WordPress 7.0 introduces one of the most exciting features in years — the AI Client is now part of Core directly, not just another third-party Plugin.
Before this, adding AI features to a Plugin meant doing everything yourself — connecting to each provider's API, managing API keys for clients, handling errors differently every time. Every Plugin did it their own way with no shared standard.
WordPress 7.0 changes all of that by introducing a central PHP API that lets a Plugin simply declare what it needs, then WordPress handles routing the request to whichever AI Model is available — whether the site is running Claude, GPT, or Gemini.
Table of Contents
Getting Started From Scratch
First, you need to connect an AI to your website. WordPress 7.0 provides a built-in UI that lets you choose a provider based on the API key you have—whether that's Anthropic, Google, or OpenAI. Simply navigate to Settings > Connectors, enter your API key, and you're ready to go!
The most important thing to do before calling AI is checking whether that site has a Provider configured, because not every site will have one set up.
$builder = wp_ai_client_prompt( 'test' );
if ( ! $builder->is_supported_for_text_generation() ) {
// No provider configured — show a notice or hide the UI
return;
}
// Reaching this line means we're good to go
Generating Text
Everything starts with a single function — wp_ai_client_prompt(), which is a Fluent Builder similar to the Query Builder pattern familiar from Laravel. Methods can be chained one after another.
$text = wp_ai_client_prompt( 'Write a Meta Description for an article about WordPress Caching' )
->using_system_instruction( 'You are an SEO Expert specializing in WordPress.' )
->using_temperature( 0.7 ) // 0 = precise, 1 = creative
->using_max_tokens( 200 )
->generate_text();
if ( is_wp_error( $text ) ) {
error_log( $text->get_error_message() );
return;
}
echo wp_kses_post( $text );
Getting JSON Back With a Schema
Instead of having AI return plain text that you then have to parse yourself, you can define a JSON Schema upfront to specify exactly what shape you want the data in. This is perfect for anything that needs to be rendered on a page or stored in a database.
// กำหนดหน้าตาของข้อมูลที่ต้องการ
$schema = array(
'type' => 'array',
'items' => array(
'type' => 'object',
'properties' => array(
'title' => array( 'type' => 'string' ),
'description' => array( 'type' => 'string' ),
'category' => array( 'type' => 'string' ),
'difficulty' => array(
'type' => 'string',
'enum' => array( 'beginner', 'intermediate', 'advanced' ),
),
),
'required' => array( 'title', 'description', 'category', 'difficulty' ),
),
);
$json = wp_ai_client_prompt( 'แนะนำ 5 หัวข้อบทความสำหรับเว็บ WordPress Developer ภาษาไทย' )
->as_json_response( $schema )
->generate_text();
if ( is_wp_error( $json ) ) {
return;
}
// ได้ Array พร้อมใช้เลย ไม่ต้อง Parse เอง
$topics = json_decode( $json, true );
foreach ( $topics as $topic ) {
echo '<h3>' . esc_html( $topic['title'] ) . '</h3>';
echo '<p>' . esc_html( $topic['description'] ) . '</p>';
echo '<small>หมวด: ' . esc_html( $topic['category'] ) . ' | ระดับ: ' . esc_html( $topic['difficulty'] ) . '</small>';
}
Generating Images
Before generating images, it's worth checking whether the configured Provider actually supports image generation, since not every Provider can do it.
