Same AI Question, Different Language: A 67,200-Response Audit
An audit of GPT, Claude and Gemini found that responses to the same Ukraine-war statement bank varied across 112 language and script conditions.
Tagged collection
Latest articles tagged “Gemini”
Open an article for the complete method, evidence, limitations, and a focused reader discussion.
-
-
Gemini Antigravity September Preview: Parser Changes Before October 5
Google’s September Antigravity preview replaces the May build. Local tool parsers need tests for PascalCase parameters and line-range edits before October 5.
-
Google Tests Publisher Payments for Contributions to AI Answers
Google is testing payments when publisher content significantly contributes to AI answers. Here is what is confirmed, undisclosed and worth measuring.
-
Google Election Answers Across Search and Gemini: Source Audit
Google is adding U.S. election information across Search and Gemini. This privacy-safe audit separates official authorities, answer sources and media provenance.
-
Gemini 3.8 Flash Is Live: Test Search Citations Before Switching
Gemini 3.8 Flash is generally available with Google Search grounding, but capability is not evidence of better citations. Use a matched query-and-claim ledger before migrating.
-
Google UCP Is Not SEO: Prepare for Checkout in AI Mode and Gemini
Universal Commerce Protocol is an early-access checkout integration for eligible AI Mode and Gemini experiences—not a ranking shortcut.
-
How to Test Gemini 3.7 Flash Before Switching Grounded Answers
Test Gemini 3.7 Flash with frozen tasks, claim-level citation checks, tool and schema failures, accepted-task economics, reviewer agreement and a canary rollback rule.
-
Gemini File Search Visual Citations: Audit media_id and page_numbers
Audit Gemini File Search visual citations by preserving file hashes, page numbers, media IDs, answer claims, cited evidence, and manual verification.
-
Gemini 3.6 Flash Deprecates Sampling Controls: Update Grounded-Answer Test Harnesses
Migrate Gemini test harnesses after temperature, top_p, and top_k deprecation while preserving model, grounding, citation, latency, and usage evidence.
