ThinkRank – The SEO Plugin Your AI Assistant Can Operate: Metadata, Schema, Sitemaps & Migration

Cofnod Newid

2.0.1

Release Date: 2026-08-23

• New: A Delivery Method setting for your llms.txt file — automatic, a static file in the site root, or served by WordPress. On Nginx the web server sends the file without naming its character set, which turns accented letters and curly quotes into mojibake; served by WordPress it always arrives as UTF-8. Automatic picks the right one for your server
• New: The keyword breakdown — title, meta description, content, image alt, address — now shows on posts with a single focus keyword. It was being calculated and then hidden unless you had two or more keywords, which is the least common setup. Each row now states plainly whether it Matched or not
• New: Page 2 and beyond of an archive or a multi-page post now describes itself: its own title, its own address, and previous/next links, instead of repeating page 1
• Fixed: Settings that saved and then did nothing. The breadcrumb “Show current page” switch, the %site_description% and %tagline% variables, the sitemap include toggles for anything other than the admin screen, and the Open Graph type and Twitter card type choices all persisted without changing what your site published
• Fixed: Saving settings for a category could answer with an error after the save had already succeeded, so you re-entered settings that were never lost
• Fixed: If a settings screen failed to load it silently filled the form with defaults. Pressing Save then wrote those defaults over everything you had stored — on the Robots.txt screen that emptied a saved robots.txt. The screen now shows an error with a Retry button and refuses to save until a load succeeds
• Fixed: Flipping two switches quickly made the first one silently revert and stay reverted after a reload
• Fixed: The Local SEO switch could not be turned on — the save was rejected for a business name whose field only appears after the switch is on
• Fixed: Saving Site Identity stored a copy of the whole response back into your settings, growing the payload on every save
• Fixed: A settings screen would store any stray field a client sent it, and once stored it came back in every later response and was written again on every save, so the settings table and every settings request grew and never shrank. Only settings ThinkRank actually defines are stored now, and strays already saved are cleared on upgrade
• Fixed: An SEO title or meta description saved on a category, tag or custom taxonomy term was ignored on the archive page — the theme’s own title was used and no description was published, while the admin screen showed the value as saved. Open Graph and Twitter values stored on a term now render too
• Fixed: Category, tag, author, date and search archives published no address or description to social networks, and an author archive’s title lost its separator
• Fixed: A Twitter-specific description saved on a post was never used; the card showed the Open Graph description instead
• Fixed: Pages built with shortcodes or page builders published their own shortcode source as the meta description
• Fixed: WooCommerce product pages carried two aggregate ratings, which Search Console reports as a critical error on every reviewed product. ThinkRank now stands aside from WooCommerce’s own product markup when it publishes its own
• Fixed: The sitemap listed the cart, checkout and account pages that ThinkRank’s own robots.txt blocks, so Search Console reported “Submitted URL blocked by robots.txt” on every store
• Fixed: robots.txt put a blank line between each Sitemap line, which ends the record for a crawler, and listed child sitemaps the index already covers. The screen also reported the served file as in sync when it was not
• Fixed: On WPML sites every translation in the sitemap pointed at the default language’s address
• Fixed: The readability score could report “Very Difficult (0)” on ordinary writing, because silent letters were counted as extra syllables
• Fixed: SEO suggestions are now ordered by how many points they can actually recover, and advice for a factor already scoring full marks is no longer listed
• Fixed: Core Web Vitals your site has no field data for were counted as failures, pushing the performance score down by 15–30 points for data you do not control. Unmeasured metrics are now reported as unmeasured
• Fixed: AI-drafted articles could carry heading labels (“H2: “) into the published heading text
• Fixed: Uninstalling the free plugin deleted an active ThinkRank Pro installation’s license and settings. Uninstall also left scheduled tasks and term data behind, and on a network install only cleaned the current site
• Fixed: Bulk SEO Optimization could show one post type’s settings under another and save them there if you switched post types quickly
• Fixed: An imported schema field could not be edited — the text snapped back as soon as it stopped being valid JSON — Print as PDF did nothing under a pop-up blocker, and the social preview could repaint with an out-of-date result
• Fixed: A failed save on Site Identity reported the literal word “undefined” instead of the reason
• Fixed: On sites whose database is not utf8mb4, none of ThinkRank’s tables could be created: every settings save failed with a generic error and Quick Setup could not be completed. Affected sites heal on upgrade, and the real database error is now reported instead of a generic message
• Fixed: A page that does not exist advertised your homepage as its address and carried social tags
• Fixed: Every page carried a second viewport tag beside the theme’s
• Fixed: FAQ blocks on a blog or archive listing published their own FAQ structured data beside the page’s, and a block theme could publish the same questions twice
• Fixed: The MCP connection token was stored in the database in plain text. It is admin-equivalent, so it is now stored hashed and encrypted; existing connections keep working
• Fixed: Several REST routes accepted any post or page id from a user with only a section permission, exposing titles and descriptions from other authors’ drafts, and four of them wrote settings onto content the caller could not edit
• Fixed: The protection against fetching internal addresses checked one address and then connected to whatever the hostname resolved to a moment later
• Performance: Anonymous page views make fewer database queries — a Google token refresh no longer runs on the front end, AI traffic counting is batched instead of locking a row on every visitor, and schema and settings lookups are cached properly

