What replaces the ranked list
In classic search there are ten positions and you occupy one. In a conversational mode there is an answer, assembled from some number of sources, and you are either among them or you are not. Nothing orders you against the others in a way you can observe or move.
| Classic search | Conversational mode | What to do instead |
|---|---|---|
| Position 1 to 10 | Included or not | Optimise for inclusion, not order |
| Rank tracking | No position exists | Track fetches and answer rate |
| CTR by position | No positions to bucket | Track referred sessions |
| Keyword targeting | Follow-up questions, not keywords | Answer whole questions |
This is why a dashboard showing your average position on this surface is showing you an invented number. There is nothing wrong with wanting the metric; there is simply no measurement behind it.
What decides inclusion
The same four things that decide it everywhere else, which is convenient: the work does not fragment per surface even though the vocabulary does.
- The fetch succeeds. No bot rule or CDN filter standing between the request and your content.
- The response parses without a browser. Client-rendered facts are absent from what arrives.
- The facts are cheap to reach. A page that is mostly theme spends the budget before reaching your content.
- The facts agree with themselves. Prose and structured data saying different things produces a confident wrong answer.
What to measure instead of rank
- Agent fetches by platform, from your own request log. Exact, because it is your server.
- Answer rate, the share of those requests that produced something useful.
- The miss log, which names the next thing to build in the order the market asked for it.
- Referred sessions, for the people who click through after the answer.
- A deterministic readability score, so a change you make is a diff between two numbers rather than a feeling.
None of that requires a new tool category. It requires reading your own logs, which is the one data source about this surface that is complete rather than sampled.