Description
AMDT AI Traffic Analytics watches; it does not touch. It records two things your usual analytics do not show:
- The AI bots that read your pages (GPTBot, ClaudeBot, PerplexityBot, Googlebot, Bingbot and the rest), and whether each one is who it says it is. It checks the bot’s IP address against the lists its operator publishes and against its reverse DNS, and flags as an impostor any visitor that uses another bot’s name. The Bot catalog shows how big each list is and the date its operator gave it, so you can tell a list kept up to date from one left untouched for a year. Agents that sign their requests with Web Bot Auth (Google-Agent, and ChatGPT agent, which uses the User-Agent of an ordinary Chrome) are verified by their signature, which cannot be forged the way a User-Agent can; IP analysis lists each signed visit with its three headers.
- The people who arrive from an AI assistant (ChatGPT, Claude, Perplexity, Gemini, Copilot…) after clicking a citation, and the page they land on.
With that it gives you a dashboard (its chart of content consumption also as a panel of the WordPress Dashboard), a log you can filter and export, and an analysis of the addresses it could not identify, with blocking rules already written for Apache, Nginx, Cloudflare or Wordfence. You apply those rules in your firewall: the plugin does not block anyone.
It also publishes your site’s llms.txt (an index of your content for language models), counts who reads it, and lets you leave posts out of it from the editor.
And with Agent-Readiness it tells you whether an AI agent that navigates and acts on pages (it fills in forms, presses buttons, opens menus) can use yours. It reads the HTML your server gives for the home page, a sample post, up to 8 pages and a sample product, or for any page of your site you paste, and for each one:
- scores from 0 to 100 what such an agent relies on: real buttons and links, the heading outline, the landmarks, labelled form fields, image alternatives and how much of the page is bare
div; - approximates Lighthouse’s «Agentic Browsing» audits as far as the server can tell them: the 33 accessibility rules of its accessibility-tree audit, the WebMCP ones,
llms.txtand the ARD manifest. Visual stability (CLS) needs a browser: each result links to PageSpeed Insights for it; - says how much of the text reaches the initial HTML and how much needs JavaScript;
- explains how to fix it with your page builder (Elementor, Divi, Bricks, Beaver Builder, WPBakery, the block editor or your theme), and exports everything to Markdown or CSV.
It only analyses your own site, and it never changes your pages.
A complete edition, with more modules, also serves your content to AI agents through an API and an MCP server, publishes the files that announce them, declares AI usage permissions, protects that API and analyses your content. Both editions share the same tables and settings: you can move from one to the other without losing data.
What it stores
In your own database, for the retention you choose (30, 60 or 90 days):
- For the bots in its catalogue and for whoever reads your llms.txt: the IP address, the User-Agent and the page requested.
- For people who arrive from an AI assistant: only the network of their IP address (IPv4 up to the third block, IPv6 the first 48 bits), never the full address, plus the User-Agent, the page and the address of the assistant they came from.
- For automated browsers that pretend to be an iPhone (possibly Grok): the request, with its IP address.
- For requests signed with Web Bot Auth: also their three signature headers (
Signature-Agent,Signature-Input,Signature) and the result of checking them.
Nothing is stored about ordinary visitors. If you turn on the debug log (Settings → Diagnostics), its entries also contain IP addresses and User-Agents. It is written to its own folder in uploads, amdt-ai-traffic-analytics/, in a file with a random key in its name; the folder denies direct access on servers that read .htaccess, such as Apache. It is deleted on uninstall if you choose to delete the data. Settings → Privacy → Policy guide has a suggested text for your privacy policy.
What it leaves in your visitors’ browsers
Nothing: no cookies, nothing stored in the browser. Pages carry a small script that, only when the visitor comes from an AI assistant, tells your own server about it.
External services
The plugin sends nothing to its author and has no telemetry. It connects to the services below. The IP lists and the signing keys are off on a new install until you turn them on.
