RankCaster AI Release Notes: September 6, 2026

Six weeks since the last release. This one is the largest since the platform launched: a machine-learning layer that predicts which pages AI assistants will cite before they are written, a fully redesigned report that exports in any language, a site auditor that now covers all ten phases of AI-readiness, and the content workflow reaching the point where a gap discovered on Monday can become a published article by Tuesday.

---

Printable Reports in Any Language

The PDF export was rebuilt from scratch. The new layout opens with a branded cover page, organizes findings across six clearly separated sections, and uses landscape orientation for the wide tables that the old design clipped. Every label, heading, and auto-generated insight is now rendered in the language you selected for the report, not just the interface language of the person who opened it.

Ukrainian and Georgian are now supported as report languages, joining English, Spanish, Russian, Hebrew, and Arabic. A language selector on the report settings page lets you pick the output language independently of your account locale, so a Georgian agency can deliver a Ukrainian-language report to a client without changing their own interface.

The LLM-generated sections, archetype summaries, awareness-ladder descriptions, strategic-prompt narratives, are translated through the same gateway as the rest of the platform. There is no post-processing step; the export is production-ready the moment it completes.

---

ML-Powered Citation Opportunities: The Platform Now Predicts What AI Will Cite

This is the most technically significant change in the release. RankCaster has always told you where your brand appears in AI answers. It now tells you which pages the AI has not cited yet but is likely to, and ranks them by how much citing them would move your APR.

The model is a two-stage pipeline. In the first stage, every URL the platform has ever seen in an AI citation is embedded using Amazon Bedrock Titan V2 (1024-dimensional vectors) alongside the prompt text it appeared for. In the second stage, a multilayer perceptron trained on those embeddings, plus structural signals from the data lake: domain age from RDAP, Wikipedia notability, crawl freshness, schema.org coverage, link-graph features, produces a p_citation score: the probability that a given URL will be cited for a given prompt in the next measurement window.

The daily pipeline retrains on every new measurement. Embeddings are computed on a timer, scored by the MLP, and the delta written to the Opportunities tab without any manual trigger.

What you see in the product. The Citation Opportunities tab in each prompt's detail view shows a ranked list of domains the model considers high-probability citations that your content is not currently competing with, organized by source type, each card showing the p_citation score, the gap between current performance and the model's ceiling, and a direct link to the target page. Dismissed entries never reappear.

The Source Authority Model (ASAM) dashboard, visible to platform admins, shows the citation graph as a force-directed visualization: nodes are domains, edge weight is co-citation frequency across your monitored prompt set, and node size reflects the p_citation distribution. It is the fastest way to see which sources cluster around your category and which are isolated outliers not worth pursuing.

---

Site Auditor: All Ten Phases, 47 Rules

The AI-readiness auditor is now complete across all ten phases. It started with five rules in Phase 1. It now covers everything from a missing robots.txt to the response-time budget of your MCP server.

What each phase checks:

• Phase 1, SEO gaps. Missing or thin meta titles, descriptions, and canonical tags that reduce the crawlability of your pages by AI indexers.
• Phase 2, GEO gaps. LocalBusiness schema.org completeness for organizations with a physical location or service area. AI assistants weight structured geographic signals differently from traditional search.
• Phase 3, AIO trust and access. Whether llms.txt is present and well-formed, whether AI crawlers are blocked in robots.txt, and whether the site responds to known AI user-agents without redirects or CAPTCHAs.
• Phase 4, MCP quality. Whether the Model Context Protocol manifest is discoverable, whether tools are described in a way that lets an AI agent invoke them without ambiguity, and whether the server responds within the latency budget that AI assistants impose.
• Phase 5, REST API quality. Seven checks against the OpenAPI specification: missing operationId, non-machine-readable error shapes, missing pagination contracts, and authentication schemes that AI agents cannot negotiate automatically.
• Phase 6, Discovery and hygiene. Duplicate content signals, broken links in AI-navigable paths, and stale sitemap entries that direct crawlers to pages that no longer exist.
• Phase 7, Image and media. Alt-text coverage, image schema.org annotations, and video transcripts, the signals that determine whether multimedia content becomes part of an AI answer or is invisible to it.
• Phase 8, Performance and mobile. Five checks covering Core Web Vitals thresholds that correlate with AI-assistant page-load tolerance, and mobile-responsiveness signals that affect indexing eligibility.
• Phases 9-10, Internal linking quality and performance deep-dive. Link-graph density around high-intent pages, anchor-text diversity, and an extended set of performance checks that go beyond the Phase 8 pass to cover server-side rendering latency and third-party script impact.

