In short: A goal you cannot measure precisely is not a goal, it is a wish with a number on it. Three shapes survive here: presence in a named set of questions, a named failing step closed, and a measurement repeated on schedule.
Each one is checkable by somebody who did not run it, which is the only test that matters when the reporting arrives.
The reason this is harder than in search is that the thing you are counting moves on its own. A question does not agree with itself between days, so any target has to be stated against that spread rather than against a single reading.
The Test A Goal Has To Pass
Before the shape, the test. Could somebody who was not involved check whether you hit it, using only what you wrote down at the start?
That rules out most of what gets written on a slide. "Improve AI visibility" fails it.
"Increase citations by 30%" fails it too, for a less obvious reason: nobody can check it unless the question set, the platforms, the number of runs and the date were fixed in advance. If all four were fixed, the goal is really the third shape below.
A goal here: a statement about a fixed question set, measured a fixed way, at a fixed interval, with a threshold that clears the noise you measured. Anything missing one of those four is a direction, and directions are useful in a different meeting.
Three Shapes That Work
- PresenceNamed in the answer to a fixed set of questions, on named platforms, at a stated rate across repeated runs
- A step closedA named failure from the sequence fixed and verified: reachable, extractable or resolvable, each with its own evidence
- A measurement keptThe baseline re-run on schedule, with the spread recorded. Unglamorous, and the one that makes the other two mean anything
Presence in a named set
"Named in at least four of our ten category questions, on ChatGPT and Perplexity, in at least two of three runs, by the end of Q3."
Long, and every clause is doing work. The question set is fixed so it cannot drift. The platforms are named because they cite differently. Two of three runs is the threshold that clears variance. The date makes it checkable.
A step closed
"Every page in the top twenty is served in the initial HTML and indexed in Bing as well as Google, verified by the end of March."
This is not a visibility goal and that is the point. It is the work the visibility goal depends on, and it has the rare property of being entirely within your control.
A measurement kept
"The baseline set is re-run in the first week of every quarter, by the same person, in fresh sessions, with the spread between runs recorded."
It sounds like process rather than a goal. It is the one that decides whether anything else you claim can be defended, and it is the first thing dropped when a quarter gets busy.
Thresholds Against The Spread
Every threshold has to sit above the noise you measured, not above zero.
If your ten questions disagree with themselves across three runs by a margin of two mentions, then a goal of "three more mentions" is a goal you can hit by running the test on a different afternoon. State the threshold as a multiple of the spread, or state the spread beside it so the reader can judge.
Say it in the goal, not in the footnote: "at least two of three runs" is a threshold that survives variance. "Cited more often" is not, and the difference will only surface at the end of the quarter, in the meeting where it is most expensive.
Timeframes That Match The Mechanism
The four steps move at different speeds, so a single quarterly cadence over all of them produces a report where three things look stalled.
Reachability and extraction can be fixed and verified inside a quarter. Resolution takes weeks to do and longer to show. Association is measured in quarters and sometimes years, because it depends on other people writing things.
Setting a quarterly target on association is how a team concludes that GEO does not work, four weeks before the work would have started to show.
Key Takeaways
- A goal here fixes four things: the question set, the method, the interval and a threshold. Missing one makes it a direction.
- The test is whether somebody uninvolved could check it from what you wrote at the start.
- Three shapes work: presence in a named set, a named step closed, and the measurement kept.
- Thresholds sit above the spread you measured, not above zero, and the spread belongs in the goal rather than in a footnote.
- Match the timeframe to the step. A quarterly target on association is how a team decides GEO does not work just before it would have.
Check yourself
Before you move on
Not scored, not recorded, and not part of the certificate. Both answers are settled by a sentence in this lesson, and the reasoning appears whichever option you pick.
- 01
What are the four things a goal here has to fix to be checkable?
- 02
Your ten questions disagree with themselves by about two mentions across three runs. What is wrong with a target of three more mentions?