Why the IP lists. To confirm that a visit from GPTBot, OAI-SearchBot, ChatGPT-User, OAI-AdsBot, ClaudeBot, Claude-User, Claude-SearchBot, PerplexityBot or Perplexity-User really comes from its operator, the plugin can download, once a day, the lists of IP addresses those operators publish. It is off until you turn it on: a notice asks you when you activate the plugin, and Settings has the switch.
What is sent, and when. Once a day while the switch is on, and once when you turn it on, the plugin makes a plain HTTPS GET request for each public file below. The request carries the plugin’s name and version as its user agent and nothing else: no data about your site, your visitors or you.
OpenAI (GPTBot, OAI-SearchBot, ChatGPT-User, OAI-AdsBot)
- Files downloaded: https://openai.com/gptbot.json, https://openai.com/searchbot.json, https://openai.com/chatgpt-user.json, https://openai.com/adsbot.json
- What they are: Overview of OpenAI Crawlers
- Provided by OpenAI: terms of use, privacy policy
Anthropic (ClaudeBot, Claude-User, Claude-SearchBot)
- File downloaded: https://claude.com/crawling/bots.json. claude.com is Anthropic’s domain for Claude; its own pages link Anthropic’s terms and privacy policy, below.
- What it is: Does Anthropic crawl data from the web, and how can site owners block the crawler?
- Provided by Anthropic: consumer terms, privacy policy, Anthropic Privacy Center
Perplexity (PerplexityBot, Perplexity-User)
- Files downloaded: https://www.perplexity.ai/perplexitybot.json, https://www.perplexity.ai/perplexity-user.json. These are the addresses Perplexity’s documentation gives; until 2.7.13 the plugin used www.perplexity.com, which redirects to them.
- What they are: Perplexity Crawlers
- Provided by Perplexity: terms of service, privacy notice
Google and OpenAI (the keys of Google-Agent’s and ChatGPT agent’s signatures)
Google-Agent and ChatGPT agent sign some of their requests (Web Bot Auth, an IETF draft based on HTTP Message Signatures, RFC 9421). To check such a signature the plugin downloads the public keys of the agent that signed it. It does so only when a signed visit from one of them arrives and the keys it keeps are out of date (they are kept as long as the response allows, between five minutes and a day), only from the two addresses below, never from an address a visit names, and only while the switch of the IP lists is on (Settings → Operator IP lists). The request is a plain HTTPS GET with the plugin’s name and version as its user agent and nothing else.
- Files downloaded: https://agent.bot.goog/.well-known/http-message-signatures-directory (Google-Agent), https://chatgpt.com/.well-known/http-message-signatures-directory (ChatGPT agent)
- What they are: Authenticate requests with Web Bot Auth (experimental), ChatGPT agent allowlisting
- Provided by Google: terms of service, privacy policy. Provided by OpenAI: terms of use, privacy policy
Reverse DNS
To tell which network a bot’s address belongs to, the plugin asks the DNS resolver your server uses for the name of the IP address of the bots and of whoever reads your llms.txt (never of the people who arrive from an assistant). This goes through your server’s own resolver, as any reverse lookup does. At most two lookups wait for an answer at the same time; a visit that finds both in use for more than 0.1 s is logged as unverified instead of waiting (the amdt_ait_rdns_max_concurrent filter changes the number, and 0 turns reverse DNS off).
Your own site
When you open the Dashboard or Settings, the plugin requests your own home page, at most twice a day, to find out whether a cache sits in front of WordPress, and your own /llms.txt, at most once a day, to check that visitors get it. Both are requests to your own server; the “Check again” buttons repeat them on demand.
When you run Agent-Readiness, the plugin requests the pages of your site it analyses and, for the Lighthouse audits, your site’s /llms.txt, /robots.txt and the addresses where an ARD manifest can be (/.well-known/ard.json, /.well-known/ai-catalog.json and any your pages announce). Only your own site, with its certificate checked; nothing else is contacted. The PageSpeed Insights link next to each result is an ordinary link: the plugin sends nothing to Google, and what PageSpeed Insights receives is what you send it if you open the link.
