justinholtweb / craft-sanka

5.0.1 2026-08-28 10:19 UTC

Requires

• php: ^8.2
• ext-json: *
• ext-mbstring: *
• ext-openssl: *
• craftcms/cms: ^5.3.0

proprietary

• Justin Holt

• geo
• seo
• indexing
• craftcms
• craft-plugin
• indexnow
• llms-txt
• generative-engine-optimization
• google-indexing-api
• ai-crawlers

Published on Packagist.org on 2026-08-28 10:20 UTC

README

Instant indexing and GEO for Craft CMS. Push URLs to Google's Indexing API and IndexNow the moment they change, then manage, log and audit how AI search engines read your site.

Named for the instant coffee, which is the joke — and which also names the setting you should leave on for the first week: Sanka runs decaffeinated by default, recording exactly what it would have sent while nothing at all leaves the server.

Requires Craft CMS 5.3+ and PHP 8.2+.

The problem

Publishing is fast. Being found is not. A new page waits for a crawler that turns up on its own schedule — days, sometimes weeks — and for the answer engines it is often not a wait at all, because they never come.

Sanka closes both gaps, and they are genuinely two different jobs.

Indexing is telling a search engine that a URL changed, instead of waiting to be asked. Google has an API for it. Bing, Yandex, Seznam, Naver and Yep share one between them, called IndexNow.

GEO — generative engine optimisation — is the other half. Answer engines do not crawl like search engines and are not managed like them, and almost nobody knows which of the twenty-odd "AI bots" in their logs is which. Sanka makes that explicit: who may crawl, what they are given, who actually turned up, and whether a page is written in a shape an engine can quote.

What it does

Submits, the moment something changes

Save an entry and the URL is queued. Publish, update, unpublish, delete — each becomes the right kind of notification, to the engines you chose, for the sections you chose.

• Google Indexing API — URL_UPDATED and URL_DELETED, service-account auth, real multipart/mixed batching (100 URLs in one round trip), and quota tracking against the 200-a-day default that resets at midnight Pacific.
• IndexNow — one submission reaching Bing, Yandex, Seznam, Naver and Yep. Bing's index is what ChatGPT Search and Copilot answer from, so this is the shortest path from published to an answer engine knows it exists. The key file is served from a route, so there is nothing to upload and nothing to redeploy when the key changes.
• Sitemap resubmission (Pro) — one call that can move thousands of pages, which is what you want after a migration rather than spending a day's quota one URL at a time.

Never spends quota by accident

Google gives you two hundred submissions a day. Spending them by accident is the normal failure mode of every tool in this category, so three things are built in rather than bolted on:

• Quota is checked before the call, not after the failure. The batch is cut to fit what is left.
• A save storm costs one submission, not twenty. A cooldown keyed on URL + engine + type absorbs an author saving repeatedly while they edit.
• Dry run is on by default and records what would have been sent.

Says why, every time

Every submission is a row in a ledger, and skipped is a first-class outcome with a message on it — not a silent drop. "Why didn't it send my page?" always has an answer on the screen: the cooldown, the quota, the dry run, a rule that did not match, a URL no engine could reach.

Failures are translated rather than reported. A 403 from Google says add the service account as an owner of the property in Search Console. A 403 from IndexNow says the engine could not read your key file. A 422 says the site's configured base URL does not match the domain it is served on.

Manages the AI crawlers (Pro)

Twenty-two agents, each labelled with what blocking it actually costs, because "block the AI bots" is advice that destroys traffic when followed literally:

Answer engines	OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot, DuckAssistBot, YouBot, Applebot, Amazonbot — blocking one removes you from that engine's answers, and does nothing about training.
User requests	ChatGPT-User, Perplexity-User, Claude-User — a person just asked about this page. Blocking one tells that reader it could not be read.
Model training	GPTBot, Google-Extended, Applebot-Extended, CCBot, Bytespider, meta-externalagent, and more — blocking these costs no visibility. This is the group most people actually mean.

Output is robots.txt, which Sanka will serve for you or hand to your own template as craft.sanka.robots() — with any part you already generate elsewhere left out, so a site running SEOmatic can take the crawler policy and nothing else.

Tells you who actually came (Pro)

Every request from a recognised AI agent is recorded — which one, what it read, what it got. "Is ChatGPT reading my site, and what is it reading?" stops being a guess.

Rows say something claimed to be this agent, because a user agent is one line of curl to forge. php craft sanka/crawlers/verify does forward-confirmed reverse DNS in batches and marks each one genuine, forged, or unverifiable — three different claims, kept apart.

Publishes llms.txt (Pro)

A sitemap tells a crawler what exists. llms.txt tells a model what the site is about and which pages are worth reading — a curated map, in Markdown, at a fixed path, that a model can take in one fetch instead of crawling a thousand URLs to find the six that matter. /llms-full.txt is the same map with the text inlined, for sites whose whole value is the writing.

Both are generated from Craft's own content and cached, so neither can drift from what the site says.

Audits whether a page is quotable (Pro)

Twenty deterministic checks across four dimensions, scored, remediation-first:

• Structure — one H1, a sane heading outline, an answer-first opening, passage-sized paragraphs, lists and tables an engine can lift whole.
• Attribution — author, datePublished, dateModified, publisher.
• Machine readability — JSON-LD present and of a useful @type, question-and-answer shapes, canonical, no accidental noindex.
• Citability — concrete numbers, outbound sources, definitional sentences.

