Search
Autonomous SEO Strategist
Point it at any domain. Get a full competitive SEO audit in minutes.
An 8-phase pipeline that maps your site, pulls 30 live Google search rankings, identifies every competitor stealing your traffic, and uses Claude to generate a prioritized action plan. Built with Nimble and Claude.
Inputs
- Target domain The domain to audit, entered at run time (e.g., stripe.com). No account or login required — just a public domain.
Outputs
What you get after a full run.
- Composite SEO score (0–100)
- 30 tracked search terms with live rankings
- Competitor threat map (up to 20 competitors)
- Quick wins ranked by impact-to-effort
- Page optimization briefs
- New pages to create with full content briefs
- Strategic themes & positioning recommendations
- Full markdown report
How it works
An 8-phase pipeline. Read the blog here for a deeper explanation.
- Understand Fetches the homepage as clean markdown to understand the site before crawling. This step gives Claude the context needed to make intelligent decisions in the next phase — what the company does, what its main products or services are, and how it structures its public-facing content. Without this step, page selection and term generation would be blind to the site’s actual positioning.
- Map & Select Discovers all site URLs, classifies by page type, Claude selects the most SEO-relevant pages. The map step returns every crawlable URL on the domain grouped by type — blog posts, product pages, landing pages, docs, and so on. Claude selects up to 50 pages based on SEO value, prioritizing pages that are likely to rank, likely to have keyword-rich content, or likely to reveal competitive gaps.
- Extract Batch-extracts selected pages in parallel. All selected pages are fetched simultaneously via Nimble’s batch extract endpoint, so the full content index is built in roughly the time it takes to extract one page. Each page is returned as clean markdown — images, navigation, and boilerplate stripped — so Claude is working with the actual text content of each page.
- Generate Terms Claude generates 30 targeted search queries across 6 intent categories. The 6 intent categories are: branded, core product, comparison, use case, pain point, and competitor-adjacent. Spreading terms across intent types ensures the SERP analysis surfaces the full competitive landscape rather than just direct brand overlap.
- SERP Data All 30 queries fired as live Google searches via Nimble. Each search returns the top 10 organic results with URL, title, and snippet. Queries are fired concurrently, so 30 live SERP requests complete in the time it takes to run one. The raw SERP data — including competitor URLs and ranking positions — feeds both the competitor mapping and the diagnosis phases.
- Competitor Map Claude maps every competitor and assigns threat levels. Competitors are identified from recurring domains across the 30 SERPs — a domain appearing in results for 5 or more queries is flagged as a direct competitor. Threat level reflects both how many target queries the competitor ranks for and how high they appear — a domain that ranks in the top 3 for 10 queries is a higher threat than one that appears at position 8 across 15 queries.
- Diagnose Claude scores each search term and identifies root issues. Scoring distinguishes between three failure modes: not ranking at all, ranking low due to thin content, and ranking for the wrong page type. The diagnosis prompt asks Claude to identify the most likely root cause per term given the page content it has already indexed — so the diagnosis is specific to this site, not generic SEO advice.
- Recommendations Claude generates quick wins, page briefs, new pages, and themes. Quick wins are prioritized by impact-to-effort ratio — improvements to existing pages that already have some ranking signal come first. Page briefs and new page recommendations are written as actionable content specifications, not vague suggestions, including target keyword, recommended structure, and key points to cover.
Stack
Nimble primitives plus the full runtime stack.
APIS & AGENTS
What it does
- extract Fetches the target domain’s homepage as clean markdown — used to understand the site before mapping.
- map Discovers all URLs on the target domain and classifies them by page type.
- extract_batch Extracts up to 50 selected pages in parallel — powers the full site content index.
- search Fires 30 live Google queries and returns ranked SERP results for each search term.
RUNTIME STACK
Role
- claude-sonnet-4-6 Anthropic Claude API — powers page selection, term generation, competitor mapping, diagnosis, and the recommendation phases.
- streamlit Interactive dashboard for browsing the SEO audit results across 8 tabs.
- python 3.9+ Pipeline orchestration and data processing. Async I/O for parallel extraction and search.
- MIT license Fork, modify, ship — no restrictions.
Reach out if you have any questions.
Talk to an Expert