SEO Platform — Bulk Audits, Rank Tracking & LLM Visibility

Project Info

Client Self-funded SaaS
Year 2026 → present
Category SaaS / SEO platform
Status Closed beta — used on my own projects, no public landing page yet, so there is no live link to share
Stack Next.js 16 (App Router), React 19, TypeScript, Tailwind v4, Prisma 7 + Postgres, NextAuth v5, next-intl, PM2 (two-instance zero-downtime deploy)
Integrations Google Search Console API, Google Indexing API, Ahrefs / Majestic / Moz, DataForSEO, SE Ranking, WayForPay, Brevo, Cloudflare Turnstile, Slack
AI OpenAI + Google Gemini + Anthropic Claude — model picked per user in settings

Project Description

A self-funded SEO platform that started as a bulk domain checker — paste up to a hundred domains, tick the checks you need, get one sortable table with WHOIS, DNS, CMS, backlink and traffic metrics — and grew into a full combine: Search Console analytics, rank tracking, AI-visibility monitoring, link and PBN control, technical audits and an AI agent on top of it all. Built for everyday use — running client sites and a link network.

Modules

Module	What it does
Bulk domain analysis	Up to 100 domains per batch, 8 data sources in one table WHOIS (registrar, created / expires), DNS, CMS, Wayback archive Majestic TF/CF, Ahrefs DR/UR/backlinks/ref. domains, Moz DA + spam score, traffic Async jobs with status polling, sortable results, CSV and XLSX export Per-plan quotas and a check history inside every project
Search Console	OAuth-connected properties, each user on their own credentials 16-month backfill plus daily snapshots into our own database — history keeps growing after Google’s window has rolled away Handles Google’s ~3-day data lag and catches up missed dates (up to 30 per run) Striking distance: queries knocking on the top-10 with an estimate of the clicks left on the table Opportunities: queries whose real CTR is under half of what the position should give (positional CTR baselines from the Advanced Web Ranking 2024 study) Content gap: queries at positions 21–100 with 100+ impressions — Google already shows you for the topic, but there is no page for it Cannibalization: where several pages fight over the same query AI spec engine: any report row becomes a ready technical spec for the copywriter or the SEO, with the live page metadata pulled in For cannibalization the prompt carries a 6-scenario playbook (merge, split the intent, de-optimize the secondaries, canonical, internal links, strengthen the main page) — the model lays the options out, the copywriter or the SEO picks Head- and secondary-keyword guards — a spec never hands over a keyword another page already ranks for Query intent classification, drop alerts, an email digest and report export
GEO — LLM visibility	Checks how often a brand is mentioned by OpenAI, Gemini and Google AI Overview AI prompt generator: keywords in — the natural questions real people ask an assistant out Branded and unbranded prompt sets, up to 5 variants per keyword, preview before saving Deduplication against existing prompts, up to 50 prompts per project Two-pass detection that cuts false positives Share of voice against up to 20 competitors per project AI Overview detail view and a cited-domains report Automatic re-runs by cron
Rank tracker	Keyword positions per project Keywords in folders with drag-and-drop ordering — something SE Ranking does not even offer Position history and per-folder aggregates Refresh on demand or on a cron schedule
Indexing	Submissions through the Google Indexing API and external indexing services Queue plus a scheduler for submissions Indexed / not-indexed status checks Submission history with deletion, visible to the whole team
Competitors & keyword gap	Competitor registry per project Keyword gap: what they rank for and you do not Regex filters and query clustering Select → export → spec in one pass
Sitemap scanner	Recurring crawl of competitor sitemaps Diffs: which URLs appeared since the last run New URLs pushed to Slack in chunks Run history per site
Keyword research	Ideas, suggestions and trends via DataForSEO / SE Ranking Minimum-volume filter, up to 200 ideas per query 7-day local cache so API quota is not burned twice Export of the selected keywords
Links control	Monitoring of purchased placements: link present, anchor, dofollow / nofollow Daily re-checks (at most one per link per day) through a proxy pool Records the exact date a link disappeared Dispute / “insurance” workflow with the vendor
PBN management	Registry for hundreds of network domains (WordPress / Next.js) WHOIS/RDAP expiry watch — warning at 30 days, alarm at 7 Per-domain traffic dashboards fed by an own hit-ingest endpoint One-click WordPress login Availability checks per domain
Sitemap health	Crawls up to 10,000 sitemap URLs per run Status codes, redirect chains (up to 5 hops), head metadata Polite concurrency and a wall-clock budget — a partial report instead of a failed run Weekly enqueue of all sites plus a Slack digest
Schema checker	Extracts and validates schema.org markup on a page Splits findings into errors and warnings per entity type Tells you what exactly to fix Result export
Log audit	Upload of server logs (CSV/TSV/XLSX/ZIP, up to 10 files / 50 MB) Who crawls the site and how: bots, status codes, top-20 pages Crawl freshness in buckets: 0–1, 1–3, 3–7, 7–14, 14–31, 30+ days — surfaces pages Google has not seen for months Crawl-budget waste broken down: re-fetched static, redirect chains, 404s, junk endpoints (wc-ajax, admin-ajax, add-to-cart, xmlrpc) And, most importantly, what to do about it — the report ends with prioritized P0/P1/P2 findings, not just numbers: set cache headers, disallow service endpoints in robots.txt, clean up 404s, deal with the bot eating half the budget Every finding carries an expected effect (“frees up ~N hits of crawl budget”) plus a closing “expected effect” summary for the P0–P1 block Configurable rules and thresholds, with a WordPress/WooCommerce ruleset out of the box: re-fetched-static %, 404 hits per URL, top-bot share, bot error rate, redirect hits Findings are stored as keys, not sentences — the report reads in whatever language the reader picked
Text checker	AI-origin and originality check for any text Script and encoding check (mixed Cyrillic / Latin and friends) Check history with the cost of every run Runs on the user's own API key
Uptime monitor	HTTP checks over projects and network domains Intervals of 1, 3, 6, 12 or 24 hours 5 attempts with back-off — no false “it is down” Alerting and incident history
Teams & access	NextAuth v5: email with verification, Turnstile against bots Roles and project sharing inside a team Per-user API keys AI model selectable in settings, admin panel with user management
AI agent — the layer everything is heading towards	A chat on top of your own project data, not generic advice — the model reaches for the platform’s own tools mid-answer Today it can already: list projects, pull a rank overview and per-folder aggregates, read the GEO picture (visibility, competitor visibility, cited domains), open the competitor registry and the keyword gap Every answer shows which tools it called, so the numbers are traceable rather than hallucinated Conversations are stored with full history and run on the user’s own API key Where it is going: from reading to doing — the agent should run almost the whole project on its own Planned: launch checks and crawls, add domains, keywords and GEO prompts, write the technical spec for a page and hand it over, push URLs for indexing, open a dispute over a lost link, assemble a client report — all from a single conversation The end state: you describe the goal, the agent decides which module to use and does the work; you review the result

Architecture

Next.js 16 App Router with server actions that never throw — they return { data }, { error } or { success } — Prisma 7 on Postgres, ownership checks on every resource, and admin-tunable settings that fall back to env vars when unset. Long jobs follow one pattern: enqueue, poll state, fetch the result file; the recurring side lives in /api/cron/* endpoints (scheduler, GEO, indexing, rank tracker, uptime, sitemap health, links control, subscriptions). The whole UI is localized through next-intl. Deploys are zero-downtime — two PM2 instances swap behind nginx so a running crawl never gets cut off.