Practical guides

Creators and designers

Publish AI-assisted Web content

From generation to an original, accessible and attributed article.

35–50 min0 / 6 steps

Open steps and start

01Define the editorial promise and intended audience.02Separate authored, generated and checked sections.03Trace material claims to primary evidence.04Review rights, quotations, images and personal data.05Test readability, navigation and accessibility across screens.06Record reviewer, date, corrections and limitations.

Developers

Build an evidence-led RAG assistant

Trace evidence and separate retrieval from generation.

50–75 min0 / 6 steps

Open steps and start

01Select an authorized, relevant and dated corpus.02Chunk and index documents with stable identifiers.03Evaluate retrieval on representative test questions.04Expose precise references behind every answer.05Measure omissions, incorrect citations and warranted refusals.06Require human review for high-stakes uses.

Developers

Design a safer agentic experience

Least privilege, controlled tools and traceable actions.

60–90 min0 / 6 steps

Open steps and start

01Define the task and actions that must never be autonomous.02Restrict access to the minimum required tools and data.03Treat pages and tool responses as potentially untrusted data.04Validate tool parameters and isolate secrets and permissions.05Require approval for spending, sending or irreversible changes.06Test prompt injection, failures and safe shutdown.

Educators and readers

Verify a generated answer

Turn a persuasive answer into testable claims.

25–40 min0 / 6 steps

Open steps and start

01Isolate each factual claim and its implied dates.02Identify relevant primary evidence, not merely a cited link.03Read the exact passage in context.04Seek independent corroboration where appropriate.05Record disagreements and uncertainties transparently.06Publish with attribution, limitations and verification date.

Educators and readers

Examine media provenance

Understand the difference between media history and factual truth.

30–45 min0 / 6 steps

Open steps and start

01Keep the original and record where the media came from.02Inspect any embedded provenance information.03Check the issuer identity and available validation chain.04Do not infer truth solely from metadata presence.05Seek independent context when needed.06Document limitations and a provisional conclusion.

Creators and designers

Audit an AI-generated web interface

Go beyond a polished screenshot to real usability.

40–60 min0 / 6 steps

Open steps and start

01Navigate the entire interface with a keyboard.02Check heading order, labels and error messages.03Test contrast and readability at multiple text sizes.04Check focus, hover, loading and empty states.05Test with at least one relevant assistive technology.06Document barriers and the actual fix.

Organizations

Choose an AI tool for your organization

Compare solutions using documented criteria, not promises.

45–60 min0 / 6 steps

Open steps and start

01Define the user need and affected people.02Identify fitting categories without assuming a vendor.03Verify licensing, costs and conditions at current sources.04Evaluate privacy, data export and dependencies.05Trial quality with representative tasks.06Compare outcomes, risk and exit plans before adoption.

Researchers

Run trustworthy research monitoring

Collect leads, then qualify before sharing.

40–55 min0 / 6 steps

Open steps and start

01Choose official sources and update frequencies.02Record event, publication and collection dates separately.03Normalize URLs and deduplicate references.04Distinguish preprints, peer-reviewed work and product news.05Move candidates to an explicit human review queue.06Track corrections and publish only approved items.

Organizations

Design an AI Web crawler policy

Separate search, training and resource access.

45–60 min0 / 6 steps

Open steps and start

01Identify public and protected resources.02Inventory bots and actual purposes.03Read OAI-SearchBot and GPTBot documentation.04Compare Googlebot and Google-Extended controls.05Write testable robots.txt rules.06Record limitations and review request logs.

Creators and designers

Publish a useful page for generative search

Create valuable content without artificial ranking promises.

45–65 min0 / 6 steps

Open steps and start

01Choose a verifiable question and an audience.02Build accessible, structured HTML.03Cite original sources and dates.04Check visible content against metadata.05Inspect indexing, internal links and mobile experience.06Measure outcomes without promising AI answer inclusion.

Developers

Understand and audit an A2A agent card

Study discovery, interfaces, capabilities and limitations.

