The AI calls never reach the browser. WordPress 7.0 ships on May 20, 2026 — seven days from today — with the WP AI Client built into core, letting every installed plugin call OpenAI, Anthropic, or Gemini through a single shared credential. Every one of those calls originates from PHP inside the WordPress server. GA4 doesn’t see it. Meta Pixel doesn’t see it. GTM doesn’t see it. The plugin runs a prompt, the model returns a response, and your browser-based analytics stack — the stack that powers most stores’ conversion attribution — is structurally blind to the entire transaction.
What Ships on May 20 — and Why It Was Delayed
WordPress 7.0 ships on May 20, 2026 after a six-week delay from the original April 9 release date (Make WordPress Core). The slip was deliberate. The core team used the extra weeks to harden two systems: the real-time collaboration database, and the AI surface — specifically, the WP AI Client.
The new client is a Settings → Connectors layer. The site owner configures one set of credentials. Every compatible plugin on the site uses those credentials automatically. WordPress 7.0 natively supports three AI providers at launch: OpenAI, Anthropic (Claude), and Google (Gemini) (BigCloudy). One provider is selected by default; plugins can request a different one if needed.
That single configuration step changes the plugin economy. Today, every plugin that uses AI has to ship its own provider integration, manage its own keys, and pass its own rate-limiting logic. After May 20, plugins call into the WP AI Client and inherit the site’s choice. Product description writers, abandoned-cart copywriters, image-alt generators, chatbots, customer-service agents — all of them flowing through the same connector layer, on the same credentials.
The Scale of What Just Became Default
WordPress runs on 43.5% of all websites (W3Techs, 2024). The WP AI Client becomes the largest single distribution channel for plugin-mediated AI calls on the open web overnight. Add the hyperscaler context: combined 2026 AI capex across Alphabet, Amazon, Microsoft, and Meta sits at $650–700B (Q1 2026 earnings, April 29). The AI-provider call layer is becoming a primary cost centre. WordPress sites running the WP AI Client will be a meaningful share of that traffic — with no built-in cost monitoring of any kind.
You may be interested in: WordPress 7.0 Just Gave Every Plugin a Public API for AI Agents
Two Layers Just Opened, and Browser Tracking Watches Neither
The WP AI Client opens two new data layers at the same time — and the standard analytics stack is structurally blind to both.
Layer 1: The AI cost meter. Every prompt produces a token cost. Every response produces a token cost. Multiply that across product description writers, customer-service agents, alt-text generators, and abandoned-cart copywriters running across all the plugins a typical WordPress site stacks. The bill arrives monthly from each provider. The breakdown — which plugin spent what, on which content, at what model — does not exist in the WordPress admin and does not appear in any browser analytics.
Layer 2: The AI attribution gap. A customer-service chatbot built on Claude resolves a return-shipping question. An abandoned-cart agent built on OpenAI sends a personalised recovery email. A product description writer generates copy that pushes conversion rate up by three percentage points. None of those interactions fire a Pixel event. GA4’s enhanced measurement watches scroll, click, and view in the browser — it has no view into a PHP-side OpenAI call. The AI-influenced revenue stays attributed to whatever channel the user arrived from, with the actual AI assist invisible in the funnel.
Both layers happen in the same place — server-side, inside WordPress, outside the browser. The fix has to live there too.
The Architectural Answer — Capture the Telemetry Where It Happens
If the AI call originates in PHP, the telemetry has to be captured in PHP. Every WP AI Client call carries a recoverable set of fields: model, provider, plugin caller (which plugin made the call), prompt token count, completion token count, latency, response status, and request fingerprint. A wrapper or filter sitting on the WP AI Client surface can read those fields on every call and emit a structured event.
That event has the same shape as a conversion event. It needs the same routing: enriched, batched, and sent to a first-party server that forwards it to BigQuery for analysis. The natural sink for AI-call telemetry is the same data warehouse that already holds your order events, so plugin-level AI spend can be reconciled against conversion outcomes per-product, per-campaign, per-customer.
