Lesson 3 introduced the general idea that AI systems interacting with people, or producing content people consume, tend to carry a transparency expectation: let people know they’re dealing with AI. This lesson treats that as a practical design problem rather than an abstract one. Disclosure is worth doing well beyond whatever a specific regulation strictly requires, because trust, once a reader discovers undisclosed AI-generated content, is expensive to rebuild and cheap to preserve with a simple label.
Lesson 3’s general overview of the EU AI Act’s transparency tier. This lesson assumes you’re comfortable adding a custom field or meta box to a WordPress post type.
Whether a specific disclosure requirement applies to a specific feature, and exactly what form it must take, is a legal question this lesson doesn’t answer. What follows is general, practical guidance on disclosure as good practice, consistent with widely understood transparency principles, not a compliance checklist guaranteed to satisfy any particular law. Confirm actual requirements with legal counsel.
Step 1: The general principle, stated plainly
If a person is talking to an AI chatbot, they should be able to tell it’s a chatbot. If a person is reading an article that was substantially written or drafted by AI, that’s generally worth disclosing too, even where a legal disclosure requirement is unclear or doesn’t strictly apply. The underlying reasoning holds regardless of the specific legal answer: people generally weigh AI-generated content differently than human-authored content, and letting them make that judgment honestly, with accurate information, is a reasonable baseline for any AI feature that touches public-facing content or conversation.
Step 2: Why this matters beyond strict legal minimums
The trust argument is independent of the legal one, and arguably stronger in practice. Readers who discover undisclosed AI-generated content after the fact don’t distinguish between “technically not required to disclose” and “chose not to.” Both read the same way: something was hidden. That reaction attaches not just to the specific piece of content but to the site’s credibility generally, and it’s disproportionate to the actual quality of the content itself, a genuinely well-written AI-assisted article can still produce a strong negative reaction once undisclosed AI involvement comes to light. Treating disclosure as a trust investment, not a compliance checkbox, tends to produce better decisions than treating it as a minimum bar to clear.
A site that discloses AI involvement consistently and matter-of-factly has a much easier time explaining an occasional AI mistake than a site whose readers discover, unprompted, that undisclosed AI content existed at all. Disclosure lowers the stakes of any single AI error.
Step 3: A practical pattern, a post-level disclosure field
A simple, durable pattern: add a dedicated field to track AI involvement per post, surfaced consistently wherever the content is displayed, rather than a one-off note added manually and inconsistently by whichever author remembers to.
// File: my-plugin.php
add_action( 'add_meta_boxes', function () {
add_meta_box(
'my_plugin_ai_disclosure',
__( 'AI Involvement', 'my-plugin' ),
function ( $post ) {
$value = get_post_meta( $post->ID, '_my_plugin_ai_involvement', true );
?>
<select name="my_plugin_ai_involvement">
<option value="none" <?php selected( $value, 'none' ); ?>><?php esc_html_e( 'None', 'my-plugin' ); ?></option>
<option value="assisted" <?php selected( $value, 'assisted' ); ?>><?php esc_html_e( 'AI-assisted, human-edited', 'my-plugin' ); ?></option>
<option value="generated" <?php selected( $value, 'generated' ); ?>><?php esc_html_e( 'Substantially AI-generated', 'my-plugin' ); ?></option>
</select>
<?php
},
'post'
);
} );
add_action( 'save_post', function ( $post_id ) {
if ( isset( $_POST['my_plugin_ai_involvement'] ) ) {
update_post_meta( $post_id, '_my_plugin_ai_involvement', sanitize_text_field( $_POST['my_plugin_ai_involvement'] ) );
}
} );Then render it consistently, once, in a template partial rather than leaving it to individual authors to remember:
// File: single.php (excerpt)
$involvement = get_post_meta( get_the_ID(), '_my_plugin_ai_involvement', true );
if ( 'generated' === $involvement || 'assisted' === $involvement ) {
echo '<p class="ai-disclosure">' . esc_html__( 'This article was drafted with AI assistance.', 'my-plugin' ) . '</p>';
}Step 4: For chatbots and conversational features
The pattern is simpler for conversational AI: state it once, up front, before any exchange happens, rather than relying on a small badge a user might not notice.
Apply it: audit your site’s AI surfaces
List every AI-facing surface on a site you maintain: chatbots, AI-drafted content, AI-personalized recommendations. For each, check whether a person encountering it would know, without digging, that AI was involved. Add disclosure to any surface that fails that check, even where no specific law is understood to require it.
Recap
The general, widely-understood principle is simple to state: let people know when they’re interacting with AI or consuming AI-generated content. Building that as a first-class, consistently-rendered field, whether a post-meta flag or a chatbot’s opening line, beats relying on individual authors to remember. The trust case for disclosure holds regardless of the exact legal requirement in a given jurisdiction, and it’s generally cheaper to disclose consistently than to explain an undisclosed AI mistake after the fact.
Resources & further reading
- AI Act, European Commission, digital-strategy.ec.europa.eu
- Add Meta Boxes function reference, developer.wordpress.org
- Privacy guide for plugin developers, developer.wordpress.org