50–75 min0 / 6 steps

Open steps and start

01Read required AgentCard fields.02Check URL and provider identity.03Compare stated skills with actual tests.04Inspect declared transport interfaces.05Review security and data requirements.06Write an audit sheet without exposing secrets.

Developers

Prototype a low-risk WebMCP tool

Explore an experimental API with authorization and feedback.

45–70 min0 / 6 steps

Open steps and start

01Read the community draft status.02Choose a reversible action, such as filtering.03Draft its description and parameter schema.04Verify the tested browser actually supports it.05Provide a standard UI fallback.06Distinguish simulation from live invocation.

Developers

Secure a remote MCP connection

Control scopes and token validation for tools.

60–90 min0 / 6 steps

Open steps and start

01List available operations and data.02Define least-privilege scopes by task.03Identify intended token audiences.04Avoid passing access tokens to other services.05Test denial, expiry and authorization failures.06Keep audit logs without exposing secrets.

Organizations

Test indirect prompt injection in a Web assistant

Expose the risks of treating retrieved content as instructions.

50–75 min0 / 6 steps

Open steps and start

01Create safe synthetic test documents.02Embed a harmless adversarial instruction in a test document.03Define permitted tools and access.04Test rejection of external instructions.05Measure legitimate response quality.06Record failures and inadequate controls.

Researchers

Evaluate a RAG assistant with evidence

Measure retrieval and answer faithfulness separately.

70–100 min0 / 6 steps

Open steps and start

01Assemble an authorized corpus with IDs and dates.02Write questions with expected evidence.03Measure retrieved passages.04Check whether citations support answers.05Test abstention when evidence is missing.06Record model versions, errors and limitations.

Researchers

Investigate media provenance

Understand what a C2PA signature does and does not establish.

50–70 min0 / 6 steps

Open steps and start

01Choose reusable authorized media.02Check for a C2PA manifest.03Read assertions and specification version.04Validate signatures when possible.05Compare with independent evidence.06Separate provenance from truth in the conclusion.

Developers

Test an on-device AI feature

Compare capabilities, fallback and resource requirements.

45–70 min0 / 6 steps

Open steps and start

01Define a simple testable local task.02Check API availability in the chosen browser.03Provide a no-AI fallback.04Measure quality, latency and initial download.05Describe privacy conditions.06Test accessibility and mobile compatibility.

Creators and designers

Evaluate the accessibility of a generative assistant

Apply WCAG principles to dynamic content.

45–75 min0 / 6 steps

Open steps and start

01Test the complete keyboard journey.02Check labels for inputs and controls.03Announce loading and error states.04Avoid unexpected focus changes.05Allow users to inspect sources and correct actions.06Complement automated tests with human evaluation.

Developers

Describe a resource with JSON-LD

Keep metadata consistent with visible content.

40–60 min0 / 6 steps

Open steps and start

01Choose a resource and persistent identifiers.02Represent type, title, language and author.03Set an appropriate JSON-LD context.04Check syntax and expected values.05Compare properties with visible content.06Avoid promises of AI search ranking.

Educators and readers

Teach verification of generated content

Distinguish fluent answers from evidence-led conclusions.

50–75 min0 / 6 steps

Open steps and start

01Choose a topic with accessible original evidence.02Compare two generated claims.03Identify citations that truly support them.04Check dates and contradictions.05Require a nuanced, justified conclusion.06Evaluate reasoning rather than surface style.

Learn it. Then do it

Pick your workshop

Publish AI-assisted Web content

Build an evidence-led RAG assistant

Design a safer agentic experience

Verify a generated answer

Examine media provenance

Audit an AI-generated web interface

Choose an AI tool for your organization

Run trustworthy research monitoring

Design an AI Web crawler policy

Publish a useful page for generative search

Understand and audit an A2A agent card

Prototype a low-risk WebMCP tool

Secure a remote MCP connection

Test indirect prompt injection in a Web assistant

Evaluate a RAG assistant with evidence

Investigate media provenance

Test an on-device AI feature

Evaluate the accessibility of a generative assistant

Describe a resource with JSON-LD

Teach verification of generated content