Nothing is sent anywhere and nothing is scored by a model, so running it twice gives the same answer. The HTML comes from a same-origin fetch of a URL derived from Craft's own site config, or from HTML you paste in — a URL typed by a user is never fetched.

Getting started

composer require justinholtweb/craft-sanka
php craft plugin/install sanka

Then, in Settings → Sanka:

1. Leave dry run on.
2. IndexNow — turn it on and generate a key. That is the whole setup; Sanka serves the key file.
3. Google — create a service account in a Google Cloud project with the Indexing API enabled, paste its JSON key, and then — the step everybody misses — add the service account's client email as an owner of the property in Search Console. A 403 means you have not done this.
4. Add a rule. Sanka ships with auto-submit on and no rules, so it never starts spending quota on its own. A rule says: this section, these events, these engines.
5. Run php craft sanka/diagnostics. It checks everything that could be wrong, in the order it would break.
6. Watch the Submissions screen for a week, then turn dry run off.

Cron

* * * * * php craft sanka/queue/drain          # send what is waiting
0 3 * * * php craft sanka/crawlers/verify      # confirm which crawlers were genuine
0 4 * * * php craft sanka/queue/sweep          # Pro: catch anything the save hook could not see

The drain also runs automatically as a queue job after a save, so the cron entry is a safety net for sites whose queue runner is not guaranteed.

Console commands

Command	What it does
sanka/diagnostics	Everything that could be wrong, in the order it would break
sanka/queue/status	What is waiting, and what would stop it going out
sanka/queue/drain	Send it
sanka/queue/retry-failed	Put every failed row back in the queue
sanka/queue/sweep	(Pro) Queue anything that changed recently and was not submitted
sanka/submit/url <url>	Submit one URL
sanka/submit/section <handle>	(Pro) Submit every live entry in a section
sanka/submit/sitemaps	(Pro) Resubmit the sitemaps
sanka/submit/status <url>	Ask Google what it last heard about a URL
sanka/crawlers/summary	(Pro) Which AI agents have been reading the site
sanka/crawlers/verify	(Pro) Forward-confirmed reverse DNS over unchecked visits
sanka/crawlers/robots	(Pro) Print the robots.txt Sanka would serve
sanka/geo/llms	(Pro) Print llms.txt
sanka/geo/audit <url>	(Pro) Audit one page
sanka/geo/section <handle>	(Pro) Audit a section, worst first

Twig

{{ craft.sanka.robots() }}            {# the crawler policy, for your own robots.twig #}
{{ craft.sanka.robots({ default: false, sitemaps: false, cp: false, header: false }) }}
                                      {# just the AI crawler groups, for an SEOmatic robots.txt #}
{{ craft.sanka.llms() }}              {# llms.txt, if you would rather render it yourself #}
{{ craft.sanka.llms(null, true) }}    {# llms-full.txt #}
{{ craft.sanka.indexNowKeyUrl() }}

{% if craft.sanka.isCrawler() %}…{% endif %}
{% set agent = craft.sanka.crawler() %}
{% for row in craft.sanka.history(entry) %}{{ row.engine }} — {{ row.status }}{% endfor %}

Serving an AI agent different content is cloaking and every engine here penalises it. Measurement, and dropping a cookie banner a machine cannot dismiss, are fine.

Editions

	Lite	Pro
Price	Free	$129, $99/year renewal
Google Indexing API	●	●
IndexNow	●	●
Submit console, submit on save, the ledger, quota, retries	●	●
Auto-submit rules	3	unlimited
Sitemap resubmission		●
llms.txt and llms-full.txt		●
AI crawler policy and robots.txt		●
AI crawler visit log and verification		●
GEO readiness audit		●
Scheduled sweeps, bulk submit, CSV export		●

Lite is not a trial — it does the whole indexing job, for nothing, forever.

Lite refuses to save a configuration it cannot run, so you are told rather than quietly given something else. It downgrades on read, so a lapsed Pro licence keeps indexing with the Lite feature set instead of breaking, and upgrading restores exactly what was configured.

Things worth knowing

• Google documents the Indexing API as being for JobPosting and BroadcastEvent pages. In practice it is used far more widely and it works. Sanka says so on the settings screen rather than pretending in either direction; it is your call to make knowingly.
• The 200-a-day quota is per Google Cloud project, not per site. Two Craft installs sharing one key share the two hundred.
• Google retired its sitemap ping endpoint in 2023. Any tool still offering it is selling a no-op. Sanka's sitemap engine goes over IndexNow, which still works.
• Google does not participate in IndexNow, and IndexNow's engines have no Google-style quota — only a rate limit.
• A static robots.txt in your web root beats anything Craft routes. Sanka detects one and says so, rather than letting a setting appear to work and do nothing.
• Nothing on a development domain will ever be submitted. .ddev.site, .test, localhost and private addresses are refused with a reason, because every engine would refuse them too. There is a switch to override that for testing.

Testing

Integration checks run inside a Craft install:

php /path/to/craft-sanka/tests/integration/checks.php

162 checks, self-cleaning, everything prefixed sanka-check-. Every engine is exercised through a fake transport, so the whole submission path — dispatcher, ledger, quota, batching, backoff, result interpretation — is covered without a single outbound request.

License

This plugin is released under the Craft License.