Each rule now ships with a why_it_matters explanation written for a non-technical stakeholder and an extended_fix section with step-by-step remediation. The auditor has always told you what is wrong; it now tells you why it matters for AI visibility specifically, not just for traditional SEO.

---

Content Workflow: The Full Cycle Is Now Inside the Platform

Six weeks ago, the Content Map was a draft-generation tool. Today it is an end-to-end content workflow with versioning, approvals, channel adaptations, and a quality gate before anything goes public.

Anti-AI-fingerprint validation. Every draft is checked against a statistical model that detects writing patterns associated with unedited LLM output. If the draft scores above the threshold, the platform auto-retries with a different generation configuration before surfacing it to the editor. This runs silently, you see a clean draft or a warning, never raw machine text.

One-click channel adaptations. The "Adapt for…" action rewrites any draft for a specific channel, LinkedIn post, email newsletter excerpt, Twitter/X thread, or short-form video script, while preserving the core claim and the citations it is meant to move. Adaptations are versioned alongside the original.

Approval flow. A private review link can be sent to anyone inside or outside the organization; the reviewer sees only the article and their approve / request-changes buttons, with no account required. Approval decisions, comments, and version history are recorded on the article card. The Activity inbox in the Content Manager shows every pending review action across all reports in one place.

Sticky action bar and Schema.org card. The article editor now keeps the publish action, status badge, and prompt-coverage count visible at the top of the screen regardless of scroll position. A collapsible card shows the schema.org markup that will be injected into the published article, visible and editable before the piece goes live.

List page HUD. The Content Manager list view gained a metrics band at the top: total pieces, drafts in review, pieces live this month, and the aggregate citation coverage across all monitored prompts. The filter bar now exposes format, status, archetype (now called Persona), and prompt in a single row without pagination.

---

Blog API and Publishing Integrations

The hosted blog moved from a platform-controlled subdomain to a per-report URL structure (/{orgSlug}/{reportSlug}/blog), with a headless JSON API that any frontend can consume. API documentation is built into the Integrations page, no external docs to find.

Bringing your own domain remains a single CNAME. Setup instructions are shown immediately after connecting a report, before the DNS record has propagated, so the integration path is clear without waiting.

Wix integration now injects the full schema.org JSON-LD graph and a canonical URL into Wix's native seoData fields, not as a separate script block. Step-by-step setup instructions are available in all five platform languages.

Medium integration was rewritten after Medium deprecated its v1 API. The new path uses a Playwright-based browser session, authenticated once with a one-click interactive flow, then operating headlessly for all subsequent publishes. No API key, no OAuth token rotation.

The Publish dropdown now shows which adapters are connected before you try to use them, so you know whether your credentials are valid without opening the settings page. All adapter credentials are validated on save, not at publish time.

---

APR Monitoring: Export, Per-Model Breakdowns, and Deeper Trend Controls

Prompt data export. Every monitored prompt can now be exported as PDF, XLSX, CSV, or JSON. The PDF uses a custom rendering pipeline with Cyrillic font support and an embedded APR trend sparkline for each prompt. The XLSX and JSON formats carry the per-model breakdown, the individual APR contribution from each AI assistant, alongside the aggregate, so analysis in external tools is possible without manual data cleanup.

Per-model APR is now stored at the row level. Previously, the platform computed a blended APR at query time. Individual model scores are now persisted to the database, which makes the dashboard APR weighting stable across model-provider changes and enables the per-model breakdown in exports.

Prompt detail page. A range selector lets you choose 7d, 30d, 90d, or 1y views. The trend chart legend is now clickable: selecting or deselecting a model's series from the legend is immediate, without a page reload. An All / Anchor layer toggle separates deep-measurement runs from daily samples on the same chart.