Third-party libraries
The charts of the plugin’s Dashboard, and its panel on the WordPress Dashboard, use Chart.js 4.6.1 (MIT license), bundled in admin/js/chart.umd.min.js: the minified build its authors publish on npm, without the line that points to its source map. Its source code: https://github.com/chartjs/Chart.js
The accessibility rules of Agent-Readiness use the WAI-ARIA tables of axe-core 4.14.0, Copyright (c) Deque Systems, Inc., under the Mozilla Public License 2.0, which is compatible with the GPL. They are in includes/data/aria-standards.php, generated from its lib/standards folder; the rules are written in PHP following its own, and the texts of each rule are based on its help texts. Its source code: https://github.com/dequelabs/axe-core
Installation
- Install the plugin from Plugins → Add New, or upload the
amdt-ai-traffic-analyticsfolder to/wp-content/plugins/. - Activate it.
- Decide whether to download the operators’ IP lists and signing keys (the notice after activation, or Settings). Without them, visits from OpenAI’s and Perplexity’s bots are logged as unverified, and signatures are kept unchecked.
- If a CDN or a proxy other than Cloudflare sits in front of your site, declare it in AI Traffic → Settings → Visitor IP. Otherwise every visit is logged with the proxy’s address.
- If you use a page cache plugin, see AI Traffic → Settings → Sync with cache plugins.
FAQ
Does it block bots?
No. It reports, and gives you the rule already written; deciding and applying it is up to you.
I use a cache plugin. Does it still measure?
Not entirely, and it is worth knowing: a page the cache serves without running WordPress does not go through the plugin, so that bot visit is not logged. With WP Rocket, LiteSpeed Cache or W3 Total Cache, Settings → Sync with cache plugins offers to keep bots out of the cache with one button, and tells you what it costs. With a server cache (nginx, Varnish, your host’s, Cloudflare caching HTML) or with WP Super Cache it cannot be done; the screen explains it when it detects one, which is not always. An agent that signs its requests but uses a browser’s User-Agent, such as ChatGPT agent, cannot be kept out of the cache that way: it is seen only on the pages the cache does not serve. Visits of people from AI assistants are measured on cached pages too: the browser reports them. If WP Rocket (Delay JavaScript execution) or LiteSpeed Cache (Load JS Deferred: Delayed) hold scripts back until the visitor touches the page, the plugin lets its own small script through, so a visitor who reads and leaves without scrolling is still counted. Other optimizers with a similar option have not been tested; if yours holds scripts back, exclude the one with the id amdt-ait-referral-tracker. The Dashboard also warns you when it finds a cache in front of WordPress, says where to act, and checks again when you ask it to.
What if Yoast SEO or Rank Math already serves my llms.txt?
Then that plugin serves it, and this one neither generates its own nor can count who reads it; the Dashboard says so.
Why are visits from GPTBot or PerplexityBot “unverified”?
Because the download of the operators’ IP lists is off. Turn it on in AI Traffic → Settings → Operator IP lists. ClaudeBot, Googlebot and Bingbot are still checked by reverse DNS without the lists.
Is the data deleted when I uninstall it?
Not by default. Only if you untick “Keep data on uninstall” in Settings, and never while the other edition is still installed.
