Acceptance‑Test Templates for Agentic Discovery: 8 JSON‑LD + Playback Artifacts That Make Features AI‑Citable

AI agents prefer machine‑readable entity graphs they can traverse. An acceptance test should include a JSON‑LD snippet that declares the feature as a first‑class entity (Article, SoftwareApplication, HowTo, Dataset, or Product depending on context). The snippet must appear server‑rendered on the page and validate with Schema.org and the Rich Results validator.

Acceptance criteria (pass/fail): the page returns 200, contains the JSON‑LD block, and the schema validates with no critical errors. As a quick test, fetch the page HTML and find the <script type="application/ld+json"> block — if your CI fails to return it, agents will likely skip your content.

FAQ and HowTo are the highest‑leverage schema types when you want direct answers extracted and cited. The acceptance test should include at least one clear Q→A or step list that exactly matches a user question the agent might ask.

Below are two minimal templates you can paste and adapt. Your CI should assert both appear on the page and that the text in the JSON‑LD matches visible HTML.

AI agents can treat short playback artifacts (5–15s micro‑UX videos) as evidence that a feature is discoverable and usable. An acceptance test should include an instrumented micro recording that demonstrates the exact click/tap sequence from landing to the feature outcome, plus a VideoObject JSON‑LD block that points to the hosted clip and provides a transcript and timestamps.

Acceptance criteria: the video loads in under 2s from your CDN in a simulated 4G test, the transcript contains the key intent phrase, and the VideoObject JSON‑LD is present and validated. Agents can extract timestamps and transcript lines when deciding to cite a step or example.

To prove agents can both find and verify a feature at scale, your acceptance test must include telemetry: production events with stable event names and attributes an agent (or a verification script) can query. That telemetry demonstrates real user interaction and reduces the chance an agent cites a feature that only exists in docs.

Provide a minimal telemetry schema and example queries (SQL or Elastic DSL) that return a small sample of successful feature completions within the last 30 days. Your CI should run the queries against a staging or production replica and fail if no successful events are found.

An llms.txt or small agent manifest acts like a robots.txt for AI agents — it tells agents where to find canonical JSON‑LD, OpenAPI endpoints, and the preferred contact/authority for the domain. Include a simple manifest as part of the acceptance test so agents can locate your machine‑readable assets quickly.

Acceptance criteria: the manifest is reachable at /.well-known/llms.txt (or /llms.txt), lists the JSON‑LD endpoint, and the manifest URL appears in your sitemap and root HTML within a link rel or meta tag.