AI-driven search is replacing keyword search in many uses. The shift is not yet complete, but for many ordinary queries (including queries about people), the answer presented to the searcher is now a generated summary rather than a list of links. The implications for what is found about a private person are different from what the keyword era taught.

A keyword search returns documents. The searcher chooses which to read. The documents themselves, in their original form, are what is encountered. A generated summary, by contrast, has read the documents on the searcher's behalf and produced a synthesis. The original documents are still there in many cases, but the searcher's typical reading is now of the synthesis.

The synthesis is shaped by the training of the model and by the source weighting it applies. A source that is well-represented in the training data is more likely to be drawn from than a source that is not. A source that is current is more likely to be drawn from than one that is stale. A source that is consistent with other sources is more likely to be trusted than one that contradicts them. The reader receiving the synthesis is, in effect, receiving a particular kind of consensus rather than the underlying documents.

For a private person, this changes the nature of exposure. The picture presented in a generated summary is the picture the model has synthesised. Inaccuracies in the synthesis (a misattribution, a conflation with another person, a date error, an unfounded inference) become the picture. Correction is harder because the synthesis is not a document with an author to correct.

· · ·

The direction is unmistakable. The work of the desk reflects it: the assessment includes what generated search now says, not only what the original records contain. Where the synthesis is wrong, the work of addressing it is its own undertaking, and one that the older procedures (corrections to articles, removals of pages) were not built to handle.