The Glass Agency.

🟢 Live page health (checked by your browser, right now)

Your own browser verifies these as you read — not our claim, your check.

What we got wrong (the correction log)

• October 2026: our Agent i Japan article launched with the wrong CTR baseline — it said 5.8% → 1.8% for #1 rankings. The measured figures are 2.9% → 1.8%, a 37.8% decline (Ahrefs 300,000-keyword study; Japan analysis by GMO AI Search Lab). An external reviewer caught it within hours of publishing on 4 October; we corrected it the same day everywhere it appeared, including the machine-readable FAQ, and dated the correction note on the article itself.
• October 2026 (2): the September crawler census said AI training traffic outweighed citation traffic “five to one”; the correct ratio against both reader-facing surfaces is three to one (the five-fold figure only holds against the search crawler alone). A ratio was printed without re-running the division. Caught by external review and corrected the same day; the underlying counts were never wrong.
• July 2026: our audit badge once said “100% LLM-Optimized” — an unverifiable claim. An independent reviewer challenged it; we changed it to describe what the tool actually checks.
• July 2026: a case-study claim said “every page” when one page had slipped through. Corrected to “across the site,” with the miss admitted in the case study itself.
• July 2026: we initially mis-rated a client-site finding as critical (a placeholder phone number that was actually harmless form text). Re-verified, downgraded, documented.
• July 2026: our live audit counter stored its tally inside the website folder, so every site redeploy silently reset it to zero. Roughly 20 real audits from the first week were lost. We moved the data file outside the deploy path and restarted the count from a conservative 20 — the lowest defensible figure, not the highest.

Why publish mistakes? Because a correction log is the only credential that cannot be faked.