Share of answer vs keyword rank: two ways of counting.
Keyword rank and share of answer measure visibility on two different surfaces, and are produced in two different ways. Keyword rank is a position in a results list, reported directly by search indexes and stable enough that a single check is a measurement. Share of answer is how often a brand is named across a fixed set of questions put to AI assistants, and because no engine publishes that, it has to be sampled by asking the questions repeatedly. Rank is read; share of answer is estimated, and the estimate is only as trustworthy as the prompt set and the sampling behind it.
01Side by side
Where Share of answer and Keyword rank part company.
| Dimension | Share of answer | Keyword rank |
|---|---|---|
| What it counts | How often a brand is named in generated answers across a defined prompt set. | Where a page sits in the ranked list for a given keyword. |
| How the number is produced | Sampling: the same questions are asked on a schedule and the answers recorded. | Reporting: search indexes publish position data directly. |
| Reproducibility | Approximate. Model answers vary between runs, so one check is a sample rather than a result. | High. The same query returns close to the same position, allowing for personalisation and location. |
| What defines the scope | The prompt set, which is written by whoever is measuring and can be chosen to flatter. | The keyword list, which is checkable against published volume data. |
| Granularity of a win | Binary per answer, then weighted by position and whether a citation was attached. | Ordinal: position one to one hundred, with click-through modelled from position. |
| Relationship to traffic | Weak. An answer can resolve the question entirely, so presence and sessions move independently. | Direct and well modelled, at least on queries that still produce clicks. |
| Main way it gets abused | Reporting a percentage without publishing the prompts that produced it, or screenshotting one good answer. | Reporting rank on terms nobody searches, where position one is worth nothing. |
02The overlap
Both are proxies, and both are gamed the same way
Neither metric is revenue. Rank stood in for demand for twenty years because it correlated with clicks, and share of answer stands in for consideration because it correlates with being shortlisted. Both correlations are real and neither is tight, which is why either one reported alone can be moved without anything improving.
The failure mode is identical too: choose the questions carefully enough and any brand looks dominant. A keyword list of branded terms produces beautiful rank reports, and a prompt set built around a company's own product category produces beautiful share-of-answer reports. In both cases the fix is publishing the list, so the measure can be argued with.
Read together they cover each other's blind spots. Rank shows what is happening where clicks still exist; share of answer shows what is happening where they no longer do. A brand losing rank and gaining share of answer is in a very different position from one losing both, and only one of those two numbers can tell them apart.
03Which you need
The answer depends on what is broken.
01
Use rank where the query still produces a click
Transactional and navigational searches largely still resolve to a link. On those, rank remains the sharper instrument and is far cheaper to track, so replacing it with sampling would be a downgrade dressed as modernisation.
02
Use share of answer where research has moved
Comparison and recommendation questions are the ones that migrated to assistants first. On those, rank can hold perfectly steady while a brand stops being considered, and nothing in the rank report shows it.
03
Insist on the prompt set before you accept a number
A share-of-answer figure without its questions cannot be reproduced or challenged, which makes it a claim rather than a measurement. The questions should also be recognisable as things a buyer would type, not phrased around the brand.
04Questions
Share of answer and Keyword rank, asked plainly.
- Why does share of answer change between checks?
- Model responses are not deterministic, and engines change retrieval and model versions without notice. A single check is one sample from a distribution, so the signal lives in the trend across repeated runs rather than in any one result.
- How many prompts make a usable set?
- Enough to cover the questions that actually precede a purchase in your category, which in practice is dozens rather than hundreds or a handful. What matters more than the count is that the questions are the ones buyers ask and that the list is published with the results.
- Can I get share of answer from Search Console?
- No. Search Console reports Google Search, and now surfaces some AI Overviews data, but it has no visibility into ChatGPT, Claude or Perplexity. Those have to be sampled directly.
- Should rank tracking be dropped?
- No. Search still carries an enormous share of commercial intent, and rank remains the cheapest reliable signal on the surfaces where clicks happen. Dropping it trades a good instrument for a noisier one.
- Does being cited matter more than being named?
- Both are worth counting separately. Being named puts the brand in the consideration set; being cited attaches a link a reader can follow, and only the second produces a visit. A report collapsing them into one figure is hiding the more useful half.
05Definitions
Stop comparing, start checking
Measure what analytics can't see.
Prompt-level reporting: where you're named, where a competitor is, and which sources built the answer.