What’s new in MrNiche Autoblogger.

Knowledge Anchors — give the AI real facts to cite

Articles ask the model to “be specific” while forbidding fabricated stats. Without a source pool the model either invented numbers or stayed generic. Knowledge Anchors close that gap — paste verified facts about your business, products, or topic into your style profile and the AI cites from that list verbatim.

• New: Knowledge Anchors field on every writing style profile (Pro plugin → Writing Styles → edit profile)
• New: Anchors injected automatically into both single-pass and multi-pass generation pipelines, with explicit no-extrapolation instructions
• New: Per-style anchors — different businesses or content niches each get their own fact pool
• Improved: Specific-claims rate up significantly without raising fabrication risk — the model now has a concrete source to draw from instead of falling back to abstract claims
• Compatibility: Existing style profiles work unchanged. Anchors field is optional — leave empty for v5.10 behavior.
• Full guide: /docs/writing-styles/

License revocation propagates within an hour

• Improved: Re-verify cadence reduced from 24 hours to 1 hour — revoked or expired keys stop generating articles within the hour, not the previous worst-case 8 days
• Improved: Offline grace period tightened from 7 days to 24 hours — long enough for maintenance windows, short enough to prevent unauthorized coasting
• New: “Re-verify with server now” button on the License page for instant manual checks
• New: “Last verified: N ago” indicator under the status pill so you can see your install’s heartbeat at a glance
• Improved: Auto-verify on License page load when the cached check is older than 5 minutes
• Fixed: Async re-verify path (silently broken on most hosts due to loopback POST cookie issues) replaced with synchronous inline verification — adds ~200-500ms once per hour, negligible elsewhere

WordPress-native auto-updates for Pro

• New: “Update Available” prompt appears in WP admin when a new Pro version ships — same UI as any wp.org plugin update
• New: Update payload signed and gated on valid license + activated domain — cracked or unlicensed installs see no update path
• New: 12-hour update-check cache with smart fallback (30 min on network errors, 1 hour on missing key) — minimises license-server traffic without delaying real updates
• New: Signed download URLs (24-hour TTL, HMAC-SHA256) — download links can’t be reused or leaked between sites
• Improved: All Pro improvements (Waves 1–7, security patches, anti-tamper) now reach customers automatically — no more manual re-download after every release

Wave 7 — curly-apostrophe fix + heading anti-formula

• Fixed (significant latent bug): Apostrophe regex patterns across 6+ humanization functions only matched straight quotes (U+0027) — articles with curly apostrophes (U+2019, common after WordPress wptexturize) silently bypassed roughly a third of the humanizer pipeline
• New: Automatic H2-shape variety enforcement — when more than half of an article’s headings open with imperative verbs (“Choose…”, “Set…”, “Build…”), a one-shot LLM rewrite mixes in question, definitional, data-led, comparative, and noun-led forms instead. Cost: ~$0.001-0.002 only when triggered
• New: Outline pass now bans Introduction / Overview / About / Background / Preface as H2 prefixes — the first paragraph IS the introduction
• Improved: Generic “Introduction” / “Overview” headings stripped from output (sister to the existing Conclusion stripper)
• Improved: House style rules extended with HEADING SHAPE VARIETY hard rule (max 2 H2s with same syntactic form per article)

Wave 6.1 — humanizer polish

• New: Self-narrating meta-prefix stripper — the model used to literally announce “One short sentence matters here:” before complying with the burstiness rule. Eight patterns now removed (To put it simply / Bottom line / Plain and simple / Make no mistake / Simply put / Long story short / Just so we’re clear / Look)
• Improved: Burstiness check now mid-article-only — first and last paragraphs no longer count toward the short-sentence quota (the closing paragraph was naturally short and was masking body uniformity)
• Fixed: Negation-restatement detector now matches curly quotes and parens after the restatement-start (“X isn’t (‘really’) Y, it’s Z” patterns)
• Improved: Voice pass FORBIDDEN list aligned with main humanizer banlist + post-voice processing chain extended to re-run intro/conclusion strippers

