LLMS Amplifier Changelog

A WordPress plugin for generating and managing llms.txt and llms-full.txt files to help AI language models better understand and interact with your website content.

Author: Brian Winum Website: brianwinum.com Current Version: 2.5.5

Version 2.5.5 — Data Safety, Portability & Logging Controls

Released June 2026

A reliability and operations release. Where the 2.5.x line so far has been about what you tell AI systems, 2.5.5 hardens how the plugin behaves on real sites — protecting user-uploaded data on deactivation, making the full v2.5 configuration portable between sites, running cleanly on minimal PHP builds, and giving site owners real control over the security log (which on busy sites had been growing unchecked).

Data Safety: Supplemental Files Preserved on Deactivation

Deactivating the plugin no longer deletes user-uploaded supplemental files. They are not regenerable, so deleting them on a simple deactivate/reactivate caused permanent data loss
Physical removal now happens only on uninstall, and only behind the existing opt-in delete_data_on_uninstall setting
Deactivation now performs only lightweight, fully reversible housekeeping (transient flush and rewrite-rule flush)

Full-Fidelity Export / Import (Backup Format v2)

The manual export/import bundle now includes all v2.5 data layers: JSON index settings, offers, corrections, agent guidance, disambiguation, brand voice, and proof points
Previously these were silently dropped from manual exports, so migrating a site between domains lost everything built in the v2.5 layers
Each layer is restored through its existing field-level sanitizer on import; older v1 backups still import cleanly

Graceful Operation Without mbstring

The UTF-8 engine now degrades gracefully when PHP's mbstring extension is unavailable, using PCRE/iconv fallbacks for encoding validation, conversion, and substring operations
A dismissible admin notice recommends enabling mbstring for the most reliable international-character handling, but the plugin no longer risks a fatal error without it

Double-Encoding Repair Map Cleanup

Corrected the character-repair map used to fix double-encoded (mojibake) content
Removed truncated and duplicate entries in the uppercase-accented Latin, Polish/Czech, and smart-punctuation sections that could match a bare lead byte and overwrite otherwise-valid characters

Security Log Controls

Retention window (days) — keep entries for a configurable number of days (in addition to the existing row cap), or set 0 for no time limit
Cap mode — choose lenient (preserve recent history) or strict (a hard ceiling that never exceeds the row cap, for predictable table size)
Logging level — All, Critical only (threats and errors), or Off. Rate-limiting and abuse/threat events are always recorded regardless of this setting, so security protections are never weakened
Manual maintenance — new “Trim now” and “Clear log” buttons with live row-count and table-size display, so a database tool is no longer needed to reduce a large log

Fixed: Runaway Security-Log Growth

The GitHub update checker was initializing — and writing a log row — on every front-end and bot request. It now runs only in admin and cron contexts, where update checks actually happen
The per-page-load meta-tag-injection log write (pure telemetry of normal operation) was removed
Together these two events were the dominant source of log growth on active sites
A one-time cleanup on upgrade purges the historical noise rows automatically — no manual database work required

Under the Hood

Security-log cleanup moved from running on every logged event to a once-daily scheduled task, removing a per-request database operation that was the real performance drag on busy sites (with a lightweight safety net for hosts that disable WP-Cron)

Version 2.5.4 — Page Builder Compatibility Hardening

Released June 2026

A hardening release that makes the v2.5.3 page-builder support accurate and dependable on real-world sites, where builder content, caching, and imported layouts complicate text extraction.

More Accurate Detection

Builder detection decoupled from the content-extraction path, so the compatibility report is accurate regardless of your content or SEO settings
Active-builder detection via class/function checks, not just post meta
New standalone “Scan Now” admin action runs detection only, with no file regeneration

Better Extraction

Elementor extraction uses the official display API with a renderer-authoritative guard against stale data on imported or reverted pages
WPBakery and Divi shortcode-attribute extraction, including base64 vc_raw_html decoding
Bricks legacy key, Oxygen prefixed legacy key, and Beaver Builder detection without relying on an enabled flag

Smarter Caching

The per-post extraction cache signature now hashes the builder source data, so it invalidates correctly on meta edits that don't bump the post-modified timestamp

Clearer Reporting

Admin panel now reports active builder plugins, per-builder extraction methods, per-page debug reasons, and a context-aware empty-state message

Version 2.5.3 — Page Builder Compatibility

Released June 2026

This release teaches LLMS Amplifier to read content built with visual page builders, which store their content as structured data or shortcodes rather than plain post content — meaning builder-built pages previously contributed little or nothing to your LLMS files.

