{c.text}
++ Where {COMPETITOR} ends at a tidy meeting page, Knowcap continues — capturing unstructured calls, verifying every claim, and acting on the confirmed ones. +
+| Feature | +{COMPETITOR} | +Knowcap | +
|---|---|---|
| {row.feature} | ++ + {row.them} + | ++ + {row.us} + | +
+ Where {COMPETITOR} stops at automated CRM updates, Knowcap continues — through human verification, into real-world action. +
+| Feature | +{COMPETITOR} | +Knowcap | +
|---|---|---|
| {row.feature} | ++ + {row.them} + | ++ + {row.us} + | +
+ Where {COMPETITOR} ends at a clean desktop note, Knowcap continues — capturing mobile + WhatsApp + in-person calls, verifying each claim, and acting on the confirmed ones. +
+| Feature | +{COMPETITOR} | +Knowcap | +
|---|---|---|
| {row.feature} | ++ + {row.them} + | ++ + {row.us} + | +
+ Where {COMPETITOR} stops at the transcript, Knowcap continues — through verification, into action. +
+| Feature | +{COMPETITOR} | +Knowcap | +
|---|---|---|
| {row.feature} | ++ + {row.them} + | ++ + {row.us} + | +
+ Where {COMPETITOR} stops at the summary, Knowcap continues — through human verification, into real-world action. +
+| Feature | +{COMPETITOR} | +Knowcap | +
|---|---|---|
| {row.feature} | ++ + {row.them} + | ++ + {row.us} + | +
{c.text}
++ One client conversation. One tap from you. Four agents working in parallel. By the time the + call ends, your Odoo SH ticket is in code review. +
++ No ticket form. No “please file a bug report” email. The reproduction is + captured live, in the client's own words, with the screen as evidence. +
+ ++ Categorized as a Risk, ready for one-tap human review. The claim sits in your inbox + with the exact 24-second clip attached and the words quoted verbatim. +
+ ++ An agent reads it, creates a ticket in YOUR Odoo SH instance (Customer: Demo Client A, + Module: Warehouse), moves it from New → In Progress, and notifies your dev team in + Telegram. +
+ +
+ It identifies the unindexed loop in stock.picking.bulk_receive, generates
+ the patch, opens a PR. Ticket auto-moves to Code Review with the PR linked.
+
+ Same loop for features. Client describes the feature on a call → Task in + project → PR scaffold ready. You only ever review what a named human already confirmed. +
+
` (currently empty, 2 instances) | Accessibility + image-search ranking |
+| 3 | Wrap FAQ `
` to prevent CLS.
+
+9. **Add real KPI numbers to "Measurable results from day one"** — replace generic claims with cited percentages (e.g., "37% fewer support tickets — average across 12 customers, Q1 2026"). These become AIO citation bait.
+
+10. **Internal-link expansion** — add at least 3 external authority outbound links (Anthropic MCP spec, Google search docs, an Odoo / Asana case study). External links from `/` to credible sources improve E-E-A-T scoring.
+
+### Medium (within 4 weeks)
+
+11. **Ship `/llms.txt`** following the [llmstxt.org spec](https://llmstxt.org/) — points AI crawlers at canonical descriptions of every feature/integration. Low ranking impact today per Google, but a marketing signal AI buyers will check.
+
+12. **Build out `/about`, `/pricing`, `/blog`, `/docs`, `/customers`** — these 404 today. Long-tail organic search needs a content surface. `/docs` doubles as developer-acquisition channel (MCP integration documentation).
+
+13. **GSC + GA4 setup** — confirm Search Console property verified and GA4 (already wired via `G-70G60W1TDK`) configured for conversion tracking. Then rerun this audit with `/seo google setup` to get real CrUX and indexation data.
+
+14. **Per-page H2 question rewrites** — turn the FAQ H2 and "Three levels of agent action" H2 into question form (e.g., "What are the three levels of AI agent action?") to match AI Overview query patterns.
+
+15. **Customer-story pages with `Review` / `Article` schema** — each customer story should be its own URL with `Article` + `Review` JSON-LD and a real `Person` schema for the quoted customer. Maps directly to Google AIO citations.
+
+### Low (backlog)
+
+16. Add RSL 1.0 licensing terms file (`/rsl.xml`) — only relevant if Knowcap wants to monetise AI-training access. Reddit, Yahoo, Medium have adopted it; not a ranking factor.
+17. Wikipedia stub article for Knowcap (highest single-lever for ChatGPT citations per Ahrefs Dec 2025 study) — possible once you have credible secondary-source coverage to cite (TechCrunch, etc.).
+18. Add `BreadcrumbList` schema on every sub-page.
+
+---
+
+## Methodology Notes
+
+- HTML fetched via `curl -sL` and parsed with BeautifulSoup; rendering verified with Playwright (`render_page.py --mode always --viewport mobile`).
+- `robots.txt` is **Cloudflare-managed content** — the AI-blocking is a zone-level Cloudflare default ("Block AI Scrapers and Crawlers"), not something a developer wrote into the Next.js repo. Toggling it lives in the Cloudflare dashboard for the `knowcap.ai` zone. If you want the override version-controlled, ship `public/robots.txt` in the Next.js app and disable the Cloudflare managed block; Cloudflare passes the origin file through when present.