Screenshots







Changelog
2.7.24
- Reverse DNS: at most two lookups wait for an answer at the same time. Each lookup runs inside the request of the visit being measured, so a burst of addresses never seen before, while DNS was slow, kept one PHP worker waiting per address for the resolver’s whole timeout, all at once; and a slow reverse zone behind a working resolver never triggered the ten-minute pause. Now a visit that finds both slots in use waits up to 0.1 s for one; if they are still in use, it is not looked up and is logged as unverified, never as an impostor. Nothing is cached, so the address is checked on its next visit. The slots are file locks in the server’s temporary folder, shared by every PHP worker and released even if one dies (MySQL’s
GET_LOCK()where that folder cannot be used). Theamdt_ait_rdns_max_concurrentfilter changes the number; 0 turns reverse DNS off. Settings → Diagnostics shows the limit and in how many minutes a lookup found no free slot. - Reverse DNS: the check that a host name resolves back to the address is cached under the domain without its wildcard, so
*.googlebot.comandgooglebot.comno longer make one lookup each. A Googlebot visit from an unlisted address made 2 forward lookups, now 1; with five names in its PTR record, 9, now 5. - Load test (PHP-FPM with 10 workers; 50 simultaneous requests from new addresses with Googlebot’s User-Agent, with empty caches; a DNS server that does not answer, 5 s per lookup): with a slow reverse zone, p50 from 15.9 s to 0.93 s, p95 from 26.3 s to 1.54 s, and 50 DNS lookups down to 2; the time the workers spent inside PHP, from 262 s to 22 s. With the resolver down, p50 from 5.8 s to 0.98 s. Reading /llms.txt with Chrome’s User-Agent, p50 from 15.2 s to 0.56 s. With a DNS that answers, as before. Every visit was logged as unverified, none as an impostor.
- Tested on the ZIPs of this version: every earlier battery, all passing, and 80 checks of this release on this edition: the limit and its filter; slots taken with files or, without the temporary folder, with database locks; waiting for a slot freed after 50 ms and giving up on one held for 0.4 s; no lookup and nothing cached without a slot; Diagnostics; and the lookups one Googlebot visit makes. 459 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (182 warnings). With every category, 0 errors (186 warnings), as in 2.7.23. With the settings WordPress.org uses on each upload, nothing. Correction to 2.7.23: «every earlier battery, all passing» was not so. Two checks of one battery looked for a Spanish heading as it read before the translation was revised («plugins de cache», now «de caché»); the plugin was right, and the test was corrected.
- Thanks to PeepSo for reporting the slow reverse-DNS case.
2.7.23
- llms.txt: when your SEO plugin (Yoast SEO, Rank Math or All in One SEO) marks a whole content type as noindex, its entries are no longer listed one by one to leave them out: the type is left out, or only its entries set to «index» are asked for. The file lists the same entries as before, sticky posts included. On a test site with 100,000 posts of a noindex type, each llms.txt took 335 ms instead of 68 ms and some 60 MB more memory; now it takes 7 ms and no more memory.
- The plugin’s queries of noindex entries no longer set
suppress_filterstotruethemselves. It is whatget_posts()does by default, so nothing changes: entries in every language are still seen with WPML and Polylang. Plugin Check flagged it as an error under the WordPress VIP performance rules. - Spanish translation: the 206 strings the Spanish (Spain) team changed on translate.wordpress.org to follow its glossary, so that the translation bundled with the plugin says the same as the language pack. One keeps «la cascada actual» (the note on
legacy_unknownentries), as its English text says. - Tested on the ZIPs of this version: every earlier battery, all passing, and 197 checks of this release on this edition: llms.txt and llms-full.txt with a whole content type noindex in Yoast SEO, in Rank Math and in both, listing the same entries as before, sticky posts included; and the Spanish translation against the 1,294 strings of translate.wordpress.org. 440 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (182 warnings, as in 2.7.22); with every category, 0 errors (2.7.22 had 2, the
suppress_filtersones); with the settings WordPress.org uses on each upload, nothing.
2.7.22
- A panel on the WordPress Dashboard with the chart «Content consumption» of the plugin’s Dashboard: the daily count of HTML crawling by bots and of people arriving from AI assistants, with the marks of the period. Editors and administrators see it; the link to the plugin’s Dashboard is shown only to those who can open it. Each user chooses 7, 30, 60 or 90 days in the panel, and the choice is kept for them. Like any panel of the Dashboard, it can be hidden in Screen Options.
