Which Website Tools Secretly Use AI in 2026 (And Need EU Disclosure)

By AJ Fox, A50 Check · Vendor and regulation details verified August 26, 2026

That "live chat" widget you installed three years ago might now be answering customers with generative AI — and you may not even know the setting got flipped on. Website tools have quietly rolled AI into features that used to be simple automation, and under EU law, not knowing your tool uses AI is not a defense.

This roundup covers the most common website tools with AI features people don't realize are AI, whether that AI ships on by default, and what EU AI Act Article 50 requires for each category. Always verify current settings directly in your vendor dashboard, since defaults change without much announcement. One note on scope up front: our free scanner currently detects AI chat widgets, AI bot platforms, and voice agents on your site — the table below marks exactly which tools it catches today and which it doesn't yet, because we'd rather be precise than impressive.

Which Website Tools Secretly Use AI: Chat Widgets

Chat widgets are the single biggest blind spot. Many businesses installed a "live chat" tool years ago for human agents and never noticed when the vendor added an AI layer on top.

Intercom Fin

Fin is Intercom's AI agent, and it's one of the more transparent vendors about EU compliance. Intercom states that Fin was built to meet the basic disclosure requirement by appearing as "Fin 'AI Agent'" with responses marked as AI answers (Intercom help center). However, the default toggle status depends on when you started using it — per Intercom's own changelog, the "AI Agent" label is "ON by default if you're already using Fin. OFF by default if you're not yet using Fin":

If you were already using Fin when Intercom launched the label in July 2025, the AI Agent label is on by default.
If you started using Fin after that, the label is off by default, though an editable introductory message is pre-populated.

For email conversations, Intercom ships a separate disclosure footer that is on by default for all customers, while an introductory AI sentence in Fin's first email reply is off by default — both are configurable in Fin settings (Intercom changelog).

This means two businesses using the identical product can have opposite disclosure states right now, purely based on signup date. Check your own toggle under Fin AI Agent → Fin settings → General — Intercom has changed these defaults before, so don't assume your settings match your memory.

Tidio Lyro

Tidio's AI layer is called Lyro, and it's bundled into Tidio's main plans — every new project gets a free, non-renewable quota of 50 Lyro conversations (Tidio help center) — which makes it easy to enable without a deliberate purchase decision. Lyro's messages are configurable, including its follow-up and typing-indicator text (Lyro documentation). Tidio doesn't publicly document whether Lyro shows an AI disclosure by default, so check your Lyro identity and greeting settings directly rather than assuming one is shown.

Drift

Drift built its reputation on AI-powered "Bionic Chatbots" for B2B lead qualification (Drift announcement). Businesses considering Drift should know that Salesloft, which acquired Drift in 2024, announced Drift's sunset as a standalone product in 2026 (coverage). If your site still runs a legacy Drift chatbot, confirm its current AI status and disclosure settings directly with Salesloft, since a discontinued product's compliance posture is exactly the kind of detail that stops getting updated.

Zendesk AI (Agents and Copilot)

Zendesk splits its AI into customer-facing AI agents and an internal-only Copilot for human support staff. This distinction matters for disclosure: Copilot's guidance is intended for internal users only and isn't shown to customers, so it likely doesn't trigger Article 50(1) disclosure the way a customer-facing AI agent does (vendor documentation). Zendesk itself confirms that "some individual AI features are turned on by default, and others must be turned on," without a single blanket answer (Zendesk help center). Check Admin Center → AI settings to see exactly which features are active on your account, since this varies by plan and feature.

Which Website Tools Secretly Use AI: Voice Agents

AI voice agents are newer than chat widgets and less standardized in how they disclose themselves. Tools like Vapi let businesses embed AI phone and voice assistants directly into their sites and call flows — and Vapi's own documentation shows compliance behavior is entirely deployer-configured, with no automatic spoken disclosure. If your website or phone system routes calls through an automated voice assistant, the same Article 50(1) rule applies as with text chatbots — the caller must be told they're talking to an AI, and it has to happen at or before the first real interaction (Article 50 text). With Vapi specifically, disclosure isn't a toggle: it goes in the assistant's start node or first message and is spoken exactly as written on every call, so write your AI identification line there and test it by calling your own number (voice AI disclosure guidance).