Featured-image duplication fix

• Fixed: AI-generated header image was appearing both as the WordPress featured image (auto-rendered by most themes) AND inline at the top of the post body — now defaults to featured-only, no duplicate inline copy
• New: “Insert featured image inline” toggle (default OFF). Turn ON if your theme doesn’t auto-render featured images at the top of posts
• Improved: Multi-image articles handle the featured/inline split cleanly — image #1 = featured only, images #2-3 = inline at intervals
• New: Direct URL-to-featured-image path for external image sources (Unsplash, Imagine.art, MonsterAPI) — no more intermediate inline insert step

Anti-tamper layer + 7 security patches

• New: Five differently-named license checks across all hot-path entry points — defeats naive “paste plugin into ChatGPT, ask it to disable license” attacks
• New: Two-path license verification on every check (option flag + class method + peer-existence check) — stub-class swaps and magic-method tricks both caught
• Security: SSRF guard on image downloader — refuses URLs to private/loopback/link-local/reserved IP ranges (RFC1918, 127/8, 169.254/16, ::1, fc00::/7)
• Security: Image extension whitelist (jpg, jpeg, png, gif, webp) — falls back to Content-Type when path has no extension
• Security: JSON-LD output now uses JSON_HEX_TAG | JSON_HEX_AMP | JSON_HEX_QUOT | JSON_HEX_APOS flags so AI-generated content can’t break out of the script tag
• Security: Webhooks now send only HMAC-SHA256 signature — raw secret is no longer transmitted. Breaking change for subscribers checking the raw secret; migrate to HMAC verification (hash_hmac('sha256', $raw_body, $secret) + constant-time compare)
• Security: wp_kses_post() runs on AI output before wp_insert_post — prompt injection of <script> / <iframe> / onerror= can no longer reach the published post
• Security: API key inputs no longer reflect the saved value back to the form — placeholder shows “Saved — submit a new key to replace” instead

Wave 6 — burstiness + negation-restatement killer

Verified result on test articles: 5% AI / 100% human across two AI detectors.

• New: Negation-restatement breaker — splits “X isn’t Y, it’s Z” comma-spliced contrasts into two declarative sentences. Models love this construction; humans almost never use it at the same rate
• New: Burstiness enforcer — programmatic guarantee that at least one mid-article paragraph contains a single short sentence. Breaks the uniform-medium-cadence AI fingerprint
• New: Fake-anecdote stripper — when no author bio is set (default), regex-removes fabricated first-person patterns (“One [profession] I spoke with…”, “I interviewed…”, “In my own practice…”)
• Improved: Voice pass + house-style rules rewritten to drop the “(invent plausibly if needed)” license — explicit no-fabrication rule. Concrete-but-unsourced beats specific-but-fabricated
• Improved: Banned-vocabulary list extended: valuable insights → insights, shed light on → show, stark reminder → reminder, myriad of → many

Optional Voice Pass + dashboard accuracy fixes

• New: Optional Voice Pass toggle (off by default) — applies a “sharp, slightly contrarian columnist” rewrite with required edits (mild dissent, first-person observation, pull-quote, named real example, varied paragraph length). Roughly doubles per-article API cost (~$0.012 → ~$0.024 on GPT-5.4-mini)
• New: Persona-aware voice pass — when an author bio is set, prompts for first-person observation; when empty, prompts for third-person editorial judgment. Explicitly forbids fabricating personal anecdotes
• New: “Reset usage stats” button in API Settings (decoupled from main settings form so accidental Enter doesn’t trigger it)
• Fixed: Dashboard post counter no longer overcounts — split into per-API-call token tracking and per-insert post tracking
• Fixed: Cost estimate accuracy — now uses live model-cost lookup instead of hardcoded $0.01/1K rate
• Fixed: JSON-LD moved out of post body — was appending <script> tags to post content, getting stripped by content filtering and leaving raw JSON visible. Now lives in post meta, emitted via wp_head
• Fixed: AI markdown code fences (“`html, etc.) automatically stripped from output
• Fixed: License activation now surfaces real error messages (was reading the wrong response key — customers saw blank errors)
• Improved: Token output limit raised from 4,096 to 8,192 across all models — multi-pass expand uses the headroom

