Shared source registry
11 shared definitionsFeeds, APIs, catalogue searches, and shared attribution definitions. Automated, manual, and legacy paths are labelled in the table below.
How it works
How AI Canada Pulse collects, checks, and summarises
AI Canada Pulse is an independent briefing focused on Canadian AI stories. The live site blends public data feeds, rule-based story sorting, cached AI-assisted summaries, topic matching, daily digests, world AI context, robotics coverage, reviewed quote records, and curated institutional references.
Three evidence layers
How source evidence enters the site
The site uses three source layers because a live feed, a curated evidence record, and a specialist-page adapter have different review needs. The table below describes shared automated and attribution definitions; it is not the complete evidence library. For question-led records and their limitations, see the Sources page.
Curated evidence library
52 records · 9 tracksQuestion-led records presented on Sources, grouped by subject tracks and linked to related pages. Page-specific evidence remains separate.
Specialist source systems
Page-specific registriesWorld AI, Robotics, data centres, strategy, global standing, model benchmarks, official updates, and indicators keep their own source receipts.
Shared automated and attribution definitions by type
This chart summarizes the shared registry used by feed, endpoint, status, and source-attribution components. Curated records are counted and specialist records are described separately above.
Source method and cadence matrix
Coverage is strongest when the reader can see how a source enters the site and how often it is checked.
SourceMethodScopeCadence
News sourcesCanadian News Feeds
News feedfetch method
National + regionalcoverage scope
1hrefresh cadence
New AI research2 source sets
Structured data servicefetch method
Nationalcoverage scope
On request, 6hrefresh cadence
GovernmentOpenParliament.ca
Web page checkfetch method
Nationalcoverage scope
6hrefresh cadence
JobsGovernment of Canada Job Bank
Structured data servicefetch method
National + regionalcoverage scope
24hrefresh cadence
Data registries2 source sets
Structured data servicefetch method
Nationalcoverage scope
24h, 12hrefresh cadence
Rules and oversight2 source sets
Structured data service, Manualfetch method
National, National + regionalcoverage scope
12h, quarterlyrefresh cadence
Startups and companiesCanadian AI Startups
News feedfetch method
National + regionalcoverage scope
6hrefresh cadence
Benchmarks and rankingsLegacy synthesized AI country scores
Manualfetch method
Nationalcoverage scope
annuallyrefresh cadence
News sources
| Source | Description | Refresh | Scope | Method |
|---|---|---|---|---|
| Canadian News Feeds English-only Canadian AI discovery across Google News jurisdiction searches, regional and section publisher feeds, BetaKit, CBC Technology, MobileSyrup, IT World Canada, and English Government of Canada Atom feeds; direct-source publication remains separately gated | English-only Canadian AI discovery across Google News jurisdiction searches, regional and section publisher feeds, BetaKit, CBC Technology, MobileSyrup, IT World Canada, and English Government of Canada Atom feeds; direct-source publication remains separately gated | 1h | National + regional | News feed |
New AI research
| Source | Description | Refresh | Scope | Method |
|---|---|---|---|---|
| arXiv.org Research papers in AI and machine learning with Canadian-affiliated authors | Research papers in AI and machine learning with Canadian-affiliated authors | On request | National | Structured data service |
| OpenAlex Open scholarly metadata for AI-related publications by Canadian researchers | Open scholarly metadata for AI-related publications by Canadian researchers | 6h | National | Structured data service |
Government
| Source | Description | Refresh | Scope | Method |
|---|---|---|---|---|
| OpenParliament.ca Independent, non-governmental search layer for AI mentions in House of Commons and committee transcripts; consequential claims should be checked against official Parliament records | Independent, non-governmental search layer for AI mentions in House of Commons and committee transcripts; consequential claims should be checked against official Parliament records | 6h | National | Web page check |
Jobs
| Source | Description | Refresh | Scope | Method |
|---|---|---|---|---|
| Government of Canada Job Bank Monthly AI job postings from Canada's National Job Bank open data | Monthly AI job postings from Canada's National Job Bank open data | 24h | National + regional | Structured data service |
Data registries
| Source | Description | Refresh | Scope | Method |
|---|---|---|---|---|
| Statistics Canada Official business-adoption references plus labour-force and economic indicator series served from a stored snapshot that checks Statistics Canada daily | Official business-adoption references plus labour-force and economic indicator series served from a stored snapshot that checks Statistics Canada daily | 24h | National | Structured data service |
| Government of Canada AI Register and catalogue Searches multiple public records in the Open Government catalogue for AI and algorithmic systems; the linked GC AI Register MVP is the reference dataset | Searches multiple public records in the Open Government catalogue for AI and algorithmic systems; the linked GC AI Register MVP is the reference dataset | 12h | National | Structured data service |
Rules and oversight
| Source | Description | Refresh | Scope | Method |
|---|---|---|---|---|
| LEGISinfo AI-related bills and legislation in the Parliament of Canada | AI-related bills and legislation in the Parliament of Canada | 12h | National | Structured data service |
