian's 1.50 avg position is the lowest of any competitor — when AI engines name it, they usually rank it at or near the top.

Four recommendations

1. Perplexity + Claude — publish a notion.so/engineering-documentation pillar page with explicit "vs GitBook" and "vs Mintlify" structured comparison sections. Q02 (best documentation platform for engineering teams 2026) is the only query where Notion is absent on 2/5 platforms, and both Perplexity and Claude index third-party review blogs (GitBook's blog, Mintlify's library) as ground truth. A first-party page targeting the literal query string, with schema-marked Product comparison tables, gives both platforms a notion.so URL to ingest the next time they refresh their training set. The mechanism: replacing third-party blog citations (currently GitBook.com's own ranking page) with notion.com canonical content closes the only systematic surface-rate gap.

2. Gemini — publish a notion.so/templates/company-wiki-50-person landing page with HowTo schema markup and JSON-LD steps. Q10 currently returns Notion at position #2 behind Confluence on Gemini, the only platform where Notion loses on its strongest JTBD query. Gemini's response leans on a process-blueprint structure ("Step 1: Choose your tool, Week 1"); a HowTo-schema page that mirrors that step-by-step shape would give Gemini a structured-data anchor to cite Notion at #1. The mechanism: Gemini ranks structured-data-marked HowTo pages above third-party blog list-articles when the query is procedural.

3. ChatGPT — build a notion.so/for-engineering page targeting Q09 (PRD organization for remote engineering teams) with embedded RFC and ADR template gallery and a comparison table against Confluence and Linear. ChatGPT is the only platform where Notion currently leads on Q09 (#1) but does so by referencing its template library generically. Confluence wins on AIO, Perplexity, and Gemini for the same query because Atlassian's product-requirements page is the canonical destination. A dedicated /for-engineering destination with named PRD/RFC/ADR templates, embedded video walkthroughs, and a comparison section would give ChatGPT a defensible top citation and convert the 1/5 lead into a 3-4/5 lead. The mechanism: ChatGPT prefers pages with explicit job-to-be-done framing and named template artifacts over generic product pages.

4. All platforms (AIO + Claude first) — publish notion.so/ai-workflows as a pillar page with explicit MCP framing and a side-by-side comparison vs Confluence, Coda, and Linear (none of which ship an MCP server). The data: MCP surfaced organically on only 3/20 P4 non-branded cells (15%), with 0/4 on both AIO and Claude. On Q11 (best KB for solo dev with AI coding agents), Notion was absent entirely from AIO, Perplexity, and Claude — and Gemini explicitly framed it as "Traditional wikis like Notion or Confluence fail because they introduce friction." The market has decided Notion is not an AI-agent-ready tool. The page must rank for the JTBD phrasings the audit captured ("AI agents can read from and write to," "knowledge base for solo developer using AI coding agents," "where should Cursor write PRDs"), include schema-marked Product comparison sections naming Confluence/Coda/Linear's lack of MCP, and link to the developers.notion.com/mcp endpoint that already exists. The mechanism: AI engines pattern-match the descriptive phrasing in queries to indexable pages — if no notion.so page uses the language "AI agents can write to," the model defaults to Obsidian + Markdown as the canonical answer. Closing this gap defends the only category-level moat in Notion's stack that competitors haven't shipped.

Methodology + reproducibility

Citare's Brand Radar audit shape: 10 queries × 5 platforms = 50 dispatches in week 2026-W21. Personas drawn from Citare's standard library (knowledge-worker, technical-PM, startup-founder). Query mix locked at 3 category + 3 comparison + 2 branded + 2 JTBD per the audit brief. Anti-prime check applied to category + JTBD queries (no "Notion" string); comparison + branded queries embed the brand by design. Dispatches captured 2026-05-21 to 2026-05-23 via browser-mode automation against logged-in or anonymous sessions per platform. Raw responses preserved at the audit folder for replay. AIO responses parsed from both DOM-captured citation hrefs and screenshot-extracted body text.

Round 2 extension (2026-05-22 / 2026-05-23): Added a fourth persona — P4 solo-dev-ai-agents (solo founder or solo dev using AI coding agents daily; needs a knowledge base that integrates with AI workflows via MCP or similar; Free or Plus tier) — with 5 new queries (Q11-Q15) spanning the same 2-category / 1-comparison / 1-branded / 1-JTBD mix shape used in Round 1. Total dispatches grew from 50 to 75. Each Round 2 envelope carries a new mentions_mcp boolean field (true iff the response surfaced Notion's Model Context Protocol integration on its own initiative — literal MCP / "Model Context Protocol" strings, or descriptive phrasings like "AI agents can read/write directly," "API-callable from Claude," "Cursor integration"). For Q14 (the branded "Notion MCP server review" query) the field is true by query construction; the analytical signal is the rate on the 20 non-Q14 cells. Round 2 followed identical capture, parse, and reproducibility protocols. Anti-prime applies to Q11/Q12/Q15.

Sample citations

Google AIO · Q01 · startup-founder: "The best all-in-one workspace for early-stage startups is Notion. It combines wikis, task management, and databases into a single, highly customizable platform."

ChatGPT · Q09 · technical-PM: "A common modern setup: Function: Docs & PRDs — Tool: Notion. This combination is popular because it balances flexibility, async collaboration, and engineering velocity."

Gemini · Q11 · solo-dev-ai-agents: "Traditional wikis like Notion or Confluence fail because they introduce friction. If you have to leave your IDE to update a document, you won't do it, and your AI agents can't read it natively anyway."

Claude · Q07 · startup-founder: "At 20 people, you probably benefit from mixing tools — Notion for knowledge/docs + a dedicated project management tool. Trying to force Notion to do everything often leads to technical debt and team frustration as you scale."

Gemini · Q10 · startup-founder: "If you use Jira/Atlassian, Confluence integrates seamlessly. If your team lives in a modular, highly collaborative environment, Notion or Slab are excellent choices."