2.0.0

Release Date: 2026-08-20

• New: The FAQ block has a new accordion design and each answer can now carry its own image. The question sits on its own row with a plus that becomes a minus when the answer opens, a line separates the open question from its answer, and the image you add shows with the answer — and travels into the FAQ structured data, so search engines can use it
• Fixed: While editing an FAQ block the question and answer fields did not line up — the question was pushed to the right edge while the answer stayed on the left, which made an empty item look broken
• Fixed: Changing a page’s address from ThinkRank’s Permalink field looked like it did nothing. The new address was saved, but the editor went on showing the old one until the page was reloaded, so it read as though nothing had happened. In Elementor, Divi and Oxygen it genuinely did nothing: the address was never saved at all
• Fixed: The check for whether your keyword appears in the page address was wrong in both directions. On a draft it always reported the keyword as missing, whatever the address said, and only began passing once the post was published. On a published page it could report a match that actually came from a parent page, a category or a date in the address rather than from the page’s own address. Each line of the keyword breakdown now states plainly whether it matched or not, instead of leaving you to read it out of a list of words

1.32.0

Release Date: 2026-08-18

• New: All of ThinkRank’s structured data is now published as one linked graph instead of several separate scripts. A page that had, say, an Article, a breadcrumb trail and an FAQ used to emit three unconnected blocks — search engines now receive one graph in which those entities reference each other, and duplicate or competing entries are merged away
• New: FAQ content is collected from anywhere on the page — the FAQ block, the Elementor FAQ widget and any FAQ you deployed for the post — and published as a single FAQPage instead of several competing ones with different questions
• New: The Instant Indexing Submit URLs screen tells you what will actually be sent before you send it. Line numbers down the side, and each URL checked as you type: URLs pointing at another site, lines that are not URLs and duplicates are all called out, with a counter against the 100-URL limit. Clean up tidies the list, Clear empties it, Cmd/Ctrl+Enter submits, and the success message says how many URLs went
• New: The submission history screen was rebuilt — an empty state instead of a blank table, readable response codes with an explanation of what each one means, and Refresh now really re-reads the log, so URLs you just submitted show up immediately
• New: Every feature on/off switch now saves itself. Flipping a switch is the save — there is no separate Save button to remember, the switch rolls back and tells you if the save fails, and a short message confirms the new state. The Save button stays where you are filling in a form
• New: Screens show a placeholder shaped like the content while their data loads, instead of the word “Loading” or a blank panel
• New: Role Manager can grant or withhold AI Insights and Manage Roles separately, like every other ThinkRank area
• Changed: Connecting ChatGPT and other AI assistants now works on hosts that answer /.well-known/ addresses themselves before WordPress ever sees the request — reported on SiteGround, where the connection failed with “does not implement OAuth”. ThinkRank now advertises an address that reaches WordPress on every host, and publishes the discovery files directly where a host insists on serving them itself
• Fixed: The first few sentences of a password-protected post were published as its meta description, and in its Facebook and X preview text — visible to anyone requesting the page and to every crawler and link preview, while the page itself still showed the password form. Questions from an FAQ block on a protected post could be published the same way
• Fixed: AI content briefs were displayed without sanitizing them. Brief text comes back from an AI provider and can include content pulled from competitor pages you supplied, so it is not trusted input; it is now cleaned before it is stored and before it is shown
• Fixed: A user given access to only the Social Media or Schema section could read the SEO details of posts they cannot edit — including other authors’ drafts and pending posts — by changing the post number in the request
• Fixed: A user given access to only the Settings section could write post-specific social and SEO settings onto another author’s post, changing what its public page shares. The same request also overwrote your site-wide defaults
• Fixed: Saving settings reported a server error even though the settings had been saved, on 8 of the 11 settings groups. People re-entered settings that were never lost, and a genuine failure looked identical to the permanent one. A successful save also came back empty, leaving the form blank
• Fixed: A published llms.txt file could display accented and non-Latin characters as mojibake (“Aktivitäten” as “Aktivitäten”) because the file was served without saying which character set it used. The bytes were always correct — only the declaration was missing
• Fixed: On block themes, a page with an FAQ block published its questions twice — once inside ThinkRank’s graph and once in a second block beside it
• Fixed: A Codex connection on the MCP screen showed a plain letter instead of the OpenAI mark
• Fixed: The two Schema screens showed their title three times over before any content, and the Global SEO, Crawling and Author Archives screens had the wrong icon or a missing divider

