A measurement model for SEO and answer engines

Rank tracking is not enough, and it is not useless

A rank report tells you whether classic search is willing to show your URL for a query you care about. That is still money. It does not tell you whether an AI overview replaced the click, or whether ChatGPT named a competitor and linked their guide.

Keep rank tracking for a tight query set. Stop tracking thousands of phrases nobody will act on. Pair the set with Search Console impressions, clicks, and the queries that already produce pipeline. Vanity positions on informational head terms are how teams hide a flat commercial trend.

Four numbers, reviewed together

We use four views. None of them is a single source of truth. The argument is in the combination.

• Coverage: are the priority URLs indexed, canonical, and internally linked?
• Demand captured: impressions and clicks on the commercial query set
• Answer presence: on a fixed prompt set, were you named, linked, or missing?
• Outcome: demo, call, store visit, or revenue you can honestly connect

Share of answer, defined honestly

Share of answer is the percent of tracked prompts where your brand is named or your URL is cited. Publish the denominator. “3 of 20 prompts this month” is a measurement. “Share of answer is up” is a slogan.

Sample the prompts the way a buyer asks, not the way a keyword tool phrases them. Include a follow-up if that is how the product is actually chosen (“which of these fits a 40-person team?”). Record the competitor who was cited when you were not. That list is the content roadmap.

Expect noise. Model updates move citations without a ranking change, and the reverse happens too. Call a trend only after several reviews point the same way. If you need language for stakeholders, say “named in 11 of 20 tracked prompts, up from 6,” and attach the prompt list.

What not to put on the dashboard

Skip composite “AI visibility scores” you cannot explain. Skip average position across mixed intents. Skip citation counts from a tool whose prompt set you have never seen. If a vendor will not show the prompts, the number is not yours.

Also skip last-click arguments that erase research. A guide that is cited and then not clicked can still shape the shortlist. Note it. Do not invent a revenue number you cannot defend, and do not ignore it because analytics marked it as zero.

A monthly review that fits in an hour

Open with coverage exceptions: pages that fell out of the index or started cannibalizing. Then clicks and impressions on the query set, with three queries to investigate, not thirty. Then the prompt log: new citations, lost citations, and the competitor passage that beat you. Close with one decision. Rewrite an opening, build a missing comparison, fix an entity mismatch, or leave it alone.

That hour is the operating system. Tools can collect the inputs. They should not make the decision. If you want the first version of this scorecard built for your category, start with a visibility audit. The background on why the two surfaces share a roadmap is in what AEO changes about SEO.