How to measure organic visibility beyond keyword rankings.

A single average position looks like a measurement and behaves like an average of measurements taken with different instruments. Google’s own documentation for the Search Console performance report says that position values vary by result type and that the number “can mean different things in different situations”, recommending that trends be monitored rather than absolute values compared.

It also documents how the data is grouped: metrics are assigned to a canonical URL and combined across duplicates and devices, and aggregating by property rather than by page changes impression counts substantially. Two reports of the same site can therefore disagree without either being wrong.

None of that makes position useless. It makes it one input among several, and a poor headline for a programme whose goal was never a number on a dashboard.

The question a ranking report cannot answer: for the ten questions we decided to own, do we now have a page for each, is it indexed, and does it appear for the phrasings people actually use?

If the programme began with a decision about which intents to own, the measurement follows from it directly. Each intent is either covered or not, the covering page is either indexed or not, and it either surfaces for the phrasings it was written for or it does not. Those are three yes-or-no questions per intent, and together they say more than a page of averages.

The fourth question is the one most reports omit: when a system describes the business, does it describe it correctly? That is not a ranking question at all, and on most sites nobody has ever checked.

Four questions per intent, in order:

• IS IT COVERED

  Does a page exist that is meant to be the answer? An uncovered intent is a content decision, not a performance problem, and no amount of optimisation fixes it.

• IS IT INDEXED

  Is the covering page actually in the index? A page that is not indexed cannot rank, be summarised or be cited, and this is the most commonly skipped check.

• DOES IT SURFACE

  Does it appear for the phrasings it was written for — not for one chosen keyword, but for the question as people ask it?

• IS IT ACCURATE

  When the business is described by a system rather than by its own page, is the description right? A confident wrong answer is a worse outcome than invisibility.

Search Console reports clicks, impressions, click-through rate and average position across Google Search, News and Discover. That is a genuine measurement of one surface, and for that surface it is the primary source rather than an estimate.

Its limits matter as much as its contents. The documentation notes that refinements which run a new search inside Google are not counted as clicks or impressions, that data is consolidated onto canonical URLs, and that position is not comparable across result types. A report that presents any of those figures without the caveat is presenting a partial picture as a complete one.

The practical use is comparative rather than absolute: the same query set, the same aggregation, over time. Comparing this quarter’s average position to a competitor’s reported figure is comparing two different instruments.

There is no general reporting surface that tells a business how often it was used in an answer engine’s response, quoted in a generative summary or described to somebody who never visited. Tools that report a figure for this are sampling prompts they chose, which is a survey of their own inputs rather than a measurement of the business.

That does not make the surface irrelevant, and it does not make it unobservable. It makes observation manual and qualitative: ask the questions a prospect would ask, in the systems prospects use, and record what comes back — whether the business appears, whether the description is accurate, and where the wrong answer came from.

Reported honestly, that is a finding rather than a metric: a short list of questions, what each currently returns, and what changed since last time. It is less satisfying than a percentage and it is the only version that is true. Google’s guidance is also explicit that meeting every eligibility requirement does not mean a page will be served at all.

Organic work produces signals on very different timescales, and treating them as one dashboard is how programmes get cancelled two months before they would have worked.

1. 01

   Technical state

   Owner

   Whoever owns the site

   Advances when

   Changes within days — indexation, crawl errors, canonical correctness

   Ends when

   Stops being informative once it is clean

2. 02

   Coverage

   Owner

   The content owner

   Advances when

   Changes when something is published or retired

   Ends when

   Complete when every decided intent has a page

3. 03

   Retrieval

   Owner

   Nobody — it is an outcome

   Advances when

   Moves over months, unevenly, and not in proportion to effort

   Ends when

   Never settles; competitors and systems keep changing

4. 04

   Commercial effect

   Owner

   The business

   Advances when

   Visible only once retrieval is consistent enough to produce volume

   Ends when

   The only one that decides whether the programme continues

A report assembled from whatever the tool exports will describe whatever the tool measures. A report assembled from the intent map describes whether the thing the business decided to do is happening.

1. 01

   Start from the intent list

   Decides

   Which questions did we decide to own?

   Output

   A fixed row set that does not change with the tool.

2. 02

   Mark covered and indexed

   Decides

   Does a page exist, and is it in the index?

   Output

   Two yes-or-no columns that need no interpretation.

3. 03

   Record surfacing against real phrasings

   Decides

   Does it appear for the question as people ask it?

   Output

   A consistent query set, recorded the same way each period.

4. 04

   Check how the business is described

   Decides

   When a system answers for us, is it right?

   Output

   A short qualitative record, labelled as observation rather than measurement.

5. 05

   State what changed and what caused it

   Decides

   Which movement can we actually attribute?

   Output

   Attribution where it exists and an explicit “unknown” where it does not.

6. 06

   Name the next interval’s decision

   Decides

   What will this report cause us to do?

   Output

   A decision to extend, revise or stop — the only reason to have produced it.

The uncomfortable use of this data is deciding that an intent is not worth continuing to pursue — the page is good, it is indexed, it has been given time, and the positions above it are held by sources the business is not going to displace. That is a legitimate outcome and it should be reportable.

Building that possibility into the report is what makes the rest of it credible. A report that has never recommended stopping anything is not measuring; it is justifying.

It also protects the work that is succeeding. Capacity spent on an intent that will not move is capacity not spent on the one that is already responding.

1. Google documents that Search Console position values vary by result type and “can mean different things in different situations” — trends, not absolute comparisons.
2. Measure per intent: is it covered, is it indexed, does it surface for real phrasings, and is the business described accurately.
3. Search Console data is consolidated onto canonical URLs and excludes refinements that run a new search inside Google; a report without those caveats is partial.
4. There is no reporting surface for AI-answer visibility. Tools reporting a figure are sampling prompts they chose — observe it manually and label it as observation.
5. Technical state, coverage, retrieval and commercial effect move at four different speeds; reading one at another’s cadence is how programmes get cancelled early.
6. Build the report from the intent list rather than from the tool export, and end each row with the decision it supports.
7. A framework that can never recommend stopping an intent is not measurement; it is justification.