1.31.0

Release Date: 2026-08-16

• New: Submission Coverage for Instant Indexing — compares every published URL against the last successful IndexNow submission and tells you what is stale, what failed and what was never announced, then closes the gaps on its own once a day. There is a Reconcile Now button for when you do not want to wait
• New: A noindex you set on a category, tag or custom taxonomy archive is finally applied. The setting saved and read back correctly before, but the archive still carried your site-wide default — and the archive is now dropped from the XML sitemap too, so the sitemap cannot advertise a page that asks not to be indexed
• New: The Robots.txt screen shows what crawlers actually receive right now, and warns you when a robots.txt file sitting in your site’s folder is being served by the web server instead of the content you saved
• Changed: ThinkRank now runs on PHP 7.4 and higher, instead of requiring PHP 8.0
• Changed: WordPress’s own sitemap is switched off while ThinkRank’s is enabled, so your site stops publishing two competing sitemaps and pointing search engines at both
• Changed: Core Web Vitals now reports Interaction to Next Paint, the metric Google replaced First Input Delay with
• Fixed: /sitemap.xml led search engines to a dead page on sites using a sitemap index — the one sitemap URL crawlers guess. It now goes to the sitemap you actually publish, and while your sitemap has not been generated yet WordPress’s own sitemap is left in place rather than leaving that URL answering nothing
• Fixed: Cleaning up sitemaps deleted any file in your site’s folder with “sitemap” in its name — including WordPress’s own and other plugins’ — and then left your site with no sitemap and nothing scheduled to rebuild it. It now removes only ThinkRank’s files and queues the rebuild
• Fixed: Turning off a content type stopped its sitemap from being updated but left the old file serving and listed in the index. Sitemaps you no longer publish are now removed
• Fixed: Regenerating the sitemap from anywhere other than the settings screen could republish one flat sitemap over your sitemap index, or republish files that had lost their styling and their image entries
• Fixed: When a Core Web Vitals check fails, ThinkRank now tells you why — your site could not be reached, no Google connection, or the daily quota is used up — instead of reporting a server error for all three. Pressing refresh after a failure also really re-measures now, and a site set up with only a PageSpeed API key works without a Google account
• Fixed: The historical performance charts drew an empty card for a metric with no measurements, most visibly for the new INP metric on sites with older data
• Fixed: Open Graph told Facebook and LinkedIn your site was in US English no matter what language it was actually in
• Fixed: Pages built with Elementor and similar builders contributed no links, images or headings to SEO analysis, so scores and recommendations were based on text alone
• Fixed: SEO Insights failed with a server error on sites that had connected PageSpeed
• Fixed: With some caching or optimization plugins, ThinkRank’s screens and the Dashboard widget could stay stuck on “Loading” forever
• Fixed: The AI usage overview reported empty fields on a site that had not used any AI features yet

