A general Philippine job postings scrape, pre- vs post-ChatGPT, cross-checked against PSA employment data.
595 Philippine job postings were scraped from Kalibrr across thirty-one general sectors: marketing, retail, logistics, HR, healthcare, education, real estate, IT, finance, e-commerce, media, and others. 317 are archived (combining Wayback Machine search-result pages with a larger pull from individual job-posting permalinks) from Oct 2021–Sep 2022, before ChatGPT's Nov 2022 launch. 278 were scraped live in Jul 2026. Each posting was tagged against explicit AI terms (ai, chatgpt, ai tools, and similar) separately from a broader, fuzzier automation bucket (automation, automated, rpa), since that bucket kept catching pre-ChatGPT martech terms like "marketing automation." The explicit-AI shift (2% → 5%) is statistically significant (Fisher's exact test, p=0.0495). The broader automation bucket (9% → 10%) isn't (p=0.89), consistent with that bucket being diluted by pre-existing martech and RPA terminology.
Total economy-wide employment is too blunt a comparison for a scrape of specific job categories, so PSA's Labor Force Survey (pulled directly from PSA's own PXWeb statistical API) is narrowed here to the six PSIC industry groups closest to the sampled sectors: information and communication, professional/scientific/technical, financial and insurance, administrative and support, education, and human health. Together they cover about 91% of the current-side sample by job function, and their combined employment keeps climbing every year since the 2020 COVID dip, ChatGPT launch included.
Every posting behind the 2% to 5% figure, with the matched term and surrounding text. Default view is explicit AI terms only, the statistically significant half of the finding. The automation bucket can be added to see the noisier, non-significant terms it contains. Click a row to see the matched text in context.
One real 2021 posting, rewritten using Claude Sonnet 5 in one shot as if written today, grounded in the phrasing of real 2026 AI-mentioning postings from this dataset rather than free-form guessing. Diffed against the original below. Not evidence for the statistic above, just a check on whether a grounded prediction matches what actually shows up in real postings.