Skip to content

TRUE Score™: the benchmark metric for measuring the experience of your AI.

Technical metrics check the model. The TRUE Score™ measures the experience, on four dimensions scored from 0 to 100: trust, resolution, usefulness and effort. Compare your AI features, track every release and prove their value.

  • 4 dimensionsscored from 0 to 100
  • 6 types of AI experiencesconversational or not
  • 2026 benchmark7 industries, 40 combinations

Go beyond technical logs.

Latency, costs, error rates: your metrics can be green while the actual experience is red. A disappointed user doesn't open a ticket. They give up and work around it.

Blind spot Without Screeb With Screeb
Automated response evaluation It scores the model's answer, not what the user did with it. User feedback, cross-referenced with their behavior after the interaction.
Thumbs up or thumbs down A click with no nuance: it says it didn't work, not why. Four distinct dimensions, plus an open question to understand the score.
The quarterly email survey The memory is reconstructed, and the model has already changed. Continuous measurement, at the moment of the interaction, tied to each release.

The four dimensions of the TRUE Score™.

Each dimension is scored from 0 to 100, and the TRUE Score™ is their average. Together, they decide whether a user will trust an AI experience again tomorrow.

  • Trust

    Do they believe what the AI tells them?

    Verification signals, direct statements, willingness to follow a recommendation.

  • Resolution

    Did they get what they came for?

    From their own point of view: stated feedback cross-referenced with downstream behavior.

  • Usefulness

    Beyond the immediate request

    Did the interaction add value? An answer can be accurate and useless.

  • Effort

    How much did it cost them?

    Rephrasing, back-and-forth, cognitive load. The smoother the experience, the higher the score.

How to read a score

  1. Below 45Destructive experienceAn experience that damages your brand.
  2. 45 to 59Endured experienceAn obstacle in front of the real service.
  3. 60 to 74Tolerated experienceUsed when it's the shortest path.
  4. 75 and aboveAdopted experienceUsers come back and recommend it.

Connect the TRUE Score™ to your technical data.

Your observability tools, like Langfuse or Datadog, tell you what the model does. The TRUE Score™ tells you what the user experiences. Bring the two together and you see the effect of every technical choice on the experience.

  • Attach technical data. Model version, prompt version, latency: sent to Screeb with every interaction.
  • Break down each dimension. Trust by model, resolution by prompt, usefulness by latency.
  • Find the cause of a drop. A release that pulls a score down shows up in the report.
  • Decide on facts. Is a faster model worth a few points of trust? The report answers.

A single evaluation framework for all your AI use cases.

The TRUE Score™ isn't limited to conversational agents. It tracks any AI experience your users encounter.

  • Conversational agents

    Web and mobile

    Chatbots, AI assistants, virtual advisors.

    Agent Analytics
  • Semantic search

    Find without the exact word

    Augmented search, smart filtering, content discovery.

  • Product recommendations

    Conversion-focused

    Dynamic, personalized suggestions.

  • Smart onboarding

    Guide first use

    Product tours, smart checklists, augmented help centers.

  • Dynamic personalization

    Real time, by segment

    Adapting the interface and content.

  • Simulators and configurators

    Help people decide

    Product builders, calculators, decision-support tools.

Evaluate your AI based on what your users think of it.

At Edifice, a Screeb survey shown right after each generated text guided the tuning of the prompt and the LLM.

Portrait de Romain Gentet
“Measuring satisfaction matters more because we don't control exactly what the AI will produce, unlike a classic, more algorithmic feature.”
Romain Gentet · Lead Product Manager · Edifice
79%satisfaction with its AI rewording feature, up from 34% in June
10 safter the generated text: when the survey appears, for immediate feedback
400,000parents or so, the platform's audience in the Hauts-de-France region
Read the Edifice story

Compare your performance to your industry's standards.

Your measurement gives you a score; a benchmark tells you what it's worth. The 2026 TRUE Score™ benchmark evaluates 6 types of experiences across 7 industries, or 40 combinations, with one sheet per industry. It's a first edition, to be read as a reference point: the scores are estimates, not averages from a representative sample.

  • 62/100Average TRUE Score™ of AI experiences: usable, but not convincing
  • 58/100Trust, the weakest dimension, across all 7 industries studied
  • 56/100conversational agents, the lowest score of the 6 experience types, versus 67 for smart onboarding
  • +13 to +16 ptslead of the top quartile over the median, across the 40 combinations

Source: 2026 TRUE Score™ Benchmark.

What Screeb does for you

Screeb instruments your priority AI experiences, sets your baseline within 4 weeks and places you against the median and top quartile of your industry, with a diagnosis for each dimension and a prioritized action plan.

Get your TRUE Score™
Frequently asked questions

Your questions.

What is the TRUE Score™?

The NPS of AI: a score from 0 to 100, the average of four dimensions (trust, resolution, usefulness, effort) that measures the perceived quality of an AI experience.

How is it different from an observability tool?

They complement each other. Observability monitors the technical health of your AI: latency, costs, traces, errors. The TRUE Score™ measures what the user experienced and what they do next. An agent can pass all your technical evaluations and still leave your users frustrated.

Does the TRUE Score™ only apply to conversational agents?

No. It also covers semantic search, recommendations, smart onboarding, dynamic personalization and simulators.

How does it relate to Agent Analytics?

Agent Analytics analyzes your agents' conversations. The TRUE Score™ is the measurement method, for any AI experience, conversational or not.

How do I know if my score is good?

With the reading scale (75 and above: adopted; 60 to 74: tolerated; 45 to 59: endured; below 45: destructive) and by comparing yourself to the median and top quartile of your industry in the benchmark.

Are the benchmark figures market averages?

This first edition is a pilot: the scores are estimates drawn from early measurements at Screeb customers, expert calibration and public literature. Focus on the orders of magnitude and the gaps between experience types or industries, not the exact figure in any one cell.

Can we measure without Screeb?

Yes, in-house: you get a number. Measuring with Screeb also tells you what it means, since no company can measure its peers' experiences.

Is your AI adopted, tolerated or endured? Measure it.

Screeb instruments your priority experiences and sets your baseline within four weeks, with a diagnosis for each dimension.

Shall we set your baseline?