jobshift

Is AI language showing up in Philippine job ads yet? A small scrape comparing job postings on Kalibrr from before ChatGPT existed against a current sample, cross-checked against PSA employment data.

Live site: ralphptorres.github.io/jobshift

Writeup: docs/writeup.pdf

Finding

Postings mentioning explicit AI terms ("AI," "ChatGPT," "AI tools") rose from 2% to 5% (7/317 archived, 15/278 current; Fisher's exact test, p = 0.0495). A broader "automation" bucket (9% to 10%) is not statistically significant, consistent with that bucket being diluted by pre-ChatGPT martech and RPA terminology. See the writeup for the full method, two real methodology corrections made along the way, and honest limitations.

Repo layout

• src/ - scraping and analysis scripts (see below for the pipeline order)
• data/ - scraped postings and computed results (JSON)
• fig/ - charts used in the writeup
• docs/ - the GitHub Pages site, plus the writeup source (writeup.typ) and compiled PDF

Reproducing

Requires uv. Runs the full pipeline end to end:

• scrape.py - current + archived (search-page) postings from Kalibrr
• scrape_detail.py - archived individual Kalibrr job-detail pages from the Wayback Machine, to grow the pre-ChatGPT sample past what search-page snapshots alone give
• merge_archived.py - merges the two archived pulls above
• psa.py - PH employment-by-industry annual series from PSA's PXWeb API
• analyze.py - tags postings against the AI-term lists
• build_evidence.py - builds the evidence table (every posting that matched an AI-adjacent term)
• chart.py - generates the writeup/site charts

uv run src/scrape.py
uv run src/scrape_detail.py
uv run src/merge_archived.py
uv run src/psa.py
uv run src/analyze.py
uv run src/build_evidence.py
uv run src/chart.py

To view the site locally:

uv run src/serve.py

Data files

• data/current.json, data/archived.json - deduplicated, cleaned Kalibrr postings (see src/scrape.py and src/scrape_detail.py for exclusion rules: BPO staffing-agency spam and non-Philippine postings are filtered out)
• data/archived_detail.json - the raw individual-job-permalink pull, before merging with the search-page archived set
• data/results.json - term-match counts and totals, produced by src/analyze.py
• data/psa_employment.json - PSA Labor Force Survey data by industry group, produced by src/psa.py
• data/ai_hits.json - every posting that matched an AI-adjacent term, with snippets; powers the evidence table on the site

License

MIT, see license.txt.