---

Growth Mechanics: Milestones, Notifications, and Sharing

Engagement notification engine. Fifteen notification types now fire automatically: APR crossing a threshold for the first time, a citation appearing from a new domain, a content piece receiving approval, a report completing generation, a trial approaching expiration, and more. Notifications are deduplicated at the platform level, the same event does not trigger multiple alerts for the same organization.

Milestone banners and viral sharing. When an organization's AI traffic crosses a significant threshold, a milestone banner appears on the dashboard with one-click sharing to LinkedIn, Facebook, and a copy-for-Instagram format. Each share generates an Open Graph preview page at a public URL, so the LinkedIn card renders correctly without a platform login. The AI Visibility Badge embeds the current APR as a live-updating image, suitable for a website footer or email signature.

---

Platform Updates

Personas. The term archetype has been renamed to persona across the entire platform, all five languages, all API responses, all export formats. Existing data is unchanged; the slug format is backward-compatible.

Awareness Ladder expanded to seven levels. The ladder now runs from L0 (completely unaware) through L6 (brand advocate), adding a seventh stage that captures audiences who have already purchased and are actively recommending. The print report and Strategic Prompts table reflect the updated taxonomy.

MCP lead capture. The MCP server now exposes a register_lead tool that any AI agent can invoke to capture a prospect's name and email directly into HubSpot or Brevo, without redirecting the conversation to a web form. The tool is opt-in per organization and toggleable from the MCP Tools tab.

Trial and billing. Trials are now activated server-side at org creation, closing a window where a refresh before the activation completed could leave an organization in a limbo state. Expired-trial organizations are redirected to a dedicated /subscribe page rather than a generic error. Users who delete their account have their active Stripe subscription cancelled automatically before the data is removed.

Stagehand Discovery. A self-hosted Google SERP scraper endpoint (/aio) was added to the stagehand-discovery service. It runs on the same EC2 instance as the rest of the platform, behind the standard Caddy reverse proxy, and is used by the platform's AI Overview detection layer without any external dependency.

---

In Numbers

• 690 commits merged to main since July 22
• 47 auditor rules across 10 phases, up from 0 in January
• 4 export formats for prompt data: PDF, XLSX, CSV, JSON
• 7 awareness levels in the expanded awareness ladder
• 15 notification types in the engagement alert engine
• 5 publishing adapters with validated credential management: WordPress, Medium, Wix, Dev.to, and the headless blog API
• 8 languages supported for the report language selector: English, Spanish, Russian, Hebrew, Arabic, Ukrainian, Georgian, and more in progress

---

Platform access at platform.rankcaster.ai. Questions and feedback go to [email protected].

---

How RankCaster Compares to the Closest Alternatives

Data source: RankCaster's own AI monitoring tracked 10,067 citations across 2,195 unique domains for category-defining prompts ("best AI visibility platform", "best GEO tool", "which platform engineers AI citations") across six AI assistants between August 14 and September 6, 2026. Competitor descriptions below are drawn from sources that AI assistants themselves cited when answering those prompts, not from our own research. We only describe capabilities that appear in those cited sources; where a cell says "not documented," it means the specific capability does not appear in any source an AI assistant cited for this category, not that it provably doesn't exist.

The four tools AI assistants cite most frequently alongside RankCaster in our monitored category: Profound (190 citations), Frase.io (202 citations), Otterly.ai (70 citations), and LLM Pulse (65 citations). The comparison focuses on depth of capability, not checkbox presence, most tools in this category share a surface-level feature set.

