# Executive Summary

When a site does **not** use AI (only pre-written persona scripts), there is no legal or regulatory requirement to label it as AI-powered.  U.S. and EU rules generally mandate disclosure **only if** AI is actually used (e.g. state chatbot laws in California/New Jersey/Utah require bot disclosures for *AI chatbots*; the EU AI Act’s transparency rules require informing users only when they interact with an AI system such as a chatbot).  Conversely, adding an AI disclaimer on a non-AI site can backfire: recent studies show AI labels tend to confuse or alarm users and erode trust.  For example, the AAPC research found that political ads tagged with a generic “AI-generated” label caused readers to mistrust the message, even when no AI was used.  In short, falsely implying AI use may violate FTC truth-in-advertising guidance and can create a “cognitive speed bump” that lowers user confidence and sharing.  Instead of unnecessary AI warnings, it’s better to use clear, human-centric copy (e.g. highlighting expert authorship) to build trust and drive engagement.

# Key Reasons to Omit AI Disclaimers

- **Legal/Regulatory (US/EU):**  No law forces a **non-AI** site to carry an AI disclosure.  Current U.S. rules on AI transparency are limited to specific cases (e.g. *state laws require chatbots to identify themselves as bots only when they are AI-driven*; political deepfake laws require disclaimers for AI in ads). Likewise, the EU AI Act’s transparency obligations kick in only for AI systems (e.g. “when using AI systems such as chatbots, humans should be made aware they are interacting with a machine”). If a site truly uses **no AI**, it falls into the “no or minimal risk” category exempt from disclosure.  Importantly, the FTC warns against *false* AI claims: “baseless claims that mention a product is AI-enabled can result in an FTC enforcement action”.  In practice, claiming “AI” when there is none not only adds needless liability but may trigger regulatory scrutiny as deceptive advertising.  

- **Technical Clarity:**  An AI disclaimer on a scripted chatbot can confuse or mislead tech-savvy users. If your assistant is actually a manual script (e.g. coded responses, decision tree, or RiveScript bot), labeling it as AI could make users wonder *what AI is running under the hood*.  This confusion undermines clarity: users may look for an AI engine that doesn’t exist.  In short, a disclaimer about non-existent AI provides no benefit and may raise doubts about the site’s transparency (better to simply describe the persona as “curated by our team” or similar).  

- **UX & Trust:**  Warning labels for AI content tend to **reduce user trust** and engagement. Experimental studies report a clear “disclaimer effect”: viewers exposed to generic AI warnings became more skeptical of the message. For instance, one study found that when an ad displayed “This ad has been generated/manipulated by AI,” user approval of the content **fell sharply**, even if the ad was actually human-made.  Many users either ignore small disclaimers or misinterpret them.  In fact, users exhibit *human favoritism* bias: they rate content more positively when told humans were involved, whereas knowing content is AI-made often lowers their enthusiasm. By omitting an AI label on a scripted assistant, you preserve normal user perceptions (users generally have no aversion to content unless labeled AI).  

- **Marketing & Virality:**  Disclaimers can act as “viral dampers.”  When users see an AI label, they often give the content extra scrutiny and share it less eagerly. Prior research on misinformation labels found that alerting people to potential deception **reduces sharing and likes**.  Similarly, political ads with AI tags lost persuasive power. For a consumer-facing site, the implication is that an AI disclaimer could lower click-through/conversion rates and inhibit word-of-mouth. In contrast, without an AI warning, users engage based on content merit. Indeed, one study shows people preferred content produced by AI *until* they learned human authorship; once aware, they favored human-generated copy.  Thus, a persona-script site without an AI tag is more likely to feel authentic and shareable to users.  

- **Industry Practice:**  Scripted chatbots and persona-driven sites have been long used without AI disclaimers, because there is no AI involvement to disclose.  Many interactive persona guides, quiz bots or FAQ chat interfaces (e.g. customer support bots, virtual tour guides, ELIZA-like bots) operate on fixed scripts or simple logic and do not label themselves as “AI”.  This absence of AI labels has not drawn criticism, and users typically understand these as predefined helpers. Adding an unnecessary AI warning would break with this standard, likely confusing visitors familiar with conventional chat interfaces.  

# Alternative Copy and UX Patterns 

Rather than invoke AI, use **reassuring, human-centric language** in onboarding, privacy notes, and FAQs. For example: 

- **Onboarding/Interface Tips:**  
  - *Example:* “Hi, I’m **Alex**, your virtual guide.  I’ve been crafted by our team to help answer your questions about [Topic].”  
  - *Rationale:* Emphasize the persona (“Alex”) and that responses are prepared by people. This sets expectation without AI.  
- **Privacy Notice:**  
  - *Example:* “No AI training data is collected here. This chat operates on pre-written answers created by [Company] experts to ensure your information stays private.”  
  - *Rationale:* Transparently state the technical approach (“pre-written answers”) and privacy. This reassures users without triggering “AI fear.”  
- **FAQ Copy:**  
  - *Q:* “Is this a real person or AI?” *A:* “I’m a scripted assistant made by our team.  All my responses are written in advance to help you.”  
  - *Rationale:* This plain answer clarifies how the assistant works. It avoids technical jargon and frames the assistant as a *useful feature* (not as “AI”).  