Page Builder Content Extraction

New LLMS_Amplifier_Builder_Content engine extracts readable text from seven major builders: Elementor, Bricks, Oxygen, Breakdance, Beaver Builder, Divi, and WPBakery
Three-tier strategy per post: render-API extraction (primary), recursive data-structure / shortcode harvest (fallback), and title/metadata-only (last resort)

Performance

Per-post extraction cache keyed on the post's modified time — unchanged builder pages are not re-rendered on regeneration
Cache is durable across LiteSpeed / Redis object caches

Admin Visibility

New “Page Builder Compatibility” panel (Settings → Integration Settings) reporting detected builders, the extraction-method breakdown, and any pages that yielded no body text

Version 2.5.2 — Guardrails Layer for AI Hallucination Control

Released June 2026

This release introduces a complete control layer for how AI assistants represent your business. Where earlier versions focused on telling AI systems what your site is, v2.5.2 gives you explicit control over what to say, what NOT to say, and how to be quoted — directly addressing the hallucination and misrepresentation problems site owners face when AI assistants describe them inaccurately.

The Five Guardrails Sections

Disambiguation — Explicit “we are NOT X” statements that separate you from confusable entities (people, companies, brands with similar names). Up to 10 entries.
Corrections — The ARP “Pink Elephant Fix” pattern: pair correct facts with the incorrect claims AI assistants may have learned about you. Up to 20 entries.
Agent Guidance — Explicit instructions for how AI should represent your business in different scenarios, with optional “do not” rules. Up to 10 entries.
Brand Voice — Tone, voice attributes, preferred terms, terms to avoid, and example phrasing constraints in a single object.
Proof Points — Verifiable claims AI can cite with confidence: certifications, awards, dated metrics, sourced statistics. Up to 20 entries, with ISO 8601 date validation.

Dual Output

All five sections appear as a new ## Guardrails for AI Assistants Markdown section in both llms.txt and llms-full.txt, positioned before the Offers section so corrections precede promotional content
Same data is rendered as structured arrays/object in llms-index.json, positioned between permissions and offers
Empty sections are silently skipped — the Guardrails section appears only if you've configured at least one entry

Admin Experience

New “AI Guardrails” admin section with a collapsible repeater UI for each list-style category
Generic repeater pattern with expand/collapse, live title preview, and automatic reindexing
Five new sanitize methods with controlled-vocabulary validation and per-section entry caps

Developer Extensibility

Per-section filters: llms_amplifier_index_disambiguation, _corrections, _agent_guidance, _brand_voice, _proof_points
Per-item filters for fine-grained customization
All five new options (llms_amplifier_corrections, _agent_guidance, _disambiguation, _brand_voice, _proof_points) are included in export/import bundles, pre-uninstall backups, and the data-deletion routine

Version 2.5.1 — Agent-Addressed Offers

Released June 2026

A new way to surface incentives directly to AI assistants helping users evaluate vendors. While traditional offers target human visitors browsing your site, agent-addressed offers are written specifically for the AI in the middle of a research conversation — giving you a way to incentivize AI recommendation alongside traditional marketing.

