What RivalNote watches and how it scores it
The sources it reads, what an insight is, how the 0 to 100 score is set, and where a scan stops.
Where it looks
- The competitor's site
- The home page, then the root-linked pages ranked as likely pricing, changelog or careers pages, plus feed entries and one nested sitemap.
- Feeds and sitemaps
- Only those the site advertises, through a
link rel="alternate"tag or aSitemap:line in robots.txt. - Public job boards
- Boards linked from the competitor's pages on common applicant-tracking platforms, such as Ashby, Greenhouse, Lever, Workable, Personio, Recruitee, SmartRecruiters and Teamtailor.
- Other linked public pages
- Release notes, marketplaces, analyst and award pages, status and trust pages, documentation, package registries and review pages, when the competitor's site links to them.
- Sources you add
- Any public URL, such as a pricing page or an RSS feed, added to a competitor by hand.
What it writes
From the difference between two snapshots, Claude writes between 0 and 5 insights per changed page. Each is one of these categories:
- pricing-change
- product-launch
- hiring-signal
- content-shift
- leadership-change
- partnership-signal
- trust-signal
- customer-signal
- marketing-campaign
- financial-signal
- other
Every insight carries a headline, a body, evidence from the changed lines, a recommended action, a confidence score, a direction (threat, opportunity or neutral) and an owner team. If nothing meaningful changed, it writes nothing. Footer, navigation, cookie and legal edits are skipped unless they change pricing, product, hiring, security, leadership, partnerships, customer proof or positioning.
How the score works
A second pass scores each insight from 0 to 100 for a B2B SaaS product manager who has to react.
| Score | What lands there |
|---|---|
| 80 to 100 | Pricing changes, major product launches, leadership changes, hiring spikes in core roles |
| 50 to 79 | Meaningful content or messaging shifts, secondary feature launches |
| 20 to 49 | Minor copy edits, blog posts, content additions |
| 0 to 19 | Cosmetic noise, broken links, irrelevant updates |
Your watch terms raise the score of matching insights, muted categories remove their category, and a competitor's relationship and priority change how urgent a material insight looks. Slack alerts fire at or above your threshold, 70 by default.
Where a scan stops
- A scan makes at most 8 requests to the competitor's own domain, the home page and robots.txt included, and 3 to every other host in total. A failed request still uses its slot.
- Requests to the same domain are at least 10 seconds apart.
- The first scan of a competitor only records a baseline, so there is nothing to report yet.
- A page the scan did not reach is carried over from the last snapshot. A section is reported as removed only on evidence: a 404 or 410, or a permanent redirect to the home page.
- It reads public pages and never signs in to a competitor's site.
- A model can misread a change. Check the evidence line before you act on an insight.
Talk to the founder
Early access starts with a WhatsApp conversation. Say which competitors you track.
Talk to the founder on WhatsApp