- Tested on the ZIPs of this version: every earlier battery, all passing, and 125 checks of this release on this edition: the chart’s data, the same as the plugin’s Dashboard got in 2.7.21; who sees the panel (an administrator with the link, an editor without it and with a period of their own, an author not at all, nobody with the plugin deactivated); the four periods with JavaScript and without it, and a period, a nonce or a visitor that must not change anything; a real browser drawing the chart, changing it to 7 days without leaving the page and keeping the choice after a reload; English and Spanish; the uninstaller removing each user’s period. 367 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (182 warnings, as in 2.7.21).
2.7.21
- IP analysis: the introduction lists the levels as they are since 2.7.20, «Verified by signature» first, instead of speaking of a «5-level cascade».
- Dashboard → Activity by bot / source: ChatGPT Agent counts as «User request», like Google-Agent, and «Signed agent» (a signer outside the plugin’s list) as «Other». Until now neither had a function, so they were left out of the reads by function.
- Agent-Readiness → Analyze all templates: a page set as the static front page is analysed once, as «Home», not a second time as a page.
- Spanish translation: the plugin’s name and description on the Plugins screen, and two strings that must not be translated (an identifier pattern and an example address).
- Tested on the ZIPs of this version: every earlier battery, all passing, and 85 checks of this release on this edition: the function of ChatGPT Agent and Signed agent, with real signed visits, in Activity by bot / source and its filters, in English and Spanish; Analyze all templates with a static front page, with the latest posts and with no front page chosen; the introduction of IP analysis; the header on the Plugins screen and the Spanish translation files. 353 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (182 warnings, as in 2.7.20).
2.7.20
- Web Bot Auth: agents that sign their requests are verified by their signature. Some agents sign every request (an IETF draft based on HTTP Message Signatures, RFC 9421): Google-Agent, as
agent.bot.goog, and ChatGPT agent, aschatgpt.com. Anyone can write a User-Agent; only whoever holds the key can make a valid signature. The plugin keeps the three headers (Signature-Agent,Signature-Input,Signature) of every signed visit it logs and checks the signature (Ed25519). A valid one verifies the visit, whatever its address, unless its User-Agent claims to be a bot of another operator: IP analysis has a new level, «Verified by signature». - ChatGPT agent is logged. It uses the User-Agent of an ordinary Chrome, so until now its visits were not logged at all; now they are, by its signature. A request signed by an agent that is not on the plugin’s list is logged as «Signed agent», with its headers, and is not checked.
- IP analysis → Signed requests: who signs, what the check said and why, and the latest requests with their headers.
- The public keys are downloaded only from
agent.bot.googandchatgpt.com, never from an address a visit names, only when a signed visit needs them, and only while the switch of the IP lists is on (off on a new install until you turn it on). A key not downloaded yet does not delay the page: the visit is checked a moment later, by WP-Cron. See External services. - readme: the keys under External services, the headers under What it stores, the cache FAQ (an agent that signs with a browser’s User-Agent cannot be kept out of a page cache), and new screenshots 3 and 5.
- Tested on the ZIPs of this version: every earlier battery, all passing, and 150 checks of this release on this edition: Cloudflare’s published vectors and requests signed by an independent library (pyauth http-message-signatures), what makes a signature invalid or not checked, which keys are asked for and how, real signed requests to the site (ChatGPT agent, Google-Agent, a broken signature, an unknown agent, a key checked later by WP-Cron, the switch off) and the screens in English and Spanish. 343 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (182 warnings, as in 2.7.19).
2.7.19
- Agent-Readiness, until now only in the complete edition, is in this one too (AI Traffic → Agent-Readiness). It analyses the pages of your own site, never another one. Details in the description.
- Agent-Readiness checks the certificate of every address it requests: a site with a self-signed certificate gets an error that says so, not an analysis.
- readme: Agent-Readiness in the description, its requests to your own site under External services, axe-core under Third-party libraries, and a sixth screenshot.
