What this run actually covers
This is the first run, and it is partial. Two categories, and not the four engines the method calls for. Rather than quietly present it as complete, here is exactly what happened.
| Engine | Result |
|---|---|
| ChatGPT | Answered. AI writing and AI voice. |
| Gemini | Answered. AI writing. |
| Perplexity | Could not run. Refuses the query without an account, both signed out and in a clean browser. |
| Copilot | Could not run. Not reachable in this run. |
Why there are no percentages worth trusting yet
Share of recommendation only means something across several engines and several months. With two engines on one category and one engine on another, the figures above are counts dressed up as percentages, so read them as positions rather than as a market share.
What is already worth noting: Claude was named first by both engines on the writing question, and Jasper appeared in both. On voice, ElevenLabs came first. Whether that holds is exactly what a second and third month will tell us.
One methodological finding, published because it matters
The first ChatGPT run was done in a signed-in account and the answer came back personalised. It referenced the kind of content the account holder produces and offered a setup tailored to their business. That is not a measurement of what buyers are told, so it was discarded for the purposes of ranking and the question was re-asked in a way that requested a general answer.
This is the single biggest hazard in this kind of exercise, and anyone publishing AI visibility numbers without saying which account they were logged into is not measuring what they claim to be measuring.
What changes next month
Clean sessions for every engine, all categories covered, and saved answer links against every row. Until those exist, this page is a starting point rather than a finding.