Documentation

Redirecta is an AI-powered tool that automatically matches source URLs to destination URLs during site migrations, CMS transitions, or domain changes.

Create an Account

Visit the Dashboard and register with your email address. No credit card required.

Your First Redirect Mapping Job

1. Prepare your CSV. You need at minimum two columns: source (your old URLs) and destination (your new URLs). You can include optional metadata columns to improve matching accuracy.

2. Upload your file. Drag and drop or click to select your CSV file. The tool will detect your columns and show a preview.

3. Configure settings. Choose your matching model, set the number of alternatives (Top-K), and review the detected columns.

4. Start matching. Click “Start Matching” and the engine will process your URLs. For large datasets this may take a few minutes.

5. Review results. Browse your matches sorted by confidence score. Filter by match type, sort by any column, and export the final mapping as CSV.

Required Columns

Your CSV must include these two columns:

Optional Metadata Columns

Providing metadata significantly improves matching accuracy. The engine uses these for content-level similarity scoring:

Accepted Aliases

The tool recognizes common variations of column names:

• source_url, old_url, from_url, old are treated as source
• dest_url, new_url, target, dest, new are treated as destination
• dest_title, title_source, title_dest are treated as title columns
• dest_h1, h1_source, h1_dest are treated as H1 columns
• source_meta_desc, dest_meta_desc, source_meta, dest_meta are treated as meta description columns

Tips for Clean Data

• Normalize trailing slashes. Be consistent with whether URLs end in / or not.
• Use absolute URLs. Include the full https:// prefix for best accuracy.
• Duplicates are removed automatically. We collapse only exact duplicates and tracking-parameter variants of the same address (so /foo and /foo?utm_source=newsletter become one row, since a single redirect covers both). Genuinely distinct resources are kept so each gets its own redirect: a different scheme (http vs https), host (example.com vs www.example.com), trailing slash (/foo vs /foo/), language path (/en/foo vs /de/foo), or file extension. You do not need to pre-dedupe.
• UTF-8 encoding. Save your CSV with UTF-8 encoding to support international characters.
• No empty rows. Remove any blank rows from your spreadsheet before exporting.

Redirecta uses a multi-signal scoring system that combines structural, semantic, and content-based analysis to find the best destination URL for each source.

How It Works

1. Candidate Generation. For each source URL, the engine identifies a pool of potential destination matches using URL type classification and fast similarity heuristics.

2. Multi-Signal Scoring. Each source-destination pair is scored across 5 composite signal groups (see Signals & Scoring for details). Each group combines multiple sub-signals that capture different aspects of URL similarity: structure, text, token overlap, content, and identity.

3. Weight-Based Aggregation. Group scores are combined using carefully tuned weights. Groups that are more discriminative (like token matching and content) receive higher weight. If a group has no data (for example, no page metadata), its weight is redistributed to the active groups.

4. Ranking and Classification. Candidates are ranked by aggregate score. Each match is classified into one of nine tiers: exact, cross-domain exact, high confidence, medium confidence, low confidence, review, parent fallback, no match, or no redirect needed. See Results & Export for the meaning of each tier.

Two-Phase Candidate Filtering

For large datasets, the engine uses a two-phase approach:

• Phase 1: URL Type Priority. URLs are classified by type (product pages, category pages, blog posts, etc.) and candidates of the same type are prioritized.
• Phase 2: Fine-Grained Scoring. The top candidates from Phase 1 are scored across all 5 signal groups for precise ranking.

This approach keeps matching fast even with thousands of URLs while maintaining accuracy.

Multilingual Support

The engine uses the paraphrase-multilingual-MiniLM-L12-v2 model for semantic embeddings, supporting 50+ languages. Combined with 12 multilingual entity dictionaries covering 12+ languages, the tool handles non-English URLs and content natively.

The matching engine evaluates each source-destination pair across 5 composite signal groups. Each group combines multiple sub-signals into a single score, covering a different dimension of URL similarity.

1. URL Structure

Measures whether the URLs are structurally compatible.

2. URL Text

Measures how similar the URL strings are in meaning and character composition.

3. Token Match

Measures overlap of meaningful tokens extracted from URLs and metadata.

4. Content

Measures similarity of page content when metadata is provided.

5. Identity

Matches on hard identifiers that strongly indicate the same entity.

Why Metadata Improves Accuracy

Without metadata columns, the engine relies on URL-based and semantic signals only. Adding titles, H1s, and meta descriptions unlocks the Content signal group. In testing, providing full metadata typically improves match accuracy by 15-25% on content-heavy sites.

Before starting a matching job, you can configure how the engine processes your URLs.

Matching Mode

Choose which embedding model the engine uses for semantic similarity:

The model determines how URL meaning is interpreted. Multilingual mode is the default and works well for both English and non-English sites.

Thoroughness (Candidate Depth)

Controls how many candidate destination URLs are evaluated for each source URL. Adjustable from 10 to 100 via a slider:

• Low (10-30): faster processing. Evaluates fewer candidates per source URL. Good for simple migrations where URLs are structurally similar.
• Balanced (40-60) (default: 50): recommended for most cases. Balances speed and accuracy.
• High (70-100): evaluates more candidates per source URL. Use for complex migrations with large destination lists or ambiguous URL structures.

Higher values increase processing time but improve accuracy, especially when the best match is not among the most obvious candidates.

When to Use Which Configuration

• Simple domain migration (same structure, different domain): Fast mode, low thoroughness
• CMS migration (URLs change structure): Multilingual mode, balanced thoroughness
• Full site redesign (new IA, new URL patterns): Multilingual or Accurate mode, high thoroughness
• Multilingual site with translated slugs: Multilingual mode, balanced thoroughness

After matching completes, the results page shows all source-destination pairs with their confidence scores and match details.

Understanding the Results Table

Confidence Tiers

Filtering & Sorting

• Filter by match type: focus on low-confidence matches that need review.
• Sort by confidence: find the weakest matches quickly.
• Sort by source URL: group related pages together for batch review.

CSV Export

Three downloads are available from the results page:

• Summary CSV: source URL, destination URL, confidence score, and match type. No Redirect Needed rows are excluded by default to prevent deploying self-redirect loops.
• Detailed CSV: all of the above plus individual signal scores and alternative destinations (if thoroughness was set above minimum).
• Gap Report CSV: two-section report. Section 1 lists source URLs with no confident match (each annotated with the closest-but-too-low candidate the engine considered). Section 2 lists destination URLs that no source URL maps to (new content with no inbound redirect path).

The Summary and Detailed CSVs are ready to import into your server configuration, .htaccess file, or redirect management tool. The Gap Report is for manual review during migration planning.

A full developer API is on our roadmap, allowing you to integrate Redirecta’s matching engine directly into your own tools, platforms, and workflows.

Planned Features

• REST API with JSON and CSV input/output
• API key authentication with per-key rate limits
• Webhook callbacks for async job completion notifications
• Synchronous mode for small datasets (instant results)
• Usage tracking and quota management

Stay Updated

The developer API is coming soon. In the meantime, you can use the web interface for all matching jobs. Check the Releases page for updates on API development progress.