AI Skill Report Card
Tracking Product Feature Updates
Quick Start14 / 15
Given a jumbled set of search results/articles about a product (e.g., "Gemini Spark"), produce:
- A one-paragraph "what is it" summary
- A chronological timeline of confirmed updates (with dates and sources)
- A current feature list, separated from rumored/upcoming features
- Explicit flags for conflicting or unverifiable claims
Don't just list links back to the user — extract and structure the underlying facts.
Recommendation▾
Add a second, shorter example (e.g., a smaller/simpler product update) to show the workflow scales down, not just to a 30+ source mega-example
Workflow14 / 15
Progress:
- Step 1: Identify the core entity (product/feature) and its parent product
- Step 2: Extract every dated claim and tag it with source type (official/press/blog/social/forum)
- Step 3: Deduplicate — many sources repeat the same announcement
- Step 4: Build a chronological timeline from earliest to latest
- Step 5: Separate "what it does now" from "what's planned/rumored"
- Step 6: Flag contradictions, unverified claims, and low-credibility sources
- Step 7: Write the summary, timeline, and caveats
Step 2 detail — source tiering:
- Tier 1 (authoritative): official blog.google, support.google.com, gemini.google — treat as ground truth
- Tier 2 (press): TechCrunch, CNET, PCMag, Mashable, CNBC — reliable for reporting, may lag or simplify
- Tier 3 (secondary/blogs): jetstream.blog, igmGuru, Devoteam, Medium — often just re-reporting Tier 1/2, verify against those
- Tier 4 (social/UGC): Reddit, YouTube, X, Instagram, Facebook — useful for user experience/sentiment, NOT for factual claims unless corroborated
- When a claim appears only in Tier 3/4, mark it "unconfirmed" unless it traces to Tier 1/2.
Step 3 detail — dedup by event, not by article. Many articles describe the same launch (e.g., 5 articles all describing the May 19 I/O announcement). Collapse these into one timeline entry with multiple sources cited.
Step 6 detail — contradiction flags. Note when sources disagree (e.g., availability tied to "Google AI Ultra" in one source vs. "Google AI Pro" in another) rather than silently picking one.
Recommendation▾
Include a template/skeleton output structure separate from the fully-worked example, so Claude can quickly reference the expected format without parsing the long example
Examples17 / 20
Example 1: Input: 30+ search results about "Gemini Spark" spanning May–September 2026, mixing official blog posts, press coverage, and Reddit/YouTube commentary. Output:
Recommendation▾
Clarify how to handle cases with very few or no Tier 1/2 sources at all (i.e., when the entire corpus is Tier 3/4) since that's a plausible edge case not explicitly addressed
What is Gemini Spark?
A 24/7 personal AI agent from Google, announced at I/O 2026, that works in the background across Gmail, Calendar, Workspace, Chrome, and Google Photos even when the user's devices are off.
Timeline
- May 19, 2026 — Announced at Google I/O 2026 alongside Gemini 3.5 Flash; positioned as agentic assistant with Gmail integration. [blog.google, TechCrunch, CNBC, CNET, Mashable]
- Jun 2, 2026 — Made available in the US. [PCMag]
- Jun 21–30, 2026 — Adds Dropbox, Zillow Rentals connectors; more (Adobe, GitHub, Notion, Slack, Samsung, Spotify, CapCut) "confirmed for summer 2026." [igmGuru, Devoteam — Tier 3, corroborate before citing as fact]
- Jul 15–17, 2026 — Workspace upgrades: >50% faster, "smarter sourcing," task/skill setup support. [9to5Google, jetstream.blog]
- Jul 16, 2026 — Rolled out to Google AI Pro subscribers (previously Ultra-only). [jetstream.blog, 9to5Google]
- Jul 30–31, 2026 — Chrome integration ("auto browse"), global expansion. [blog.google Gemini Drop]
- Sep 4–9, 2026 — Google Photos integration (edit/curate albums, photo→ calendar events); connected with Chrome. [TechCrunch, blog.google, BGR]
- Sep 15, 2026 — Gemini 3.8 Live models announced (related but distinct from Spark itself). [blog.google]
Confirmed capabilities (current)
- Background task execution across Gmail, Calendar, Workspace
- Chrome auto-browse for web errands
- Google Photos editing/curation
- Custom connectors (Dropbox, Zillow Rentals confirmed; others pending)
- Available to Google AI Pro and Ultra subscribers in the US, expanding globally
Unconfirmed / conflicting
- Pricing tier gating varies by source ($100/mo Ultra-only vs. later Pro-tier rollout) — reflects a real timeline change, not an error, but should be dated explicitly.
- Full connector list (Adobe, GitHub, Slack, etc.) reported only in Tier 3 sources as of the given results — flag as "reported, not verified via official channel."
Best Practices
- Always date-stamp claims; "recently" or "now" from an article is meaningless without the article's publish date.
- Prefer the official product/support page for current-state feature lists; use press/blog timeline for history.
- When a product is mid-rollout (regional/tier-gated), state the rollout status explicitly rather than a single "available" verdict.
- Cite sources inline or in brackets so the user can verify — don't present synthesis as if self-evidently true.
- Distinguish the core product update from adjacent announcements (e.g., new model versions bundled into the same event) that get conflated in coverage.
Common Pitfalls
- Treating every repeated headline as independent confirmation — five articles about the same I/O announcement is one data point, not five.
- Citing Reddit/YouTube titles as factual product claims ("10x better," "insane update") — these are engagement bait, not specs.
- Ignoring date gaps that indicate a claim is now stale (e.g., "coming this summer" from a June post vs. what actually shipped by September).
- Collapsing distinct events (initial launch vs. tier expansion vs. feature add) into a single blurred "release."
- Missing explicit tier/access gating (Ultra vs. Pro vs. region) when summarizing "availability."