Which Website Tools Secretly Use AI: Video and Image Generators

Video generation tools sit under a different part of Article 50 — mostly Article 50(2), which requires the provider (the tool itself) to mark outputs in a machine-readable format, and Article 50(4), which requires the deployer (you, if you publish the video) to visibly disclose deepfake-style content (Article 50 text).

Synthesia

Synthesia handles machine-readable marking automatically, applying C2PA Content Credentials to relevant videos. But Synthesia is explicit that visible, on-screen disclosure to viewers — telling your audience "this video was made with AI" — is the deployer's job, not something the platform can do for you. Content generated before August 2, 2026 does not need retroactive labeling. (All per Synthesia's Article 50 guidance.)

HeyGen

HeyGen commits to an invisible watermark, embedded provenance metadata, and a verification mechanism for its AI-generated content by December 2, 2026. But HeyGen is direct about the limits of what it can do: as the deployer, you must decide how to declare to your audience that your content was created with AI — HeyGen cannot make that labeling decision for you (HeyGen's EU AI Act compliance page).

An honest caveat: automated scanners — ours included, today — detect video generators far less reliably than chat widgets, because an embedded AI video often looks identical to a normal video embed in a page's code. If you publish AI-generated video, don't wait for a scanner to flag it: check your own content library against the disclosure rules above.

Quick-Reference Table: Which Website Tools Secretly Use AI

Tool	Category	AI on by default?	Detected by A50 Check scan?	Source
Intercom Fin	Chat widget	On if you used Fin before the July 2025 label launch; off otherwise	Yes	Intercom
Tidio Lyro	Chat widget	Not publicly documented — check your settings	Yes	Tidio
Drift	Chat widget	Product discontinued 2026 — confirm status with Salesloft	Yes	Coverage
Zendesk AI Agents	Chat widget	Varies by feature — check Admin Center	Yes	Zendesk
Zendesk Copilot	Internal agent tool	On for eligible plans, but not customer-facing	Yes (widget detected)	Docs
Vapi	Voice agent	No default — set it in the start node	Yes	Guidance
Synthesia	Video generator	Machine marking automatic; visible label is manual	Not yet — check manually	Synthesia
HeyGen	Video generator	Watermark/metadata automatic by Dec 2026; visible label is manual	Not yet — check manually	HeyGen

Check your own site: run a free scan at a50comply.com — it takes about 60 seconds, no email required, and it flags the AI chat widgets and voice agents your pages are actually loading.

Frequently Asked Questions

How do I know if my chat widget uses AI or just rule-based automation? Check your vendor's admin settings for any feature named "AI Agent," "Copilot," "Lyro," "Fin," or similar branded AI names — these are the giveaway labels. If unsure, contact vendor support directly and ask explicitly whether generative AI or machine learning powers the customer-facing responses.

Do I need to disclose AI if I only use it for internal staff tools, like Zendesk Copilot? Generally no, because Article 50(1) is about AI systems that interact directly with natural persons — if the AI output is only visible to your internal team and never shown to the customer, the disclosure trigger is less clear-cut. But if AI-drafted replies reach customers with only light human review, treat that as customer-facing and disclose. When in doubt, disclose — over-disclosing carries no penalty.

Does my AI video tool or my website need to add the disclosure label? Both, in different ways. The video tool (like Synthesia or HeyGen) is responsible for machine-readable marking; you, as the person publishing the video on your site, are responsible for the visible, human-readable disclosure.

What if my chatbot vendor changed its AI default after I signed up? This happens more than people expect — Intercom's Fin label defaults, for example, depend on whether you were using the product when the label launched in July 2025 (changelog). Don't assume your settings are still what they were when you first configured the tool; check current settings periodically.

Which tool should I check first if I only have time for one? Start with whatever handles the highest volume of customer conversations — usually your main chat widget — since that's the most visible and highest-traffic point of AI interaction on your site.

For the underlying legal requirement behind all of this, see our companion article on AI chatbot disclosure requirements in the EU, which breaks down exactly what Article 50(1) requires and how to fix a missing disclosure quickly.