Agent-Addressed Offers

Configure up to 10 offers: discounts, free trials, consultations, audits, pilots, referral credits, bonuses, and more
Each offer renders as both a Markdown ## Offers section in llms.txt and a structured array in llms-index.json
Schema.org Offer sub-object included with each entry for broader crawler compatibility

Audience Targeting

Mark offers as general, agent-addressed (writing directly to AI assistants), new customer, existing customer, or referral
AI assistants can prioritize the right offer for the user's context

Tracking & Attribution

Claim URLs (designed for UTM parameters) for measuring AI-referred conversions
Claim codes for offers that don't use URLs
Either claim URL or claim code is required for an offer to be valid

Validity Windows

Optional valid_from and valid_until dates
Expired offers are automatically filtered from output — no manual cleanup required

Developer Extensibility

llms_amplifier_index_offers filter for the full array
llms_amplifier_index_offer filter for per-offer customization

Version 2.5.0 — Structured JSON Index & AI Discoverability Standards

Released May 2026

A major release that adds a complete machine-readable layer alongside the existing Markdown files. Where llms.txt is human-readable Markdown that AI systems can parse, llms-index.json is structured JSON that AI systems can consume directly without parsing — and the supporting protocol layer (vendor MIME types, link relations, .well-known paths) brings the plugin into alignment with emerging AI discoverability standards.

Structured JSON Index (llms-index.json)

Auto-generated machine-readable JSON index of your site
Sections: site identity, summary, pages, contact, products, people, FAQ, permissions, and a files manifest
SHA-256 content hashes on every listed file for integrity verification
Served at both /llms-index.json and /.well-known/llms-index.json
Developer filters for every section (llms_amplifier_index_data, llms_amplifier_index_pages, etc.)

Vendor MIME Types

text/vnd.llms-amplifier.manifest+markdown for llms.txt
text/vnd.llms-amplifier.full+markdown for llms-full.txt
application/vnd.llms-amplifier.index+json for llms-index.json
Plus supplemental file variants
Sets the plugin apart from generic markdown files and signals to crawlers that the content is purpose-built for AI consumption

HTTP Link Headers

rel="canonical", rel="describedby", and rel="license" headers on every served file
Allows AI crawlers to follow protocol-standard links before parsing file contents

IANA Link Relations

HTML meta tags now use officially registered IANA link relations: rel="alternate contents", rel="describedby index", rel="license"
Better compatibility with standards-compliant crawlers

.well-known Dual-Path Serving

/.well-known/llms.txt, /.well-known/llms-full.txt, and /.well-known/llms-index.json all served correctly
Aligns with the IETF .well-known convention used by robots.txt, security.txt, and other discovery standards

robots.txt Integration

New LLMS-Index: directive added when the JSON index is enabled
Sitemaps continue to be referenced

Admin UI: AI Index Settings

New “AI Index” settings section in the admin dashboard
Configure business identity, location, social links, summary, contact points, and Schema.org type mappings per post type
Drives the content of llms-index.json site identity section

AI Crawl Indexing Toggle

Optional setting (default OFF) to remove the X-Robots-Tag: noindex header from llms.txt, llms-full.txt, and supplemental files
Allows search engines to index these files alongside AI crawlers
Sitemaps are unaffected by the toggle

Version 2.4.1 — Data Safety, Recovery & Critical Fixes

Released May 2026

This release combined major new safety features with several critical bug fixes — including one that was silently corrupting custom header and footer content for any site that had used markdown special characters.

Critical Fix: Markdown Corruption in Custom Header/Footer Content

Custom header and footer content (which is markdown, not HTML) was being passed through WordPress's wp_kses_post() sanitizer, which HTML-entity-encoded every markdown special character it didn't recognize as valid HTML
Visible corruption included: apostrophes appearing as \', quotes as \", markdown specials backslash-escaped (\#, \*\*, \[, \]), ampersands encoded as & (sometimes double-encoded), em-dashes flattened to ---, en-dashes flattened to --, and bullet lists flattened into single paragraphs
New sanitize_custom_markdown() method replaces wp_kses_post() for all custom header/footer save paths, stripping genuinely dangerous patterns (script tags, iframes, event handlers, dangerous URI schemes, PHP tags) while leaving markdown intact
Automatic one-time repair migration runs on upgrade: html_entity_decode() plus reversal of backslash-escaping, iterating up to 3 passes to handle content corrupted multiple times. A single scheduled regeneration is triggered so repaired content lands in llms.txt within 60 seconds of upgrade — no manual click required