| Federal and provincial AI rules Curated jurisdiction records for federal rules and six represented provinces; the linked page is the federal digital-government hub | Curated jurisdiction records for federal rules and six represented provinces; the linked page is the federal digital-government hub | quarterly | National + regional | Manual |
Startups and companies
| Source | Description | Refresh | Scope | Method |
|---|---|---|---|---|
| Canadian AI Startups A curated company list matched against the broader Canadian news feed for funding, acquisition, product, partnership, expansion, and IPO signals | A curated company list matched against the broader Canadian news feed for funding, acquisition, product, partnership, expansion, and IPO signals | 6h | National + regional | News feed |
Benchmarks and rankings
| Source | Description | Refresh | Scope | Method |
|---|---|---|---|---|
| Legacy synthesized AI country scores Legacy normalized scores synthesized from Stanford AI Index, Tortoise Global AI Index, and OECD material; current comparisons use the source-specific Global Standing page | Legacy normalized scores synthesized from Stanford AI Index, Tortoise Global AI Index, and OECD material; current comparisons use the source-specific Global Standing page | annually | National | Manual |
Adoption evidence
Why the adoption rates are kept separate
The adoption page uses Statistics Canada as the official baseline, then labels broader survey evidence separately. A worker using generative AI, a small business trying an AI-enabled tool, and a firm's formal deployment of AI inside service delivery are all meaningful clues, but they are not the same measurement.
Latest official business AI use
19.2%AI used to produce goods or deliver services.
Source: Statistics Canada, June 11, 2026
Businesses planning to use AI
14.5%Businesses expecting to adopt AI over the next 12 months.
Source: Statistics Canada, August 27, 2025
Small-business task use
45%Small-business use of generative AI for tasks such as drafting, marketing, ideation, or translation.
Source: Canadian Federation of Independent Business, April 2, 2026
Employees using generative AI
51%Individual employees who report using generative AI at work.
Source: KPMG Canada, November 27, 2025
What executives say
93%Business leaders saying their organization uses AI in some form.
Source: KPMG Canada, November 27, 2025
Government AI use
Thinly measuredFederal AI systems, strategy, register entries, and concrete examples.
Source: Government of Canada, March 2025
Refresh pipeline
What runs automatically
Five-stage refresh pipeline
The pipeline separates heavy publishing work from lighter background summary refreshes and reviewed quote intake.
- 1Daily full refresh
Fetches Canadian stories, refreshes stored AI summaries and topic matches, generates the latest digest, checks for longer explainers, updates Job Bank data, warms rules-and-oversight data, and writes section summaries.
- 2World AI enrichment
Runs on the same protected refresh path so public page views can read stored world-feed summaries without triggering AI generation.
- 3Same-day digest retry
Runs only if the scheduled digest is missing or errored, so a failed morning generation has a second path without duplicating successful work.
- 4Background summary refresh
Refreshes cached story summaries, topic matches, and briefing data every four hours without running the heavier digest and explainer steps.
- 5Quote candidate review
Classifies and deduplicates candidate government AI quotes, then holds them for admin review before public display.
The weekly email is compiled on Mondays during the full refresh and uses a Redis dedupe key so a retry or manual trigger does not send duplicate mail.
The live feed's operational status panel and the public status data service report feed count, digest freshness, AI-summary coverage, and source-list state. Protected status endpoints expose deeper AI-refresh and weekly-email readiness checks for the operator without sending mail.
Manually maintained
Curated datasets
Some data is maintained from known public references rather than pulled from a live data service on every request. These records are reviewed periodically and may lag changes between review cycles. The Robotics pulse combines optional live adapters with typed seed records and deterministic labels so physical-AI coverage remains available even when a live source or API key is unavailable.
| Dataset | Records | Last Verified | Review Schedule |
|---|---|---|---|
| Canadian AI Startups | 48 companies | March 28, 2026 | Every 6 months |
| University AI Programs | 29 programs | March 28, 2026 | Annually |
| Regional AI Reference Data | 13 jurisdictions | March 28, 2026 | Annually |
| AI Events & Conferences | 12 events | April 1, 2026 — review due | Every 3 months |
| Canadian Data Centres | 87 facilities, with sourced first-online years for the AI-era buildout | June 10, 2026 | As cloud, telecom, and AI-compute facilities change |
| Topic Registry | 21 topics | May 2026 | As public topic coverage changes |
| Global AI Standing Sources | 12 source references | July 13, 2026 | As publishers release new editions or benchmark data |
| Robotics pulse | 13 source adapters, 12 seed items, 28 watched entities | July 14, 2026 | Daily adapters plus reviewed snapshot and seed maintenance |
| AI laws and oversight | Federal plus six represented provinces | July 16, 2026 | As legislation, enforcement, or guidance changes |
| Canadian Government AI Quotes | Approved Supabase archive plus pending review queue | Ongoing | Candidate ingest daily; approval by admin review |
AI processing
How AI is used on this platform
AI-assisted writing is precomputed on scheduled or guarded background paths and served from cache. Normal public page requests attach stored results and do not call the AI provider directly.