- Tested on the ZIPs of this version: every earlier battery, all passing, and 61 checks of this release on this edition: the screen, its analysis and exports in English and Spanish, an address of another site refused before anything is requested, the certificate checked, and nothing of the complete edition named. The accessibility rules give the same violations as axe-core 4.14 here too (21 checks of 2.7.18 on this edition). 325 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (182 warnings, as in 2.7.18).
2.7.18
- No changes in what this edition does: 2.7.18 fixes a loop in the accessibility rules of the complete edition’s Agent-Readiness screen. Both editions carry the same version number.
- Tested on the ZIPs of this version: every earlier battery, all passing, and 2 checks of this release on this edition (none of it is here: no accessibility rules, no Agent-Readiness). 311 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (182 warnings, as in 2.7.17).
2.7.17
- No changes in what this edition does: 2.7.17 changes how the complete edition’s Coherencia screen judges a home page served from a page cache, and adds one discovery link tag there. Both editions carry the same version number.
- Tested on the ZIPs of this version: every earlier battery, all passing, and 3 checks of this release on this edition (none of it is here: no api-catalog link, neither header nor tag, and no Coherencia screen). 304 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (182 warnings, as in 2.7.16).
2.7.16
- No changes in what this edition does: 2.7.16 changes how the complete edition declares its permissions for AI (content signals, TDM reservation, ai-policy.json). Both editions carry the same version number.
- Tested on the ZIPs of this version: every earlier battery, all passing, and 4 checks of this release on this edition (none of it is here: no TDM reservation header or meta, no content signal in robots.txt). 294 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (182 warnings, as in 2.7.15).
2.7.15
- No changes in what this edition does: 2.7.15 gives a sanitize callback to the content signal settings, which only the complete edition has. Both editions carry the same version number.
- Tested on the ZIPs of this version: every earlier battery, all passing, and 5 checks of this release on this edition: no content signals and none of their settings, and every
register_setting()call with itssanitize_callbackwritten in it. 281 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (182 warnings, as in 2.7.14).
2.7.14
- Changes asked for in the third WordPress.org review:
- The plugin’s tables are declared in the format
dbDelta()reads: the primary key on a line of its own (PRIMARY KEY (id)),KEYfor the indexes and the column types as the database reports them (int(10) unsigned). With the key written next to its column, every reactivation on MariaDB madedbDelta()add the primary key again; the query failed with “Multiple primary key defined”, shown on screen withWP_DEBUG_DISPLAYand written to the log otherwise. The tables keep the same columns and indexes, and existing tables need no change. - Each setting is registered by a
register_setting()call of its own, with its name and itssanitize_callbackwritten in it (includes/settings/core.phpandincludes/settings/llms.php). The callbacks are the same as in 2.7.13. - Perplexity’s two IP lists are downloaded from www.perplexity.ai, the address Perplexity’s documentation gives; www.perplexity.com, used until now, redirects there. A site updated from an earlier version moves its two Perplexity lists to the new addresses; an address changed by hand is left as it is.
- The plugin’s tables are declared in the format
- readme: under External services, Anthropic’s entry links its Privacy Center too, and Perplexity’s lists are given at their new addresses.
- Tested on the ZIPs of this version: every earlier battery, all passing, and 68 checks of this release on this edition. They compare the table statements of 2.7.13 and 2.7.14: the same tables, and nothing for
dbDelta()to change in the old ones, where those of 2.7.13 fail five times with “Multiple primary key defined” on MariaDB. They activate the plugin from Plugins with WP_DEBUG and find no output, no database error and no PHP notice, read everyregister_setting()call of this edition with itssanitize_callback, and check the update of Perplexity’s lists and this readme. 275 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (182 warnings, 179 in 2.7.13: the three new ones are the queries of the update of Perplexity’s lists, on the plugin’s own table).