Fix: Stale Timestamp Display in Admin

Displayed timestamp now derives from the physical llms.txt file's modification time (filemtime()), immune to object-cache staleness
New get_last_generated_time() and get_last_generated_human() helpers in the core class
Write path hardened with fresh re-read and explicit object-cache deletes

Fix: False “Mismatch” in File-Serving Diagnostics

The Diagnose File Serving and Force Clean Regeneration tools were reporting every file as “✗ mismatched” even when content was being served correctly
Root cause: the plugin writes files with a UTF-8 BOM but strips the BOM when serving them, so byte comparisons were off by exactly 3 bytes
Verification now strips the BOM from both the local file and served body before comparing, plus a trailing-whitespace tolerance for proxy-introduced newlines

Uninstall Data Protection

Settings, custom content, and supplemental files now preserved by default on plugin uninstall
Data deletion on uninstall is strictly opt-in via a new setting (defaults to OFF)
Existing installations keep all their data on upgrade
When deletion is enabled, an automatic backup is written to /wp-content/uploads/llms-amplifier-backups/ before anything is removed
Backup directory protected with .htaccess (deny all) and an index file

Settings Export / Import

Export complete configuration as a portable JSON bundle
Includes all settings, custom content, and supplemental file contents (base64-encoded)
Import validates structure, sanitizes all fields, rejects executable content, and enforces size limits

Backup Manager

Admin interface to list pre-uninstall backups with filename, size, and timestamp
Restore configuration from any prior backup
Strict filename validation and path boundary checks on deletion

Force Clean Regeneration

Audit tool detects orphan llms.txt / llms-full.txt files at the web root that can shadow plugin-served content
Two-step audit-then-destroy flow: review what will be removed before taking action
Verifies served files after regeneration and clears third-party caches

Other Improvements

Supplemental file size limits raised: 1 MB per file (up from 500 KB) and 10 MB total (up from 5 MB)
WooCommerce PHP 8+ compatibility fix using isset() and explicit casting for safer property access

Version 2.4.0 — Supplemental AI File Uploads

Released March 2026

A major feature release that lets you upload supporting documents directly from the dashboard — no FTP or file manager required. These supplemental files help AI systems understand your brand beyond what's in your main llms.txt files.

Supplemental AI File Uploads

Upload .md and .json files directly from the admin dashboard
Supports multiple file categories: brand info, review roundups, FAQ/PAA documents, products, services, and custom files

Supplemental File Management

Full management table with view, replace, enable/disable, delete, and bulk actions
Per-file controls for fine-grained management

Dedicated Supplemental Sitemap

Separate llms-supplemental-sitemap.xml for your uploaded AI optimization files
Per-file inclusion toggle to control what's published
Automatically registered in Yoast SEO's sitemap index
Dedicated Rank Math supplemental sitemap provider
Added to robots.txt automatically when enabled

Security

Secure file storage with path validation and MIME type checking
Content scanning for script tags, PHP code, and event handlers
JSON syntax validation on upload
Directory protection via .htaccess and index files
Per-file storage tracking

Version 2.3.7 — Hybrid Content Filtering

Released March 2026

This release introduces a multi-layered approach to content filtering, giving site owners fine-grained control over which content appears in their LLMS files.

Hybrid Content Filtering Engine

Three-layer filtering system for noindex detection:
- Layer 1 — Database-level meta queries supporting 8 SEO plugins: Yoast, Rank Math, AIOSEO, SEOPress, The SEO Framework, and SmartCrawl
- Layer 2 — Optional wp_robots() fallback for unknown or unsupported SEO plugins (WordPress 5.7+)
- Layer 3 — Developer filters (llms_amplifier_noindex_meta_query, llms_amplifier_filter_posts) for custom filtering logic

Per-Post Exclusion Controls

New “Exclude from llms.txt” checkbox in the post, page, and product editors
Admin list column showing LLMS.txt inclusion status, with quick edit and bulk edit support

