AI in Varify, and how you control it
Varify has five AI features. Four of them are optional and switched on per user, and two of those can run on your own AI key. The AI Support Assistant answers in the chat widget. This page shows what each feature sends to a model, which provider processes it, how to switch AI off and what the EU AI Act means here.
2,700 teams worldwide already rely on Varify
AI that takes work off your CRO process
Varify has five AI features: the AI CRO Audit (Get Test Ideas), Prompt-Based Experimentation, the AI Results Summary, the Varify MCP and the AI Support Assistant. From the first test idea to the final result, they do the time-consuming parts, so your team can test more and faster.
AI CRO Audit (Get Test Ideas)
Analyzes a page and returns CRO insights, concrete recommendations and 10 test ideas. No more starting from a blank page.
Prompt-Based Experimentation
Describe a change in plain language, and the AI builds the variation in your site's design. No developer needed for the first draft.
AI Results Summary
Explains your results in plain language: what won, how certain it is and what to test next.
Varify MCP
Connects your AI assistant, such as Claude or ChatGPT, to Varify. Set up and check experiments right from the chat.
AI Support Assistant
Answers your questions in the chat widget on the website and in the app. It draws only on the user documentation and published pages.
Which subprocessors Varify's AI tools use
Without your own key, all AI features run on Anthropic (Claude) models. Firecrawl is not an AI service. It loads the website content that PBX and the AI CRO Audit work on. According to the DPA, both process data for these features only when a user activates them. The AI Support Assistant also runs on Anthropic, whenever someone writes in the chat widget.
| Subprocessor | Purpose as stated in the DPA |
|---|---|
| Anthropic (Claude) | „Provision of AI-supported language models (Claude) for processing text inputs and generating text outputs within the optional AI features of the Processor.“ |
| Firecrawl (SideGuide Technologies, Inc.) | „Provision of a web scraping and data extraction API for preparing web content for AI-supported features within the optional AI services of the Processor.“ |
Source: Varify DPA, Annex 5 (v9, September 3, 2026). Annex 5 also lists where data is processed and the transfer bases.
Use your own AI key
For teams with strict requirements: connect your own API key from Anthropic, Google or OpenAI. Prompt-Based Experimentation and the AI CRO Audit then run through your own provider account.
Your account, your contract
With your own key, these features call the model provider through your own account. Your own contract with that provider applies. Varify's account and contract are not used for these requests.
Why teams use it
Use the AI provider your company has already approved, keep AI usage under your own agreement with that provider, and see the usage on your own bill.
Which features
- Prompt-Based Experimentation (PBX)
- AI CRO Audit (Get Test Ideas)
The AI Results Summary and the AI Support Assistant keep running on Varify's Anthropic models. The Varify MCP works with the AI assistant you connect yourself.
How Varify is assessed under the EU AI Act
The EU AI Act sorts AI systems by risk. Here is where Varify stands, and the rules we follow for every AI feature.
- EU AI Act
What the EU AI Act is
The EU AI Act is Regulation (EU) 2024/1689. It sorts AI systems by risk into four groups:
- Prohibited practices
- High risk
- Limited risk, with transparency duties
- Minimal risk
Article 4 on AI literacy applies since 2 February 2025. Article 50 on transparency applies since 2 August 2026. The Digital Omnibus moved the high-risk obligations to 2 December 2027 and 2 August 2028.
Varify's rules for AI
- Varify's AI features and workflows run only on Anthropic (Claude) models, with a fixed model identifier.
- Every AI application has an entry in Varify's model register and a policy before it goes live.
- A model change needs approval and a register entry.
- The Varify team is trained in AI literacy (Article 4).
Risk analysis
Varify's role: deployer
Varify uses AI models from Anthropic in its product. It does not develop its own AI model or place one on the market.
No high-risk AI
None of Varify's AI features falls into the prohibited or high-risk groups. Features that talk to people, such as the AI Support Assistant, are limited risk with transparency duties.
Every feature on record
Each AI feature has an entry in our model register and its own policy before it goes live. A change of model needs approval and a new entry.
No personal data from Varify to the AI
Varify does not send personal information to an AI model on its own. The AI only gets what a feature needs to do its job.
What the AI gets
Your prompt, the page content and the element you select, screenshots you upload, the result data of an experiment, or your messages in the support chat. Nothing else.
Our recommendation
Do not enter personal information into AI features. Build variations with a test account, so no names or email addresses are visible on the page.
You stay in control
AI features are off until a user turns them on. „Use AI features“ is a separate permission, so you decide who may use them. The AI Support Assistant is the exception: it answers whenever you write in the chat widget.
See Varify AI in action
Talk to our team about AI, data and the EU AI Act, or try Varify yourself.
Book a call
We show you the AI features and answer your questions about data and compliance.
Book a call