2.7.13
- Changes asked for in the second WordPress.org review:
- The plugin’s cron events (the daily cleanup of the log and, when it is turned on, the daily download of the IP lists) are scheduled on
initinstead of while the plugins load. Scheduling an event reads the list of cron schedules, and other plugins translate the names of their own schedules there (WooCommerce does). Asked for that early, WordPress 6.7 and later report those translations as loaded too early, which the review saw on the first screen after activation. register_setting()is called with thesanitize_callbackwritten in the call. Each setting keeps the callback its module declares; one without it would be sanitized as text.- The page that shows the progress of a reclassification of the log loads its style from a file and its redirect through the script functions of WordPress, and the help texts of Settings are added to the admin script. No script or style tag is written by hand in the plugin’s code.
- The plugin’s cron events (the daily cleanup of the log and, when it is turned on, the daily download of the IP lists) are scheduled on
- readme: under External services, each service lists the files the plugin downloads from it as links, with its documentation, terms and privacy policy, in the format of the plugin guidelines. Perplexity’s privacy link points to its current address.
- Tested on the ZIPs of this version: every earlier battery, all passing, and 45 checks of this release on this edition. They activate the plugin on WordPress 7.1 with WP_DEBUG next to a plugin that translates the name of its cron schedule as WooCommerce does, and find no notice on activation, the WordPress dashboard, the plugin’s screens or a public page. They also cover the cron events, the settings’ callbacks, the absence of hand-written tags and this readme. 261 deliberate faults, all caught by the tests. Plugin Check 2.1.0: 0 errors (179 warnings, as in 2.7.12).
2.7.12
- No changes in what this edition does: 2.7.12 changes the complete edition’s WebMCP, its files for agents and its audit screens. Both editions carry the same version number. The shared files change only where this edition never reaches: the Dashboard leaves out WebMCP test rows (which only the complete edition writes), and the Spanish translation has the complete edition’s new strings.
- Tested on the ZIPs of this version: every earlier battery, all passing, 5 checks of this release on this edition (none of the new screens or files is here), and Plugin Check 2.1.0 with 0 errors (179 warnings, 175 in 2.7.11: two new queries of the shared Dashboard).
2.7.11
- Changes asked for in the WordPress.org review:
- The debug log (off by default) moves from the root of the uploads folder to its own folder,
uploads/amdt-ai-traffic-analytics/. The folder has an emptyindex.phpand an.htaccessthat denies direct access on servers that read it, such as Apache. A log left at the root is moved there after the update. Uninstalling with “Keep data on uninstall” unticked removes the folder too. - The referral script’s endpoint no longer answers to the action name used before 2.7.0 (
ag_track_referral), whose prefix was not the plugin’s. Pages served since 2.7.0 useamdt_ait_track_referral. - Admin notices only appear on the plugin’s own screens and on Plugins. Until now, seven of them showed on every admin screen: the six that report what an update or the automatic reclassification did, until dismissed, and the one about the rewrite rules.
- Every setting declares its sanitization callback where it is registered; one that did not would be sanitized as text.
- On requests for llms.txt, the names of the query parameters passed to Yoast SEO’s allowlist (so that it does not strip them) are sanitized, and only names made of the characters of a parameter name, up to 64, are passed.
- Request parameters read only to decide what a screen shows (the notice after a redirect, the filters of the log, the window of the IP analysis) are unslashed, and say why they need no nonce.
- Chart.js, bundled for the Dashboard charts, updated from 4.4.4 to 4.6.1.
- The debug log (off by default) moves from the root of the uploads folder to its own folder,
- readme: the External services section lists OpenAI’s
adsbot.json(OAI-AdsBot), which was missing, and the plugin’s daily request for its own llms.txt; the Anthropic privacy link points to its current address; a new section names the bundled library. - Tested on the ZIPs of this version: 78 checks of this release on this edition, every earlier battery, all passing, and Plugin Check 2.1.0 with 0 errors (175 warnings, from 211). The checks cover the debug log’s folder, also on a real Apache 2.4, uninstall, the old action name, the settings’ callbacks, the notices, Chart.js and this readme.
2.7.10
- No changes in this edition: 2.7.10 only changes the complete edition’s AIO analysis. Both editions carry the same version number.
