PondralThe paraphrase problem
Models rarely cite verbatim. They paraphrase. Pages built around a single clever turn of phrase or a specific keyword cluster get distorted in transit; pages built around durable facts and clear comparisons survive.
Test for paraphrase resilience by asking three different models the same question three different ways. If your page is cited correctly across all nine variants, your positioning is resilient. If it falls apart on the second rephrasing, it isn't.
Writing for resilience
Anchor every claim to a specific, verifiable number or comparison. "Faster" gets paraphrased away; "2.4× faster than the median competitor on a 1k-record benchmark" survives.
Avoid metaphors that depend on industry insider context. They paraphrase badly into adjacent contexts and your brand picks up associations you didn't intend.
State the limitation. Pages that include a clear "X is not a good fit when…" paragraph are cited more often *and* more accurately than pages that only list strengths. Models reward honesty because honesty looks like reference material.
- Test with paraphrased prompts, not just the literal one.
- Anchor to verifiable numbers.
- Naming your limitations is a citation strategy.