+- A/B test middleware is producing the four variants; `x-matched-path: /b` etc. in response headers confirms server-side rewrite. The `kc-landing-variant` cookie sticks the assignment for 30 days. Either canonicalise the variants or move the test to a query parameter (`?v=b`) which Google handles better via canonical hint or URL Parameter handling in GSC.
+- PageSpeed Insights field data not collected — rerun with `GOOGLE_API_KEY` set or after running `/seo google setup`. Once available, this changes Performance (CWV) section from estimated to factual.
+- Brand-mention checks: Wikipedia (not found), Reddit (API blocked unauth — manual check needed), YouTube `@knowcap` (200), X `@knowcap` (200), LinkedIn `/company/knowcap` (999 anti-bot, manual check needed).
+- The skill the user requested as "Cloudy SEO" / `AgriciDaniel/cloud-seo` does not exist under that name; the actual repo is `AgriciDaniel/claude-seo` (v2.0.0). Installed via the documented Windows path.
+
+---
+
+## Score Breakdown
+
+| Category | Weight | Score | Weighted |
+|----------|--------|-------|----------|
+| Technical SEO | 22 | 4 / 22 | 4 |
+| Content Quality | 23 | 14 / 23 | 14 |
+| On-Page SEO | 20 | 8 / 20 | 8 |
+| Schema / Structured Data | 10 | 0 / 10 | 0 |
+| Performance (CWV — estimated) | 10 | 5 / 10 | 5 |
+| AI Search Readiness | 10 | 5 / 10 | 5 |
+| Images | 5 | 2 / 5 | 2 |
+| **Total** | **100** | | **38 / 100** |
+
+After fixing the 5 critical items: projected score **72 / 100**.
+After all High + Medium fixes: projected **88 / 100**.
diff --git a/research/competitors/README.md b/research/competitors/README.md
new file mode 100644
index 0000000..cec635b
--- /dev/null
+++ b/research/competitors/README.md
@@ -0,0 +1,49 @@
+# Competitor Analysis
+
+Sales-facing competitive positioning. One folder per competitor. Each contains:
+
+- `architecture.md` — how the product actually works under the hood
+- `positioning.md` — the angle Knowcap leads with vs them, grounded in `docs/VISION.md`
+- `index.md` — folder index + source recording pointers
+
+## Real threat envelope (per [knowcap/docs/POSITIONING.md](https://github.com/Knowcap-V2/knowcap-marketing/blob/main/brand/POSITIONING.md))
+
+After the 2026-05-19 trust-layer reframe, our competitive picture is sharper. **Read.ai is not the Tier-1 threat** — Glean, Zep, and Tana are. Read.ai is the most-encountered alternative in conversations and we still need the angle for it, but our build priority on additional competitor docs should match the actual threat hierarchy.
+
+| Competitor | Tier | Why it matters |
+|---|---|---|
+| **Glean** | 1 — closest threat | $7.2B valuation; Fellow meeting integration shipped Jan 2026; one quarter away from a "verified" pill |
+| **Zep / Graphiti** | 1 — architectural twin | Open-source typed-edge memory with contradiction detection; anyone could wrap it in a meeting UX |
+| **Tana** | 1 — spiritual sibling | New Tana launched March 2026: meetings + collaborative graph + agents |
+| **Otter.ai** | 2 — pivoted in April 2026 | "Conversational Knowledge Engine" reframe pushed them into Glean's category |
+| **Read.ai** | 2 — the noisy one | Notetaker with MCP + agents; loud but not architecturally close to our thesis |
+| **Mem0 / Letta** | 3 — infra layer | Sold to developers, not end users; integration target, not competitor |
+| **Fathom / Granola / Fireflies / Tactiq** | 3 — productivity tier | Pure notetakers; commodity category; we don't fight here |
+
+## Folders
+
+| Competitor | Status | Last refreshed |
+|---|---|---|
+| [`read.ai/`](read.ai/) | ✅ Built (rewritten 2026-05-19 with trust-layer framing) | 2026-05-19 |
+| `glean/` | Queued — **highest priority** (Tier 1) | — |
+| `zep-graphiti/` | Queued — Tier 1 architectural twin | — |
+| `tana/` | Queued — Tier 1 spiritual sibling | — |
+| `otter.ai/` | Queued — Tier 2 pivoter | — |
+| `recall.ai/` | Queued — Tier 3 | — |
+| `fathom.video/` | Queued — Tier 3 | — |
+| `granola.ai/` | Queued — Tier 3 | — |
+| `tactiq.io/` | Queued — Tier 3 | — |
+
+## Source material
+
+Raw research lives in `~/Github/knowledge/llm-wiki/wiki/Knowcap/competitors/` — Apify scrapes, transcripts of competitor onboarding walkthroughs, video downloads. This folder copies only the synthesized analysis. Don't bloat marketing with research artifacts.
+
+## How to use these
+
+Before any client conversation where the prospect names a competitor:
+
+1. Open the matching `