2.7.9
- No changes in this edition: 2.7.9 only adds the complete edition’s A2A interface and rewrites its agent card. Both editions carry the same version number.
2.7.8
- llms.txt can say what your site is for: a new setting, “What the site is for”, one use per line (up to 8, of 200 characters each). It is published as a “When to use this site” section under the description. Left empty, as it comes, nothing is published: the plugin does not make it up. With WPML it is translated in String Translation. When an SEO plugin serves llms.txt, the section is added to its file.
- WebMCP detections, which only the complete edition records, are no longer counted as consumption on the Dashboard, and the rows it wrote are relabelled from “(agéntico)” to “(WebMCP)” once, on update. In a site that only ever had this edition, nothing changes.
- Tested on the ZIP of this version: the “When to use this site” section with and without text, WordPress’s redirects for every client, and Plugin Check 2.1.0 with 0 errors.
2.7.7
- No changes in this edition: 2.7.7 only changes the complete edition’s MCP server, its WebMCP tools and its files for AI agents. Both editions carry the same version number.
2.7.6
- llms.txt leaves out the entries your SEO plugin marks noindex: Yoast SEO, Rank Math and All in One SEO (the entry’s own setting, or else that of its content type and, in Rank Math and All in One SEO, the global one), and SEOPress (the entry’s setting). An SEO plugin counts only while it is active. WordPress’s “Discourage search engines” setting does not count.
- The products section of llms.txt now respects the “Exclude from the AI surfaces” box, like the other sections.
- The “AI access” box in the editor says when the SEO plugin already leaves the entry out.
2.7.5
- No changes in this edition: 2.7.5 only adds checks to the complete edition’s content analysis. Both editions carry the same version number.
2.7.4
- Marks on the Dashboard’s chart: dated changes of the AI ecosystem that can move the figures (model changes in ChatGPT and Gemini, Search Console’s AI report, Cloudflare’s new defaults, Google’s spam update), each with its reach and its source, and marks of your own that you can add and delete. The list comes with the plugin; no server is asked for it.
- A notice when an operator’s agent that reads for people (ChatGPT-User, Claude-User, Perplexity-User…) stops reading while its search crawler keeps reading, after reading every week of the three before: the trace of a block on the way, such as Cloudflare’s new defaults. It says when the site seems to be behind Cloudflare, and it can be hidden for 7 days.
2.7.3
- Every bot in the catalog says what it reads for (search, a person’s request, training, ads, or a mix), and the Dashboard adds up the reads by function and filters by one. Google-Extended and Applebot-Extended are switched off and marked as robots.txt tokens: no crawler sends them. Applebot and OAI-AdsBot are added.
- Visits from AI assistants: copilot.com, deepseek.com and meta.ai are added, and a visit without a Referer but with the utm_source=chatgpt.com that ChatGPT adds to its links counts as ChatGPT.
- Settings → Diagnostics checks once a day that /llms.txt answers as Lighthouse expects (a title, a link and 50 characters), and the Dashboard warns when it does not.
- A blocking snippet never includes the range of an operator with search or user agents (OpenAI, Anthropic, Perplexity, Google, Microsoft, Apple, Meta, Amazon, DuckDuckGo), and one made for such an operator starts with a warning.
- Help texts that claimed more than is known were corrected: llms.txt, the help of three bots, and the one of human visits from assistants.
2.7.2
- The notice about a cache in front of WordPress names the header that gives it away and says what to do in that case: Cloudflare keeping the HTML, the nginx cache of the hosting (with where Plesk turns it off), or another hosting or CDN cache. It has a “Check again” button: after that cache was turned off, the notice used to stay for up to 12 hours.
- “More details in Settings” now goes to the panel of that cache.
2.7.1
- Visits from AI assistants are counted even when WP Rocket (Delay JavaScript execution) or LiteSpeed Cache (Load JS Deferred: Delayed) hold scripts back until the visitor touches the page: the plugin lets its own small script through, and nothing else of theirs is touched.
2.7.0
- First public release.