WooCommerce Visibility Filtering

Hidden catalog products automatically excluded
Functional WooCommerce pages (Cart, Checkout, My Account, Shop) automatically excluded

Other Improvements

Password-protected content excluded by default
New Content Filtering Settings section with strict filtering toggle
SEO plugin detection display showing which plugins are active and recognized

Version 2.3.6 — Activation Timeout Fix

Released January 2026

Resolved activation timeout caused by running full content generation during WordPress's activation window
File generation now deferred to 60 seconds after activation via wp_schedule_single_event()
Improved error handling and security logging for first-time file generation
Enhanced stability for large sites with thousands of posts, products, and custom fields

Version 2.3.5 — WordPress Compatibility Fix

Released late 2025

Fixed fatal error caused by calling wp_cache_delete_many(), which is not a standard WordPress function
Fixed wp_cache_flush_group() calls to check function existence before invoking (only available in WordPress 6.1+)
Improved compatibility with older WordPress versions and shared hosting environments

Version 2.3.4 — Cache Management Fix

Released December 2025

This release fixed a critical caching issue where regenerated LLMS files were serving stale content instead of pulling fresh data.

All caches cleared before file generation — main content cache, post query cache, and subdirectory caches
Added cache_results => false to all WP_Query calls to bypass WordPress internal query caching
Switched from get_posts() to WP_Query for better cache control
Single shared content generator instance prevents stale cache state conflicts between file types
Automatic cache invalidation when posts are created, updated, or deleted via clean_post_cache()
Fixed Safari checkbox and button clickability issue in the admin CSS

Version 2.3.3 — Cache Implementation & Timestamp Merge

Released December 2025

Initial cache clearing implementation for regenerated files
Shared content generator instance introduced
Merged the v2.3.2 timestamp display fix and v2.3.0 full functionality into one consolidated admin JS file

Version 2.3.2 — Timestamp Display Fix

Released mid-2025

Fixed timestamp display to update in the admin dashboard without requiring a page reload after regeneration
Auto-refresh timestamp display using jQuery fade transitions

Version 2.3.1 — Authentication Token Update

Released mid-2025

Updated GitHub authentication token for the automatic update system
Improved update mechanism reliability and token expiration handling

Version 2.3.0 — Automatic Meta Tag Injection

Released July 2025

A discoverability-focused release that helps AI crawlers find your LLMS files automatically.

Injects <link rel="alternate" type="text/plain"> meta tags into all page headers, pointing to your LLMS files
Configurable options for which files to reference (llms.txt, llms-full.txt, subdirectory files)
Customizable title attributes on meta tags for better context
Full multisite compatibility
Settings migration for existing installations — meta tags disabled by default on upgrade to avoid unexpected changes

Version 2.2.2 — Automatic Update Improvements

Released July 2025

Improved the automatic plugin update distribution pipeline
Removed the need for users to configure GitHub tokens or edit wp-config.php
Seamless, zero-configuration updates through the standard WordPress update system

Version 2.2.1 — Update Mechanism Fixes

Released July 2025

Resolved issues with the GitHub-based update mechanism
Improved reliability of version detection and download URL resolution

Version 2.2.0 — Creative Commons Attribution

Released July 2025

Added optional Creative Commons Attribution 4.0 International (CC BY 4.0) license headers
When enabled, the license header is automatically prepended to all generated LLMS files
Helps clarify content reuse permissions for AI systems and crawlers

Version 2.1.13 — Update Checker Stabilization

Released July 2025

Removed custom force-release functions that were interfering with natural update detection
Simplified update checker initialization for reliability
Set a reasonable 12-hour check frequency (previously 1 hour)
Fixed GitHub Actions workflow permissions for uploading release assets
Resolved issue where updates downloaded source code instead of the built release ZIP

Version 2.1.1 — Bug Fixes & Defensive Programming

Released June 2025

All class instantiations now check for class existence before loading
Graceful degradation — missing integration classes log errors but the plugin continues operating
Atomic file write operations ensure LLMS files are either fully written or not written at all
Optimized database operations to avoid performance issues on large datasets
Comprehensive and consistent path validation throughout the codebase

Version 2.1.0 — WooCommerce & ACF Integration

Released June 2025

A major feature release adding deep integration with two of the most popular WordPress ecosystems.