$builder = wp_ai_client_prompt( 'test' );
if ( ! $builder->is_supported_for_image_generation() ) {
echo 'Provider ที่ตั้งค่าไว้ไม่รองรับการสร้างรูปภาพ';
return;
}
Once confirmed, generating an image is straightforward:
use WordPress\AiClient\Files\Enums\FileTypeEnum;
use WordPress\AiClient\Files\Enums\MediaOrientationEnum;
$image_file = wp_ai_client_prompt( 'ภาพ Hero Banner สำหรับเว็บ Tech Blog สไตล์ Minimal สีน้ำเงินเข้ม' )
->as_output_file_type( FileTypeEnum::inline() )
->as_output_media_orientation( MediaOrientationEnum::from( 'landscape' ) )
->generate_image();
if ( is_wp_error( $image_file ) ) {
return;
}
echo '<img src="' . esc_url( $image_file->getDataUri() ) . '" alt="Hero Banner">';
If you want to give clients multiple variations to choose from, you can request several at once:
$images = wp_ai_client_prompt( 'โลโก้สำหรับร้านกาแฟสไตล์ Minimalist พื้นหลังขาว' )
->as_output_media_orientation( MediaOrientationEnum::from( 'square' ) )
->generate_images( 4 ); // ขอมา 4 แบบพร้อมกัน
if ( is_wp_error( $images ) ) {
return;
}
echo '<div style="display: grid; grid-template-columns: repeat(2, 1fr); gap: 1rem;">';
foreach ( $images as $index => $image ) {
echo '<div>';
echo '<img src="' . esc_url( $image->getDataUri() ) . '" style="width: 100%;">';
echo '<button onclick="selectImage(' . $index . ')">เลือกแบบนี้</button>';
echo '</div>';
}
echo '</div>';
Choosing a Model With Fallback
If a Plugin needs a specific Model, you can define a fallback order and WordPress will pick the first one available on that site. If none of them are available, it automatically falls back to the most suitable Model it can find.
$result = wp_ai_client_prompt( 'วิเคราะห์ Tone ของบทความนี้และแนะนำวิธีปรับปรุง' )
->using_model_preference(
'claude-sonnet-4-6', // อยากได้ตัวนี้มากที่สุด
'gemini-3.1-pro-preview', // ถ้าไม่มีก็ตัวนี้
'gpt-5.4' // หรืออันนี้ก็ได้
)
->generate_text_result();
if ( is_wp_error( $result ) ) {
return;
}
// ดูได้ว่า Model และ Provider ไหนตอบกลับมาจริงๆ
$model = $result->getModelMetadata();
$provider = $result->getProviderMetadata();
$tokens = $result->getTokenUsage(); // ดู Token ที่ใช้ไป
error_log( 'ใช้ Model: ' . $model->getName() . ' จาก ' . $provider->getName() );
error_log( 'Token ที่ใช้: ' . $tokens->getTotal() );
Using generate_text_result() instead of generate_text() gives you this extra metadata, which is extremely useful for debugging or tracking AI usage costs across a site.
Connecting to the REST API
For anything that needs to be called from JavaScript on the front-end, the recommended approach is to create a dedicated REST Endpoint and manage permissions on the PHP side. Letting JS call AI directly is not advisable because access control becomes much harder to enforce.
add_action( 'rest_api_init', function () {
register_rest_route( 'myplugin/v1', '/generate-description', array(
'methods' => 'POST',
'callback' => 'myplugin_generate_description',
'permission_callback' => function () {
return current_user_can( 'edit_posts' );
},
) );
} );
function myplugin_generate_description( WP_REST_Request $request ) {
$product_name = sanitize_text_field( $request->get_param( 'product_name' ) );
$result = wp_ai_client_prompt(
sprintf( 'Write a product description for "%s" in approximately 100 words.', $product_name )
)
->using_system_instruction( 'You are a Copywriter specializing in E-commerce.' )
->using_temperature( 0.8 )
->generate_text_result();
return rest_ensure_response( $result );
}
Then on the JavaScript side, it's just a simple fetch:
const response = await fetch( '/wp-json/myplugin/v1/generate-description', {
method: 'POST',
headers: {
'Content-Type': 'application/json',
'X-WP-Nonce': wpApiSettings.nonce,
},
body: JSON.stringify({ product_name: 'Handmade Genuine Leather Bag' }),
} );
const data = await response.json();
console.log( data );
This isn't just a new feature being added to WordPress — it fundamentally changes how developers work with AI inside the platform. Everyone now speaks the same language, uses the same API, and site owners manage everything from one place. No more juggling API keys across a dozen different Plugins.
Source: WordPress Core Dev Notes — Introducing the AI Client in WordPress 7.0 Introducing the AI Client in WordPress 7.0
