CiteUnseen

Rubric

Measurement readiness

Rubric v0.8 · 3 checks · 5 of 100 points

Will you be able to tell whether any of this worked?

Without a baseline fixed before results are known, any later movement can be explained after the fact. This is what makes a re-audit mean something.

Last updated · Written by Chris Brock, who wrote the rubric and signs the audits that use it.

Why this dimension exists

The last five points are about whether you can tell if any of this worked.

A site that makes twenty changes and sees a number move cannot say which change moved it, or whether it moved on its own. The three checks ask for the minimum that makes the question answerable: a written set of questions you care about, recorded before the changes; a way to see traffic arriving from answer engines; and a stated interval for looking again.

Measurement here is hard. Answer engines are non-deterministic: the same question asked five times returns five different answers, so one check is an anecdote. Referral headers from AI surfaces are incomplete and differ between platforms. The rubric solves neither. It scores whether you can notice a change at all, a lower bar than attribution that most sites still fail.

The guide for this dimension.

  • Audits and monitoring: what a monitoring tool measures that an audit does not, and the order a buyer wants them in.

The 3 checks

None of these is deterministic. Every one is a reading rather than a measurement, so the free scan names them as not covered rather than guessing.

MR-01 — Baseline question set

2 points · Rated by a person

What is tested. Define a versioned set of customer questions, brand variations, comparison questions, and geographic variations.

What is recorded as evidence. Question set, rationale, target engines, baseline date.

What fails it. Random prompts selected after results are known; questions unrelated to real customer intent.

How to fix it.

  1. Write down the questions you want to be the answer to, before you change anything.
  2. Record today's answer from each engine you care about, with the date and the exact prompt.
  3. Keep the wording fixed. A question you reword is a baseline you have lost.
  4. Store the results where you can compare against them later.

How to confirm it is fixed. Re-run one question a week later and confirm you can tell whether anything moved.

Reference.

MR-02 — Referral and conversion attribution

2 points · Rated by a person

What is tested. Review whether analytics can isolate relevant AI referrals and connect them to meaningful site outcomes.

What is recorded as evidence. Approved analytics configuration or implementation backlog.

What fails it. Reporting only impressions or mentions without any connection to qualified behaviour.

How to fix it.

  1. Set up analytics that can distinguish AI-referred traffic from search and direct.
  2. Record the referrer where the engines send one, and watch direct traffic for the ones that do not.
  3. Define what a conversion is before you measure it.
  4. Connect Search Console so you can see queries alongside the referrals.

How to confirm it is fixed. Check that a visit arriving from an AI answer shows up somewhere you will look.

Reference.

MR-03 — Recheck discipline

1 point · Rated by a person

What is tested. Define when the rubric, crawl sample, prompt set, competitors, and evidence will be rechecked.

What is recorded as evidence. Versioned schedule, change log.

What fails it. Changing methodology between reports without disclosure; treating one volatile result as a trend.

How to fix it.

  1. Put a recheck in the calendar rather than leaving it to memory.
  2. Re-run the same baseline questions, unchanged, on a fixed interval.
  3. Record what changed on the site between rechecks so a movement has a candidate cause.
  4. Keep the history. A single reading is not a trend.

How to confirm it is fixed. Look at two consecutive rechecks and confirm you can say what changed and why.

Reference.