Real people · Not synthetic data · UK-wide
If your product includes a chatbot, AI copilot, recommendation engine, or any other AI powered feature, the people who matter are the real humans who will use it, not a simulated panel or another AI playing the role of your customer. We test AI features with real recruited participants, so you find out how people actually understand, trust, and use your AI before they do it on your live product.
AI generated personas and simulated users are trained on existing patterns. They can describe what a "typical" user might do, but they cannot show you what happens when a real person with a real task encounters your AI for the first time, gets an answer they do not trust, or asks a question your model was never designed for.
Testing AI with AI also creates a circularity problem. If both the product and the evaluator are generative models, you risk validating your AI against its own blind spots, rather than against the people it is actually built for. We set out the evidence in why AI personas cannot replace real user research.
We bring real recruited participants into contact with your AI powered product and watch what actually happens: where they get value, where they get confused, where they stop trusting the system, and where they need a human to step in.
From customer facing chat assistants to AI features quietly working in the background, we design research around how real people experience your AI.
Test how real users phrase questions, react to responses, recover from misunderstandings, and decide whether to trust or escalate to a human.
Watch how people use AI suggestions inside your product. Do they accept, edit, or ignore them, and does the assistance speed people up or get in the way?
Understand whether recommendations feel relevant, how people react when they are wrong, and whether users notice or care that an algorithm is involved.
Test AI generated text, images, or summaries for clarity, accuracy perception, and whether users can tell, or need to know, what was generated.
See whether AI generated answers and summaries help people find what they need faster, or introduce new confusion and verification work.
For tools that recommend or make decisions, test whether users understand the basis for a decision, when they would override it, and what evidence they need to feel comfortable.
The same rigorous process as our usability testing, with extra attention to how people interact with AI generated output.
We discuss what your AI feature does, who uses it, and what "working well" looks like, including any known failure modes or edge cases. Free of charge, usually 30-45 minutes.
We design tasks and scenarios that reflect real use, including situations where the AI might get something wrong, so we can see how people respond.
We recruit real users matching your audience. For regulated or clinical contexts, this can include patients, carers, or specific professional groups.
Moderated sessions where participants use your AI feature on real or realistic tasks. We probe how they interpret, trust, and act on what the AI produces.
Beyond standard usability findings, we analyse how participants understood the AI's reasoning, noticed errors, and decided whether to rely on it.
Findings covering usability, trust, and AI specific risks, with video evidence and prioritised recommendations your team can act on.
Beyond standard usability, AI features raise questions about trust, errors, fairness, and when a human should take over.
Do people trust the AI the right amount, not too much and not too little?
Best when You are launching a feature where over reliance or under reliance on AI could cause real harm or missed value.
What you get Evidence of how confident users feel, when that confidence is misplaced, and what changes it.
You leave with Recommendations for framing, disclaimers, and confidence indicators that calibrate trust appropriately.
What happens when the AI gets it wrong, and can users tell?
Best when Your AI sometimes produces incorrect, irrelevant, or unhelpful output, as most do.
What you get Evidence of whether users notice errors, how they react, and whether recovery paths work.
You leave with Design recommendations for error states, fallback options, and human handoff points.
Does the feature work as well for everyone who needs to use it?
Best when Your AI feature serves a diverse audience, including people with different digital confidence, language needs, or accessibility requirements.
What you get Evidence from inclusive recruitment that reflects your real user base, not just confident early adopters.
You leave with Findings highlighting where the experience breaks down for specific groups, and what to fix.
When should the AI step back and let a person take over?
Best when Your product combines AI with human support, whether that is a customer service team, a clinician, or an account manager.
What you get Evidence of where users want or need a human, and whether your current handoff points match that.
You leave with Recommendations for where to add, simplify, or remove escalation points.
From startups building AI native products to established healthcare and regulated organisations adding AI features carefully.
Transparent pricing, the same as our other usability and UX research services.
Quick Validation
Focused testing of a single AI feature or flow. Right for validating a chatbot, copilot, or recommendation feature before or shortly after launch.
Comprehensive Testing
Thorough testing across multiple AI features or user segments, with enough sessions to spot patterns in trust, errors, and edge cases.
Multi-Phase
Ongoing research partnership as your AI features evolve. Test across iterations, combine with PPIE or design validation, or run continuous trust monitoring.
All packages include: Participant incentives, session recording and storage, synthesis and analysis covering both usability and AI specific findings, video clip highlights, and clear documentation your team can act on immediately.
Combining services: AI product testing pairs well with design validation for regulated products, or PPIE when AI features are used in healthcare and patient facing contexts.
We fully recognise the effort your team invested in recruitment, moderation, and analysis, and we genuinely appreciate the quality of the discussions and reporting.
Trusted PPI and PPIE delivery partner to the NIHR HealthTech Research Centre in Accelerated Surgical Care.
Tell us what you have built and what you are unsure about. We will suggest the right approach, whether that is AI product testing on its own or combined with usability testing, UX research, or PPIE.
Start a conversationWe usually respond within one working day.
Book a free 20-minute call. No obligation, no sales pitch. We'll tell you honestly whether research is worth it for your decision, and what it would cost.