GPT-5 model line-up + Wave 4 structural enforcer

• New (v5.2.0): GPT-5.4 Mini (default, ~$0.0036/1K tokens), GPT-5.4 Nano (cheapest, ~$0.0010/1K), GPT-5.4 (~$0.0125/1K), GPT-5.5 (flagship, ~$0.0250/1K)
• Critical fix (v5.2.0): OpenAI retired gpt-4o, gpt-4o-mini, gpt-4-turbo, gpt-4, and gpt-3.5-turbo from the API on Feb 16, 2026. MrNiche Autoblogger now auto-migrates customers stuck on retired model IDs to the closest current replacement on next plugin update
• Hotfix (v5.2.1): GPT-5 family rejects max_tokens — request payload renamed to max_completion_tokens (also accepted by older models, no version branching needed)
• New (v5.2.2 / Wave 4): H3 asymmetry enforcer — when more than two-thirds of H2 sections have H3 sub-points, the wrapper tags are stripped from random middle sections (paragraph content preserved beneath). Defeats the symmetric-tree AI fingerprint that prompts alone couldn’t fix
• Improved (v5.2.2): Generic “Conclusion: …” / “Final Thoughts: …” headings with arbitrary subtitle now stripped (previous version only caught bare “Conclusion”)

Humanizer rewrite — Waves 1, 2, 3

First major rewrite of the humanizer pipeline to defeat AI-detector fingerprints. The pivot: stop relying on prompts to police style — enforce rules in code on the output instead.

• Fixed: Multi-pass writing now respects the Article Prompt from Prompt Studio. Previous code hardcoded outline + expand specs and silently ignored your saved prompt
• New: Filler-opener stripper — regex-deletes “Here’s the thing”, “Let’s face it”, “Think about it”, “In today’s world”, “Nowadays”, “It’s no secret”, and 8 more
• New: Banned-vocabulary substitution — replaces overused AI words with neutral alternatives (delve→explore, leverage→use, robust→strong, elevate→improve, harness→use, seamless→smooth, cutting-edge→modern, game-changer→major shift, embark on→start, navigate the landscape→understand the field)
• New: Generic conclusion-heading stripper — removes <h2>Conclusion</h2> and Final Thoughts / Wrap-Up / Summary / Closing Thoughts variants, keeps the prose underneath
• New: Per-article voice variation — picks one of six mild stylistic constraints at random per article (no bulleted lists / open with question / open with anecdote / skip Conclusion / include pull-quote / include numeric example) so a single site doesn’t read identically article-to-article
• New (v5.1.1 / Wave 3): House-style auto-injection — canonical rule block auto-appended to every brief regardless of customer’s saved prompt. Single source of truth in code: paragraph-length variance, structural asymmetry, banned openers, banned closers, em-dash cap (max 2), stakes/nuance directive
• New: TOC restricted to H2 only + auto-skipped on articles under 3 sections
• Improved (v5.1.1): FAQ HTML now runs through the AI-tells scrub before append (was previously raw — banned vocabulary leaked through verbatim)

Major architecture rewrite — multi-file class-based plugin

• Refactored the entire plugin into modular class files (Logger, License, API, Images, SEO, Queue, Generator) for maintainability and faster updates
• New: Background queue system — articles generate via WP-Cron every 2 minutes, no PHP timeouts on shared hosting
• New: Multi-pass writing pipeline — outline → expand → humanize, producing 1,400–3,000-word articles
• New: Domain-locked licensing with 7-day grace period for network reliability
• New: Article + FAQPage JSON-LD schema emitted via wp_head (survives content filtering)
• New: HMAC-signed webhooks for Zapier, Slack, Make, and custom endpoints
• Improved: Dashboard now correctly counts one article = one post (was overcounting per API call)
• Improved: Per-model OpenAI cost estimation with 2026 pricing
• Improved: AI output stripped of stray markdown code fences before publishing

Image provider expansion + style profiles

• New: Imagine.art and MonsterAPI image providers (alongside DALL·E 3 and Unsplash)
• New: Custom writing style profiles with system-prompt-level voice control
• New: Visual content calendar with bulk-management actions
• Improved: Yoast and Rank Math meta injection now toggleable per-site
• Fixed: Featured image fallback when DALL·E quota is exhausted

Humanizer pass + AutoPilot scheduler

• New: Humanizer rewriting pass — varied sentence length, contractions, natural transitions
• New: AutoPilot scheduler with daily/weekly cadence + per-category title generation
• New: AI-driven internal linking from new articles to existing posts
• New: Auto-generated Table of Contents block