The Entity and Answer Passes

Pass Three: Resolution

Two parts, and the first takes twenty minutes.

The naming inventory

List every form of the company and product name in use, across page titles, the footer, schema markup, the careers site, social profiles, invoices and conference bios. Then pick one canonical form and note where the others appear.

The finding is the count. Four or five live variants is common and nobody has ever written them down in one place, which is why nothing gets consolidated: the problem is invisible until it is a list.

The brand question

Ask each assistant about the brand by name, three times, in fresh sessions. Record what comes back.

Pass Four: Association

The prompt run from module seven: ten questions, no brand names in the category ones, three fresh runs per platform.

The test plan builder will generate the set and the running order from your category and markets. You run it yourself, in the actual assistants, because an API call with a search tool attached is a different product and gives a different answer dressed up as the same one.

Record the sources, not just the answer

For every answer, write down which sites were named. That list is the most actionable thing the audit produces and it is the easiest thing to fail to capture, because the answer is what you were looking at.

It tells you where the category is being read from, which is the target list for the off-site work. Your own presence or absence tells you whether there is a problem; the source list tells you where to go about it.

Turning Four Passes Into One Page

The audit report has five parts and fits on a page.

The failing pass, named. One, not three.

The evidence, in a form somebody else could check: the log line, the index count, the passage that does not stand alone, the list of name variants, or the answers with the source lists.

What it costs, stated as the thing that cannot happen while it is unfixed rather than as a percentage nobody can verify.

The fix, and who does it. Technical passes name a developer. Editorial passes name a writer. Association names whoever owns relationships with the publications on the source list.

The date the same measurement runs again, with the spread from this run recorded beside it so the next one has something honest to compare against.

• The failing passNamed. One, not three
• The evidenceIn a form somebody else could check: the log line, the index count, the passage that does not stand alone, the name variants, the answers with their source lists
• What it costsThe thing that cannot happen while it is unfixed, rather than a percentage nobody can verify
• The fix, and whoA developer for the technical passes, a writer for the editorial ones, and whoever owns relationships for association
• The dateWhen the same measurement runs again, with this run's spread recorded beside it

What Not To Put In It

No score. No competitor ranking table. No list of twenty recommendations sorted by effort, which is the format that makes an audit feel thorough and guarantees that the blocking problem gets the same attention as a nice-to-have.

And no findings from passes that were not assessable. If reachability failed, the association pass has no verdict, and writing one anyway is the fastest way for the whole document to stop being trusted.

Key Takeaways

• Resolution is a naming inventory plus the brand question. The finding is usually the count of live name variants, which nobody has written down.
• Confidently wrong when asked by name means resolution failed, and no association work lands until it is fixed.
• Record which sites were named in every answer. That list is the target for off-site work, and it is the part an audit most easily loses.
• Three runs per question, minimum. The disagreement between your own runs is the bar every future comparison has to clear.
• The report is one page: failing pass, evidence, cost, fix, and the date it is measured again. No score, no ranking table, no list of twenty.