Stored-output workflow
The visual version of the pipeline separates source collection, rule-based sorting, cached AI assistance, and review limits.
- 1Fetch public evidence
News feeds, data services, manual registries, and curated references enter through source-specific clients.
Source list - 2Sort before summarising
Category, region, and sentiment labels are assigned before AI-assisted summaries or topic matches are served.
Story sorting rules - 3Cache AI-assisted layers
Summaries, briefs, topic matches, digests, and longer explainers are precomputed and read from cache on public pages.
Upstash cache - 4Keep review limits visible
Quote candidates, curated datasets, and uncertain indicators keep human review or limitation notes attached.
Admin review and public limitations
Briefing and explainer text
OPENAI_BRIEF_MODEL defaults to gpt-4o-mini and is used for the daily digest, longer explainers, section summaries, world-feed summaries, and Canadian live-feed brief text unless the environment overrides it.
Story summaries
OPENAI_ARTICLE_MODEL defaults to gpt-4o-mini for per-story summaries, with the brief model used as the fallback path.
Topic fallback
OPENAI_TAGGER_MODEL defaults to gpt-5-nano only for ambiguous stories that do not match the rule-based topic filter.
Quote classifier
OPENAI_QUOTE_MODEL defaults to gpt-4o-mini for candidate quote classification before the Supabase review queue.
Rule-based sorting
Story category, region, and tone use keyword and regex rules. No AI model is involved in that baseline sorting.
Story summaries
OpenAI summaries are generated from story headlines and snippets, date-grounded, cached in Upstash Redis, and read by id first with a headline fallback for older cache entries.
Daily brief
OpenAI generates the headline, intro, main story items, and top original-source links from the top 10 current stories. Daily briefs are stored in Redis for 90 days with error sentinels when generation fails.
What this means posts
OpenAI drafts 400-600 word explainer posts when significance thresholds are crossed, or when the feed needs a fresh post after four or more days.
Section summaries
OpenAI generates one-sentence summaries for live-feed sections such as Stories, Trends, New AI research, Parliament, Jobs, and Stocks.
Topic matching
A hybrid pipeline uses regex keyword matches first, then gpt-5-nano resolves remaining ambiguous stories into topic pages.
World AI summaries
The world feed uses the same guarded refresh pattern as the Canadian feed: summaries are generated in the background and stored before public rendering.
Weekly email
The Monday email is compiled from stored digest and feed data, sent through the newsletter path, and deduped in Redis so retries do not send duplicate messages.
Quote review aid
OpenAI classifies candidate government quotes before they enter the review queue. Public quote records still require approval or manual curation before display.
AI does not:
- select which source sets exist
- modify original upstream records
- replace original-source checks
- approve public quote records by itself
Data quality and limits
What the live feed is good for, and what it is not
- AI summaries, digests, section summaries, topic matches, and longer explainers are navigation aids. They can miss nuance, omit background, or flatten uncertainty.
- The daily digest is generated by the scheduled refresh pipeline and has a same-day retry. If the scheduled digest is missing or errored, the digest page labels the state and falls back to current public-source headlines.
- Story summaries are stored in a cache and can be refreshed through a protected operator-only path when cached summaries are missing.
- News feeds and public data services can change structure, publish duplicates, or omit context. The app deduplicates and filters aggressively, but false positives and missed stories remain possible.
- Market, jobs, benchmark, and economy indicators are context only. They are not investment advice, economic forecasts, or causal evidence about AI effects.
- Curated reference datasets and global AI indices reflect the latest verified snapshot in the repository and may lag real-world changes between review cycles.
- The AI Adoption Compass is a static, source-labelled evidence page. It separates official business use from worker use, small-business task use, survey claims from executives, and productivity scenarios.
Disclaimer
AI Canada Pulse is an independent informational monitoring product with no affiliation to the Government of Canada or any federal, provincial, or territorial agency. It does not provide financial, legal, regulatory, or policy advice. AI-assisted summaries, daily briefs, topic matches, and longer explainers are produced from public-source headlines and short context snippets and may contain omissions or phrasing errors. Always verify important claims with source pages.
Last reviewed: