makefield

Week 11 · fixing it

Correct and Engineer: turning findings into fixes

An audit tells you what's wrong. These two stages turn that into a ranked list of changes — the record first, then the machinery.

From diagnosis to work

Two kinds of fix

An audit locates the problem; now you fix it. The fixes fall into two stages. Correct is about the narrative — the account the model reads and repeats. Engineer is about the machinery — making your information legible to a machine. Most companies need both, but in that order: there's little point engineering clean structure around a story that's wrong.

Correct

Make the true account the easy one

Most visibility failures are narrative failures. The model describes you from whatever it found — often a third-party summary that's thin, dated, or framed by a competitor. You can't edit the model, so correcting means making the true, clear account of you the easiest one for a model to find and repeat: fixing your own pages, and fixing the record on the outside sources that carry you. When the clearest, most-corroborated version of you is also the accurate one, that's the version that gets repeated.

Engineer

The machinery, applied

Then the technical work — the machinery dimensions turned into changes: documentation structured so a model can extract specific facts, valid and current structured data, and an earned presence in the sources AI trusts. This is the same three machinery dimensions from earlier in the course, now read as a to-do list rather than a diagnosis.

The discipline that matters most

Rank by impact, not by appetite

The trap here is trying to fix everything. The discipline is to rank — order the fixes by likely impact against effort, and start with the cheap ones that move the answer most. A single wrong fact carried by one heavily-cited source can outweigh a dozen schema tweaks; fix that first. The audit's per-dimension scores are what tell you where the gains are, so the fix list is derived from the measurement, not from a generic checklist — which is exactly why generic checklists underperform.

Next — Measure & Monitor: proving a fix actually moved the answer, and watching it over time. New here? Start with the overview.