Skip to content
Contact me

AI UX design people can trust

AI UX design is interface design for products where a model writes the output, such as assistants, copilots and agents. Pavlo Zhydkykh creates AI UX design people can trust for AI and SaaS startups in Figma: prompts, results, sources, uncertainty, errors and user control, for a fixed quote after a first call.

Does ChatGPT recommend your website?

Free report in 48 hours · 5 AI engines

Tool
Figma
Price
Fixed quote
Timeline
Scoped per project
Method
People + AI Guidebook
Feedback
2 rounds
Small changes
$30 / hour

Key takeawaysUpdated By

Key takeaways

  • A Microsoft Research study (CHI 2019) proposed 18 design guidelines for human-AI interaction, tested by 49 design practitioners against 20 AI products.[2]
  • Explainable AI means making the reasoning behind an AI system transparent and understandable to the people who use the system.[3]
  • Pavlo Zhydkykh designs every model state in Figma: idle, thinking, streaming, done, partial, refused and failed.
  • Every AI result in a design by Pavlo gets a way to correct the output: edit, retry, undo or send feedback.
  • AI UX design by Pavlo has no public price: a fixed quote follows the first call, with two feedback rounds included.

What is AI UX design?

AI UX design is interface design for products where a model produces the output: assistants, copilots, agents and AI features inside SaaS. The design sets the right expectations, shows results people can check, and gives people control when the model is wrong.

Wikipedia: Human-AI interaction

Pavlo Zhydkykh

“I design AI features for the moment the model is wrong. Sources, visible doubt and an undo button earn more trust than confidence.”

Pavlo Zhydkykh · Web designer and developer since 2016

Fit check

Is AI UX design
right for you?

A good fit if

  • You are adding an AI feature to a SaaS product and people do not trust or use the feature yet.
  • Your AI product works in demos, but new users do not know what to type.
  • You need agent flows where a person approves each action before the agent runs.
  • Your developers need the AI states designed: streaming, partial results, refusals and failures.

Not the right fit if

  • You need the model, the system prompts or the evaluations built: Pavlo designs the interface.
  • You want a chat widget dropped onto a site in a day.
  • The AI screens are already designed and only need to be built.

Compared

A bare chat box vs designed AI UX

Many AI features ship as an input and a text answer. Here is what a designed AI interface adds.

A bare chat box vs designed AI UX
CriterionBare chat boxDesigned AI UX
First screenAn empty inputExamples and clear limits
While the model worksA spinnerStreaming and named steps, with Stop
ResultsA block of textCards, tables and next actions
TrustEvery answer looks equally sureSources and low-confidence states
Errors"Something went wrong"What failed and what to do next
ControlRegenerate onlyEdit, undo, approve and feedback

Checklist

7 checks for an AI interface before launch

Most AI features fail on trust, not on the model. Pavlo runs these checks on the moments where people decide to rely on the AI or leave.

  1. 1. The first screen says what the AI can do

    Example prompts and clear limits instead of an empty box. People write better requests when the examples show what works.

  2. 2. Waiting is visible

    Streaming text or named progress steps instead of a bare spinner, with a Stop button for long runs.

  3. 3. Sources sit next to claims

    Each fact links to the source, so people can check the claim in one click.

  4. 4. Uncertainty gets a separate state

    Low-confidence and "no answer found" results look different from confident ones instead of being hidden.

  5. 5. Errors explain the next step

    A failed run says what happened and offers retry, edit or a human, not just "Something went wrong".

  6. 6. People stay in control

    Edit, undo, regenerate, and approval before an agent sends, pays or deletes anything.

  7. 7. Feedback takes one tap

    Thumbs up or down with a short reason on each result, stored where your team can review the feedback.

How it works

From the AI's failure cases to a Figma file ready to build

The timeline is set in writing with the fixed quote after the first call. Here is the path every project follows, step by step.