WooCommerce Integration

Product prices, SKUs, and stock status included in LLMS output
Product attributes and specifications extracted and formatted
Featured products catalog section for prominent product listings
Configurable product limit to control output size
Privacy-first approach — customer and order data is completely excluded

Advanced Custom Fields (ACF) Integration

Automatic field detection and extraction across all post types
Support for 20+ ACF field types including text, textarea, number, select, checkbox, radio, date picker, and more
Flexible content and repeater field handling with configurable item limits
Field group structure documentation in generated output
Privacy controls — ability to exclude specific sensitive fields by name
ACF Pro feature detection

Other Improvements

New settings migration system for smooth upgrades from previous versions
Integration health checks with admin dashboard status display
Configurable security log retention (default: 1,000 entries)

Version 2.0.6 — UTF-8 Encoding & Modular Architecture

Released June 2025

A significant architectural overhaul that resolved persistent character encoding issues and split the plugin into a maintainable modular structure.

UTF-8 Encoding Fixes

Implemented the proven v1.1.1 encoding approach with ensure_utf8_content() applied at the point of retrieval
Added UTF-8 BOM (Byte Order Mark) to all generated files
Fixed Nordic and special character handling (æ, ø, å, etc.)
Expanded character replacement map for double-encoded content

Modular Architecture

Split monolithic single-file plugin into dedicated class files
Each file kept under 1,000 lines for maintainability
Introduced singleton pattern for the core class and UTF-8 handler

Version 2.0.5 — Simplified UTF-8 Encoding

Released June 2025

Simplified UTF-8 encoding handling for better reliability
Fixed double-encoding issues with international characters

Version 2.0.4 — UTF-8 Refinements

Released June 2025

Further improvements to UTF-8 handling
Enhanced character encoding detection
Fixed PHP parse error in the character mapping array (malformed entry for the Ñ character)

Version 2.0.0 — Enhanced Security Edition

Released June 2025

A complete security overhaul of the plugin, addressing vulnerabilities identified in a comprehensive third-party security review.

Critical Security Fixes

Fixed CSRF vulnerabilities in all AJAX handlers with action-specific nonce verification
Resolved path traversal vulnerabilities with comprehensive validation
Enhanced input validation and sanitization across all user inputs
Improved output escaping throughout the admin interface
SQL injection prevention through prepared statements

New Security Features

Comprehensive security event logging to a dedicated database table
Tracking of failed authentication attempts, path validation failures, AJAX access attempts, file operations, and configuration changes
Content Security Policy, X-Content-Type-Options, X-Frame-Options, and Referrer Policy headers
Explicit file permissions (0644) on all generated files
IP address validation and logging for sensitive operations

Multilingual Support

Full UTF-8 encoding support with 70+ language detection
Custom language and charset override settings
Automatic language detection from WordPress locale

Version 1.2.0

Released May 2025

Basic LLMS file generation functionality
Initial multisite support
SEO plugin integration (Yoast and RankMath)
Subdirectory support for section-specific LLMS files

Version 1.0.0 — Initial Release

Released May 24, 2025

The first public release of LLMS Amplifier.

Automatic generation of llms.txt (overview) and llms-full.txt (detailed content) files
Content selection by post type with configurable post limits
Smart excerpts with configurable word limits and metadata inclusion
Custom content areas (headers and footers) for both files
Scheduled automatic regeneration (immediate, daily, weekly, or manual)
SEO sitemap generation (llms-sitemap.xml) compatible with Yoast SEO and Rank Math
robots.txt integration
Subdirectory support for creating section-specific LLMS files
Admin dashboard with file regeneration and settings management
WordPress 5.8+ and PHP 7.4+ required

Visited 36 times, 1 visit(s) today