Key points
- Motoko Hunt argues translation memory, built to maximize reuse, makes local pages look like duplicates to AI systems, which then favor the canonical U.S. or main English version.
- Retrieved passages lose their URL context, so a buyer in Germany may see U.S. availability or a Spanish query from Spain may get Mexico-oriented content.
- Her proposed fix: a fact registry by market, content split into global-invariant and market-variable parts, and local teams supplying evidence and questions while AI drafts and assembles.
- She recommends running fixed local-language prompts across major AI assistants on a schedule and tracking market leakage as a key metric.
Why it matters for marketers: Hreflang and high translation reuse no longer guarantee local visibility in AI answers, so global teams need market-specific facts, wording and measurement, not just faster translation.
In a column for Search Engine Journal, international SEO consultant Motoko Hunt argues that the standard global content model is breaking down in AI search. That model relies on translation memory, machine translation, shared templates and metrics such as cost per word and TM leverage, all aimed at publishing one message cheaply in many languages.
Her core point is that AI search builds answers from passages, not whole pages. When markets carry near-identical content, systems tend to process one canonical version in full, and a retrieved passage no longer carries the URL that signaled its market. The result, she says, is that the strongest instance of your content wins, usually the U.S. or main English site. That can mean wrong product availability in Germany or Spanish answers for Spain drawn from Mexico-focused sources.
For marketers, the implication is that using AI only to translate more pages faster produces more of the same. Hunt proposes a different setup: a fact registry that records which markets each claim applies to and whether it is approved, a split between content that must be identical and content that must vary, and local teams supplying evidence, questions, claim approval and judgment of fit. AI handles drafting, structuring and checking.
She also offers passage-level habits: name the market inside the sentence, date every figure, keep claims with their qualifiers, use question-shaped headings and keep product names identical.
Next, she suggests measuring what AI assistants actually say in each market. Build fixed prompts in local languages, run them on a regular schedule, and watch market leakage most closely, meaning how often another market's content answers a local query.
Sources
This brief is our own summary. Read the full reporting at the source.