You may be interested in: WordPress to BigQuery Without a Developer
Three Data Surfaces, One Pipeline
| Surface | Where it fires | Browser tracking | Server-side pipeline |
|---|---|---|---|
| Page view / click | Browser JS | Captured | Captured (via hook) |
| Order completion | PHP hook | Partial (front-end only) | Captured |
| WP AI Client call | PHP | None | Captured (via wrapper) |
The column that sees all three rows is the rightmost one. Browser tracking covers the top row well, the middle row partially, and the bottom row not at all. A server-side first-party pipeline is the only architecture that captures every surface a 2026 WordPress site actually produces.
Here’s How You Actually Do This
Transmute Engine™ is a first-party Node.js server that runs on your subdomain (e.g., data.yourstore.com). The inPIPE WordPress plugin already hooks into woocommerce_payment_complete for conversion events; the same plugin architecture extends naturally to a WP AI Client wrapper that captures model, provider, plugin caller, tokens, and latency on every AI call. Those events stream into the same first-party pipeline as your order events, land in BigQuery, and reconcile against provider invoices and conversion outcomes. The AI spend layer and the attribution layer arrive in the same warehouse, on the same schema, ready for analysis.
The argument for capturing this from day one is straightforward: AI usage is data you will wish you had captured later. If your customer-service chatbot starts producing measurable LTV lift, you need the telemetry to prove it. If a plugin starts spending $4,000 a month in tokens for output nobody reads, you need the telemetry to find it. Running a private AI endpoint behind the WP AI Client only sharpens the case — the cost is yours, the usage data should be too.
Key Takeaways
- May 20 is the inflection date. WordPress 7.0 ships the WP AI Client to 43.5% of websites, on PHP 7.2+ (8.3 recommended). Every plugin becomes a potential AI caller.
- Three providers, one credential. OpenAI, Anthropic (Claude), and Google (Gemini) are natively supported at launch, configured once under Settings → Connectors.
- Browser tracking is structurally blind. Every WP AI Client call originates from PHP. GA4, Meta Pixel, GTM, and enhanced measurement see none of it.
- Two layers need capturing. Cost (tokens per plugin per model) and attribution (AI-influenced revenue). Both live server-side.
- The fix is the same pipeline as order events. Capture AI-call telemetry at the PHP wrapper, batch it, route it server-side to BigQuery alongside conversion data.
Frequently Asked Questions
The WP AI Client is a native WordPress 7.0 connector layer that lets any installed plugin call OpenAI, Anthropic, or Google Gemini through a single shared credential configured under Settings → Connectors. It ships May 20, 2026. Before WP 7.0, every plugin that needed AI had to manage its own API keys, rate limiting, and provider switching. After 7.0, plugins call into the WP AI Client and the site owner controls provider, model, and credentials in one place.
WordPress 7.0 does not ship a built-in cost or usage monitor. The WP AI Client routes calls but does not aggregate observability. To monitor per-plugin token usage you need to capture telemetry at the server-side wrapper layer — model, provider, plugin caller, prompt tokens, completion tokens, latency, and response status — and route it to a data sink like BigQuery for cost reconciliation against your provider invoices.
No. Every WP AI Client call originates from PHP inside the WordPress server and never reaches the browser. GA4, Meta Pixel, GTM, and every other JavaScript-based analytics tool is structurally blind to it. The AI usage layer and the conversion attribution layer both happen server-side, which means browser tracking captures neither without a dedicated server-side pipeline.
Three at launch: OpenAI (GPT models), Anthropic (Claude models), and Google (Gemini models). Site owners configure their preferred provider once through Settings → Connectors, and every WP AI Client–compatible plugin uses those credentials automatically. Private or self-hosted AI endpoints are not in the native list but can be wrapped through a custom connector.
The original April 9 release date was pushed back six weeks specifically to harden the AI surface and the real-time collaboration database. The WP AI Client is one of the two headline systems in 7.0, and the core team used the additional time to finish security and stability work on it before shipping to the wider ecosystem.
WordPress 7.0 lands in seven days. The AI calls start the moment your plugins update. Plant the data tree now — capture the telemetry from day one, or rebuild the trail from invoices later.