1.30.0

Release Date: 2026-08-13

• New: AI alt text for images — ThinkRank can now look at the picture and describe it, instead of only rewriting the filename. Choose it under Image SEO; the filename template stays the default because AI costs you money per image, and AI runs are capped at 10 images per click
• New: One-click fixes in the Site SEO Audit. Where an issue is safely fixable — search visibility, XML sitemap, structured data, missing image alt text — a button applies it and the audit is re-scored on the spot. Fixes with consequences warn you before you click, and things that are your call (your titles, your permalinks, your server) are deliberately left alone with an explanation
• New: A warning when WordPress is set to discourage search engines from indexing your site, so the setting that quietly undoes all your SEO cannot sit there unnoticed
• Fixed: Opening the editor could hammer your site with repeated SEO requests until pages took 30-40 seconds to load, or failed outright. The panel now makes one request at a time, and only keeps checking while ThinkRank is actually writing metadata in the background
• Fixed: SEO data imported from another plugin did not appear in an editor that was already open until you reloaded the page
• Fixed: Saved content briefs could not be listed, exported or deleted on sites with no AI key configured — the screen failed with a server error, even though none of those actions need AI
• Fixed: On some hosts ThinkRank’s settings tables were never created, so every save in Site Identity, Sitemap, Social and Schema silently failed and the setup wizard dead-ended after a migration. The tables now fit the stricter index limit on those servers, and existing sites are migrated
• Fixed: A settings save that fails now records why in the error log, instead of leaving you with “Failed to update settings” and nothing to act on
• Fixed: The one-click sitemap and schema fixes reported success while saving nothing, and the audit’s sitemap check could never fail no matter what your settings said
• Fixed: Author archives printed two meta descriptions, ended their title in a stray space, ignored your chosen title separator, and would not let you clear the title or description templates — a cleared field came back on the next load
• Fixed: The alt-text coverage panel could claim “6 of 5 images have alt text” by counting trashed images on one side only, and the bulk alt fill could spin forever on sites with private images, turning one click into thousands of requests
• Fixed: The Alt text source setting accepted any value, so an unrecognised value silently fell back to the template and the AI batch limit stopped applying
• Fixed: Saving Schema settings could white-screen the site on WordPress 6.0, after the settings had already been written
• Fixed: Email reports — the Save button disappeared when you toggled the feature off, so turning it off never stuck; changing the frequency did not move the next send; a failed send still consumed the whole reporting period; the report described three different date ranges at once; “Next report” showed the wrong time outside UTC; pages and keywords that lost all their clicks were dropped from the Top Losing sections; and a report with no sections could be configured and sent empty
• Fixed: Deleting ThinkRank with “Delete all data on uninstall” enabled while ThinkRank Pro was active silently reinstalled the free plugin and recreated all of its data
• Fixed: Data Management could not be saved on sites without an AI API key
• Fixed: An author or contributor granted Schema Manager access could deploy structured data onto other people’s posts
• Fixed: Alt text beginning with an accented character was corrupted
• Changed: AI Visibility data is now removed when you delete the plugin with data deletion enabled, and old probe transcripts are aged out after 90 days instead of accumulating forever
• Changed: The Integrations tab no longer displays or re-saves settings it has no control over, and a partial save now leaves untouched settings exactly as they were

1.29.0

Release Date: 2026-08-11

• New: Show/hide your API key while typing it, and a Replace control so a saved key can be swapped without clearing it first. Saved keys are now only ever sent back to your browser masked, never in full
• New: A social card on the dashboard with the ThinkRank community and channels in one place
• Fixed: Content briefs that came back as “Unable to parse AI response”, or saved a brief full of raw API output. When the AI refuses, is cut off at its token limit, or answers in an unexpected shape, ThinkRank now tells you what happened instead of saving the failure as a brief
• Fixed: Short, Medium and Long content briefs all asked the AI for the same output budget, so a Long brief could be truncated while a Short one paid for room it never used. Each length now requests a budget that fits it
• Fixed: Briefs on GPT-5 models ran at maximum reasoning effort by default — the slowest and most expensive setting, with much of the budget spent on hidden reasoning before any text was written
• Fixed: A request that ran past the time ThinkRank allows was retried twice more, each attempt near-certain to time out again — multiplying the wait while the abandoned requests kept running, and billing, on your AI provider. A timed-out request now fails once, straight away, instead of being retried
• Fixed: “X days ago” counts on sites outside UTC were off by the site’s time offset
• Fixed: Saving settings could send your API key back to the browser in the clear, and re-saving a form that displayed a masked key could overwrite the real key with the mask
• Fixed: Importing SEO data from another plugin read serialized values from that plugin’s tables in a way that could be abused to create PHP objects on your site; those values are now parsed without ever instantiating an object
• Fixed: Fetching a schema or competitor URL could be pointed at private, internal or cloud-metadata addresses, including addresses disguised as IPv6. All outbound fetches now share one check, applied again on every redirect hop
• Fixed: Dismissing a ThinkRank admin notice, importing or resetting the whole configuration, and adding database indexes no longer accept requests from users without the capability for them
• Changed: Brand Visibility is temporarily hidden while it is reworked. Nothing is deleted — recorded runs, history and settings are kept, and the feature returns in a later release

