Free
$0on WordPress.org
- Attribution and dashboard, 30 days of history
- llms.txt and llms-full.txt
- Crawler log with citations
- Monthly digest email
- Schema enrichment and catalog feed
- Agent guard rails
Assistants send buyers to your shop today, and your analytics files most of them under Direct. This plugin counts that revenue, makes the catalog readable for agents, and holds every agent write until you say yes.
70.6% of AI referrals arrive with no referrerZero external calls in the shopper request path
Order+$183.81
Assistants open links in a context that strips the referrer. The visit lands in your reports as if the buyer typed a 60 character product URL by hand, so the channel that produced it never shows up in any report you already have.
70.6%
of AI referrals arrive with no referrer at all, measured across 446k visits by The Digital Bloom
26.5%
of the reference store's AI sessions had to be recovered rather than read off a referrer
$229.80
revenue in one month that no report attributed to any channel
Counts
First party attribution, recorded on your own server and joined to WooCommerce orders. The panel below reads the same response shape the plugin serves in wp-admin, filled with the plugin's own test fixture.
| Source | Revenue | Share | Orders | Sessions | Conv. | AOV |
|---|---|---|---|---|---|---|
| ChatGPT | $1,286.70 | 50.5% | 7 | 209 | 3.35% | $183.81 |
| Perplexity | $572.00 | 22.5% | 3 | 92 | 3.26% | $190.66 |
| Claude | $532.90 | 20.9% | 3 | 39 | 7.69% | $177.63 |
| Gemini | $32.50 | 1.3% | 1 | 22 | 4.55% | $32.50 |
| Copilot | $54.00 | 2.1% | 1 | 11 | 9.09% | $54.00 |
| Other AI | $67.40 | 2.6% | 1 | 27 | 3.7% | $67.40 |
| Proven by referrer or campaign tag | $2,315.70 | $229.80 more was recovered from visits that arrived with nothing attached | ||||
Daily AI revenue, America/New_York106 sessions (26.5%) would have been filed as Direct
Sells
An assistant can only recommend what it can parse and only close what it can call. The plugin publishes the machine surface, keeps it in sync with the catalog, and shows you who is actually reading it.
88%agent readiness, 7 of 8 checks passing
Training crawlers can be turned away without losing a sale. The assistant fetches cannot: those happen while somebody is waiting for an answer.
| Crawler | Vendor | Purpose | Hits | Address check |
|---|---|---|---|---|
| GPTBot | OpenAI | Training | 4,187 | verified |
| ChatGPT-User | OpenAI | Live user fetch | 612 | verified |
| OAI-SearchBot | OpenAI | Search index | 933 | verified |
| ClaudeBot | Anthropic | Training | 1,504 | verified |
| Claude-User | Anthropic | Live user fetch | 208 | verified |
| PerplexityBot | Perplexity | Search index | 877 | verified |
| Perplexity-User | Perplexity | Live user fetch | 143 | unverified |
| Bytespider | ByteDance | Training | 2,291 | unverified |
| CCBot | Common Crawl | Training | 318 | unverified |
| GPTBot | OpenAI | Training | 46 | spoofed |
A user agent string costs nothing to forge, so a claimed crawler is checked against the address ranges its vendor publishes. Unverified means the vendor publishes none, which is not the same accusation as spoofed.
Guards
WooCommerce ships an agent API in core, which means the write side is already open on millions of stores. Every ability runs behind a scope check, a risk rule and an append only journal.
Trailhead Merino Crew
$163.59$0.00
price-dropPrice cut is larger than 30% of the current price.
Each agent gets its own key with a written list of what it may touch. A call outside that list never reaches WooCommerce.
Scopes answer whether an agent may touch prices at all. Risk rules answer whether this particular change is the one you meant.
A write ability that cannot be reversed is not registered at all, so no configuration mistake can let one through.
Deny the held price change and roll back the discount the same agent applied a minute earlier. Two entries get appended below, and their hashes chain onto the last one.
Trailhead Merino Crew$163.59
Denied, price restored from the snapshot.
Nothing here is a mockup: the hashes were produced at build time by the same routine the plugin runs, and a test rebuilds them on every commit.
The artifact owners forward to their agency. It is assembled from the same figures shown above.
AI brought you $2,545.50
Orders attributed to assistants16
Best source, ChatGPT$1,286.70
Sessions recovered from Direct106
Agent readiness88%
Agent actions waiting for you1
Your data stays on your server. The digest is rendered from figures your store computed, and the plugin sends nothing else.
Start with the free plugin from the WordPress directory. Upgrade when the number it shows you is worth more than the subscription.
$0on WordPress.org
$19per month
$39per month
Annual billing takes 20% off. Prices are set for the first cohort and will be reviewed after the first fifty paying stores.
Visit capture writes to a buffer and flushes in the background through Action Scheduler. The budget for a storefront hit is two milliseconds, and a performance test in CI fails the build if it is exceeded. Nothing in the shopper request path calls an external service.
Your licence check, and opt in anonymous telemetry if you enable it. Order data, sessions and the journal stay in your own database. The digest email is rendered from figures your store computed.
GA4 files most assistant traffic under Direct because the referrer is stripped before the browser arrives. Attribution here happens on your server, where the request actually lands, and it joins to WooCommerce orders rather than to page views.
WooCommerce core exposes agent abilities on its own, so the surface exists whether or not you added anything. The guard is what decides which key may call them and what happens before a write lands.
Every write takes a snapshot of the record first and appends a journal line whose hash chains onto the previous one. Undo replays the snapshot, and the rollback is itself journalled. An ability that cannot be undone is never registered in the first place.
No. The plugin reads your store and writes to your store. It also lets you decide, per crawler, which vendors may read the catalog at all, and it separates training crawlers from the assistant fetches that bring buyers.