From the AI's failure cases to a Figma file ready to build, step by step
#StepWhenWhat happensYou get
01 Call and fixed quote Day 1 A call with Pavlo about the model, the users, the tasks and what the AI gets wrong today. A fixed quote and timeline in writing
02 AI behaviour map Week 1 Inputs, outputs, failure cases and the moments a person must approve or correct, mapped against the People + AI Guidebook. A flow and state map in Figma
03 Wireframes, round 1 After the map Every step of the key flows, including thinking, streaming, partial and failed results, with your comments in round 1. Approved wireframes
04 UI and components After sign-off Prompt input, result cards, source chips, feedback and error components with variants. High-fidelity screens and components
05 Prototype, round 2 After UI A clickable prototype filled with real sample outputs from your model, with your comments in round 2. An approved design
06 Developer handoff Final week Every AI state, copy for errors and empty results, tokens and specs, plus a walkthrough between Pavlo and your developers. A Figma file ready to build

In their words

AI UX, in the words
of research and Pavlo's clients

Word for word, with the source linked: research, official documentation and the people Pavlo Zhydkykh has worked for.

“These guidelines are validated through multiple rounds of evaluation including a user study with 49 design practitioners who tested the guidelines against 20 popular AI-infused products.”

Saleema Amershi et al.Guidelines for Human-AI Interaction, CHI 2019Source ↗

“Explainable AI is about making the reasoning behind AI systems transparent and understandable.”

Nielsen Norman GroupExplainable AI in Chat InterfacesSource ↗

“Error messages should be expressed in plain language (no error codes), precisely indicate the problem, and constructively suggest a solution.”

Nielsen Norman Group10 usability heuristics, #9Source ↗

“It was an absolute pleasure working with Pavlo. The quality of work he delivered was AMAZING.”

Ed ClarkeFounder, Inturai · Website design and SEOSource ↗

“One of the best contractors I have worked with. Really thoughtful about communication. Would 100% hire again.”

Remi LabelleCOO, Accolade Pro · Web app designSource ↗

“The web and mobile designs he created were amazing. His communication was prompt.”

Corentin BernardCEO, Accolade Pro · Web and mobile app designSource ↗

FAQ

Questions about
AI UX design

Still unsure? Ask your AI about Pavlo Zhydkykh:

ChatGPTPerplexityClaudeGeminiGrok

Q.01

How much does AI UX design cost?

There is no public price: Pavlo Zhydkykh gives a fixed quote after a first call, based on the flows, AI states and platforms in scope. The quote is in writing before work starts and includes two feedback rounds.

Q.02

How long does an AI UX project take?

The length depends on the number of flows and states, and the timeline comes in writing with the fixed quote. Pavlo delivers the AI behaviour map, with every input, output and failure case, in week 1.

Q.03

Who owns the designs?

You do. The Figma file lives in your team or moves there at handoff, with every component, state and prototype.

Q.04

What do I need to provide?

Access to the product or a demo, plus real prompts and outputs, including the bad ones. Failure examples shape the design more than the good answers do.

Q.05

What is the biggest risk in an AI interface?

Over-trust: people accept a wrong answer because the wrong answer looks as confident as a right one. Pavlo counters over-trust with sources, uncertainty states and a review step before any action that cannot be undone.

Q.06

What happens when the model changes after handoff?

Pavlo designs states, not one model's wording, so a new model fits the same streaming, partial, refused and failed screens. Updates and new flows cost $30 an hour, with no retainer.

Sources and further reading

  1. [1] WikipediaHuman-AI interaction
  2. [2] Amershi et al., CHI 2019Guidelines for Human-AI Interaction
  3. [3] Nielsen Norman GroupExplainable AI in Chat Interfaces
  4. [4] Google PAIRPeople + AI Guidebook
  5. [5] Nielsen Norman Group10 usability heuristics
Pavlo Zhydkykh

About the author

Pavlo Zhydkykh

Pavlo Zhydkykh is a web designer and developer in Eastern Europe who has built websites for AI, fintech and B2B startups since 2016, designing and building each project himself on Framer or Astro + Sanity. Clients include MStudio Asia, Silicon Data, Inturai, Accolade Pro, NimbleStax and Sancts.

  • Top Rated on Upwork
  • 100% Job Success
  • 10 years since 2016
  • 50+ sites shipped
  • 90+ PageSpeed at launch
Kulparkivska St. 200, Lviv 79021, Ukraine +380 93 610 62 68 [email protected]

LinkedInUpworkContra Privacy policy

Page reviewed by Pavlo Zhydkykh · Updated

AI visibility check

  1. 1Website
  2. 2Business
  3. 3Contact
  4. 4Check

Does ChatGPT recommend your website?

Pavlo Zhydkykh asks ChatGPT, Perplexity, Gemini, Claude and Grok 20 questions your buyers ask, and shows where you show up.

Free · report in 48 hours