1.28.0

Release Date: 2026-08-10

• New: AI Insights — see how much of your traffic comes from AI assistants and AI crawlers, first-party. AI platforms strip referrers so Google Analytics files those visits under “Direct”; ThinkRank reads them as they arrive. Daily totals only — no IPs, no cookies, no per-visit tracking
• New: Brand Visibility measures whether AI assistants mention your brand — asking each question several times for a real mention rate instead of a one-shot yes/no, ranking you against named competitors as share of voice, and scoring everything into a 0–100 Visibility Index
• New: Brand Visibility uses its own API key and model per AI platform, so measuring visibility never means switching the provider that writes your metadata. Adds Perplexity, which searches the web live — the closest thing to what a person actually sees
• New: Auto AI writes a missing SEO title or description when you publish, in the background, so publishing never waits on an AI call. It fills empty fields only and never overwrites what you wrote
• New: Quick Edit SEO on the posts list — edit the SEO title, meta description and focus keywords straight from the posts table, without opening the post. Cleared fields fall back to your Global SEO patterns
• New: Traffic Recovery tab on the AI Tools screen, which finds published pages losing search traffic and writes a refresh brief for each
• New: A Migration screen for moving SEO data in from another plugin, with a re-scan control that re-detects what is installed
• New: Social platform verification is now one tab per network, with per-card saving and masked verification codes
• Fixed: The AI Traffic and Brand Visibility tables were never created, so AI traffic went unrecorded and every brand check spent a paid AI call without saving any history
• Fixed: Brand Visibility failed to start on upgraded sites with “Could not start the analysis run”. A missing table is now detected and repaired on load, instead of going unnoticed until a deactivate/reactivate
• Fixed: Brand checks reported “not mentioned” for questions that could not plausibly omit your brand — the AI never actually returned an answer, and the blank was saved as a clean negative. A probe with no answer now fails visibly and is never stored
• Fixed: A Brand Visibility run could sit at 0% forever, and progress never advanced while it was running
• Fixed: A site that configured Brand Visibility on Pro and then lapsed to free kept running the larger Pro batch on the user’s own paid API keys
• Fixed: The SEO Overview column showed “Not Analyzed” forever on sites that never ran an analysis, even right after editing a post’s SEO title and description
• Fixed: Saving in the editor could wipe an SEO title or description that Auto AI, bulk optimization or an import had just written, and the editor kept showing the old empty value until a manual reload
• Fixed: Google Analytics, Search Console and PageSpeed were reported as “not configured” on sites connected through Google sign-in, and a site with a perfectly working GA4 tag was told its tracking was unverified
• Fixed: Reconnecting a different Google account kept showing the previous account’s Search Console properties for up to 30 minutes, and domain properties could not be found by searching
• Fixed: The ThinkRank data stream picker in Pro went blank on every reload
• Fixed: ThinkRank’s styles leaked into the WordPress admin and page builders

1.27.0

Release Date: 2026-08-06

• New: Connected Apps page for MCP — see every AI app connected to your site, when it last used the connection, and disconnect any of them in one click
• New: AI now writes in your site’s language. Titles, descriptions and content briefs came back in English on non-English sites; they now follow the post’s language, or the site language, and can be overridden with a filter
• New: Instant Indexing checks that your IndexNow key file is actually reachable and tells you what is wrong before you submit a single URL
• New: The MCP connection test detects hosts that block AI clients by User-Agent — the “couldn’t register with the sign-in service” failure that no other check could see
• Fixed: MCP could connect successfully and offer no tools at all, with nothing reporting a problem — the most common cause was another plugin’s copy of the Abilities API loading first
• Fixed: A connector still using an old token could lock itself out of your site permanently by retrying. The lockout window no longer extends on every retry, and the plugin now shows when clients are locked out
• Fixed: ChatGPT rejected some sites with “MCP server does not implement OAuth” — the sign-in handshake itself was tripping the rate limiter
• Fixed: Every IndexNow submission returned 403 on hosts with a read-only web root, because the key file could never be written. The key is now served directly by WordPress, and the error message names the file and what to check
• Fixed: Importing from Rank Math (and Yoast, AIOSEO, SEOPress) could fail outright on a single unexpected setting value
• Fixed: The “ThinkRank SEO” metabox was hidden by a leftover style rule — Classic Editor users never saw it, and the block editor’s panel toggle appeared to do nothing
• Fixed: Missing borders in the SEO drawer inside the Divi Visual Builder
• Improved: The plugin package now installs as an upgrade rather than a second copy, and ships the WPML config file that earlier packages left out

See changelog for all versions.