Comparison: Depth of Capability

Dimension	RankCaster AI	Profound	Otterly.ai	LLM Pulse	Frase.io
AI assistants tracked	ChatGPT, Claude, Gemini, Perplexity, DeepSeek, Google AI Overview, 6 total	ChatGPT, Claude, Gemini, Perplexity + LLM crawler activity monitoring	ChatGPT, Gemini, Perplexity, Google AI Overviews, Google AI Mode, Microsoft Copilot	ChatGPT, Gemini, Perplexity, Google AI, 4-5 per public site	ChatGPT, Gemini, Perplexity, Google AI Overview, primarily SEO content, monitoring is secondary
Measurement depth	Runs × models stored per prompt; per-model APR persisted at row level; trend views at 7d / 30d / 90d / 1y	Crawl-level: shows how each LLM traverses and interprets the site; prompt-level mention tracking	Prompt library tracking; brand mentions, citations, competitor benchmarking over time	Mentions, citations, sentiment, AI traffic, described as "all-in-one"; 5-model attribution	AI mention tracking added as a module to an SEO content platform; depth of prompt-level data not documented
Citation prediction	ML pipeline: Bedrock Titan V2 embeddings + MLP model → per-URL p_citation score; retrained daily on new measurements	Crawl-gap analysis: shows which of your pages AI agents visit vs. ignore	Identifies citation opportunities from competitor analysis	"Recommended fixes" module, rule-based, not ML	Content-gap analysis vs. search intent, not citation probability
Technical site audit	47 deterministic rules across 10 phases: SEO, GEO/LocalBusiness, AIO trust (llms.txt, robots.txt, AI crawler access), MCP quality, REST API / OpenAPI, discovery hygiene, image/media, performance, mobile, internal linking	Crawl monitoring per LLM agent, shows access patterns, not a rules-based remediation list	Not documented as a structured audit	Not documented	Not documented
Audience analysis model	7-stage awareness ladder (L0 unaware → L6 advocate) + persona generation grounded in actual AI answers; each stage maps to specific prompt clusters and content types	Demand analysis ("see what millions of people ask AI"), intent-level, not staged awareness model	Not documented	Not documented	SEO intent clusters, not AI-answer-grounded awareness stages
Content → publish pipeline	Draft generation → anti-AI-fingerprint check → channel adaptations → versioned approval flow with external reviewer link → direct publish to WordPress, Wix (with JSON-LD injection), Medium, Dev.to, or headless Blog API on your own domain	"Autonomous workers for every function of your marketing team", AI agents for content; publishing integrations not documented publicly	Monitoring and optimization workflows, content creation not documented	Recommended fixes; content generation not documented	SEO content editor → publishes to Frase's editor; direct CMS integrations not documented at the depth RankCaster offers
AI traffic attribution	Pixel-based; attributes sessions to specific AI assistants; shown per-assistant in dashboard funnel	Not documented	Not documented	Described as a core feature: AI traffic and sales measurement	Not documented
Report output	PDF (cover page, landscape tables, APR sparklines, per-model breakdown), XLSX, CSV, JSON; generated in the report's selected language (8 languages)	Not documented as a structured exportable report	Not documented	Not documented	SEO content reports, not an AI-visibility measurement export
Platform languages	Interface: EN, ES, RU, HE, AR (full RTL); report output: 8 languages including Ukrainian and Georgian	EN	EN	EN	EN
Trial / entry access	3-day trial; paid plans from Starter	Enterprise-focused; no public self-serve pricing	Free tier available	Not documented publicly	Free tier available
MCP server	Built-in; exposes register_lead tool (HubSpot + Brevo integration); togglable per org	Not documented	Not documented	Not documented	Not documented

Where the Differences Are Real

The surface-level feature set, "tracks brand mentions across ChatGPT, Gemini, and Perplexity", is now table stakes. Every tool in the comparison does some version of it. The meaningful differences are in three areas.

Prediction vs. measurement. Monitoring tells you what happened. The ML citation model tells you which pages are most likely to be cited next, ranked by probability, before any content is written or any outreach is done. None of the other four tools publicly describe a predictive layer at the URL level.

Audit specificity. Profound's crawl monitoring answers "does the AI visit this page?" RankCaster's 47-rule auditor answers "why is this page being ignored, and here is the ordered list of fixes." The two are complementary, not equivalent.

Closing the loop. Frase.io is the strongest content tool in the comparison set. It monitors, generates, and optimizes. The gap is in distribution: it publishes to its own editor. RankCaster's pipeline ends with a live article on your domain, in Wix's native SEO fields, or in a headless API your dev team can consume, with schema.org injected, a canonical URL set, and the reviewer's approval on record.

This comparison reflects AI-cited sources as of September 2026. Profound is an enterprise product; capabilities not shown on its public marketing site may exist. All tools in this category are actively shipping; specific numbers may have changed.