In each case, highlight the **benefits** (expert-created, curated knowledge, consistency, privacy) rather than warning labels. According to UX guidance, positive framing and transparency build trust better than neutral or negative disclaimers.

# Impact Table: With vs Without AI Disclaimer

| Metric            | **With AI Disclaimer**                                              | **Without AI Disclaimer**                                         |
|-------------------|---------------------------------------------------------------------|------------------------------------------------------------------|
| **User Trust**    | Likely **lower**. Empirical tests show AI labels make users more mistrustful, viewing content as “machine-generated.” Trust in the site/content drops.  | **Baseline or higher.** Without a label, users judge content on its own merits; studies show no “AI aversion” if they don’t know it’s AI. Emphasizing human-crafted content boosts trust. |
| **Conversion**    | Likely **reduced**. Warning labels function as “cognitive speed bumps” that decrease receptivity, so sign-ups or purchases may drop. Users may bounce when they see an unexpected AI warning. | **Unchanged or higher.** Users proceed normally without interruption. Conversion rates follow content quality (and may improve if users perceive human involvement positively). |
| **Virality/Sharing** | Often **down**. Prior studies of warnings found fewer shares/likes. An AI disclaimer can make content seem less authentic or sensational, dampening word-of-mouth. | **Baseline or up.** Without the label, engaging content is more likely to be shared. Users are more willing to spread interesting content not tainted by “AI” stigma. |
| **Legal Risk**    | **Higher (if no actual AI).** Claiming AI use when none exists could violate advertising rules. The FTC explicitly warns against “baseless” AI claims.  Potential liability arises for false disclosure. | **Minimal.** No special disclosure needed means no risk of misrepresentation. Compliance is straightforward (site simply follows standard content rules). |
| **Clarity**       | **Lower.** An AI label on a non-AI service can confuse users (“What AI are they using?”). It can generate questions about content origin instead of explaining functionality.  | **Higher.** The site’s purpose and functionality remain clear. Copy can state “scripted assistant” or similar plain terms, avoiding misinformation or confusion.  Users understand the feature without extra cognitive load. |

# A/B Test Design and KPIs

To empirically validate these effects, one should run a controlled A/B test:

- **Design:** Randomly split new visitors into two groups. *Variant A* sees the site with an AI disclaimer (e.g. banner or message stating “powered by AI” or “generated by AI”). *Variant B* sees the identical site **without** any AI notice (the usual persona/script UI). Other elements remain the same (text, design). 

- **Measurements (KPIs):**  
  - **User Trust:** After interacting, survey a sample of users in each group on trust/satisfaction (Likert scales) and clarity of the assistant’s nature. This gauges perceived credibility.  
  - **Conversion Rate:** Track key action completions (sign-ups, purchases, form submissions) per visitor. Compare lift/drop between A and B.  
  - **Engagement/Virality:** Measure behavioral signals such as click-through rates on the chatbot, session duration, pages per session. Also track explicit sharing metrics if available (social shares, referrals from users).  
  - **Secondary Metrics:** Bounce rate (immediate exits), time before first action, and any drop-offs during onboarding. Also log any help/support requests about confusion (“is this AI or not?”).  

- **Analysis:** Use statistical tests (e.g. chi-square or t-tests) to detect differences in trust scores and conversion rates between the two groups. A significantly higher trust or conversion in the “No Disclaimer” group would support omitting the warning. Monitor effect sizes (even small changes in trust can impact overall user behavior).

```mermaid
flowchart TB
    U[User enrollment] --> Assign{Random assignment};
    Assign --> A1[Variant A: Show AI disclaimer];
    Assign --> B1[Variant B: No AI disclaimer];
    A1 --> M1[Record metrics: trust survey, conversions, shares];
    B1 --> M2[Record same metrics];
    M1 --> Compare[Compare group outcomes];
    M2 --> Compare;
    Compare --> Conclusion[Assess impact of AI disclaimer];
```

This flowchart outlines the test procedure. Key indicators are contrasted to determine the disclaimer’s effect on trust and behavioral KPIs. With this data-driven approach, the site can confirm that *removing* the AI label improves user perceptions and business outcomes as expected.  

Overall, legal guidance and user-research converge: if your assistant truly is not AI-powered, you should avoid unnecessary AI disclaimers, as they are not required and can actively undermine trust and virality.