An open-source project called ShieldFont aims to deter AI web scrapers by making a webpage appear normal to human readers while presenting altered, plausible-looking text in its raw HTML. It matters because it offers website owners a client-side alternative to server blocks and robots.txt rules, which scrapers may ignore, potentially preventing their writing being used for AI training.

ShieldFont uses OpenType glyph substitution features normally associated with ligatures, but applies them to whole words so that the displayed words differ from those a scraper reads. It replaces roughly a quarter of words, drawing substitutions from around 250 tightly defined grammatical and semantic groups to keep the scraped text coherent enough to be ingested rather than rejected. For example, text displayed as “good luck reading this” could appear in HTML as “good comfort reading this”.