What we measure, and the two readings
We track our own site with our own product, on the same schedule a customer's site gets. Promvia asks 24 buyer questions that do not name us, plus 4 that do, on the five default AI assistants. The live scoreboard publishes the result, losses included, and counts only the 24: when a question names us, an answer naming us is an echo, not a recommendation. Checks run every 3 days, with a full check on every assistant each week.
| Reading | Questions naming us | 95% interval | Where it was published |
|---|---|---|---|
| 7 November 2026 | 1 of 24 | about 1–20% | Why isn't my brand in ChatGPT? |
| 12 November 2026 | 4 of 24 | 7–36% | Live scoreboard |
Which four questions, and which AI assistant
The scoreboard shows every question against every AI assistant. On 12 November one assistant named us on each of four questions. "Linked to our site": the answer linked our page as a source. "Named us": it named us without a link.
| Tracked question | AI assistant | How | Our guide on it, published |
|---|---|---|---|
| why my products are getting shown on chatgpt | Gemini | Linked to our site | Products in ChatGPT answers, 8 October 2026 |
| AI visibility tool with a free plan | Claude | Linked to our site | AI visibility tools with a free plan, 7 October 2026 |
| AI visibility tool with an MCP server for Claude | Gemini | Linked to our site | AI visibility MCP servers for Claude, 20 October 2026 |
| affordable AI visibility tool for small businesses | ChatGPT | Named us | Affordable AI visibility tools for small businesses, 25 October 2026 |
All four are narrow questions about a specific need. None of the "best tools" or "alternatives" questions names us. Across all 24, the answers cite reddit.com for 21 questions and youtube.com for 19, per the same scoreboard. That matches what our seven-check guide says to expect from a brand a few months old.
What we did, with dates
- August 2026: the site went live, with robots.txt open to AI crawlers, an llms.txt and structured data from the first public deploy.
- 24 August: the live scoreboard started publishing our own results.
- 3 September: the first AI Source Index page, weekly since; the first-eight-weeks report followed on 23 October.
- 7, 8, 20 and 25 October: the four guides in the table above.
- 2 November: Promvia listed in ChatGPT's plugin directory (changelog).
- 7 November: the WordPress plugin listed on wordpress.org, and the guide that published the first reading.
- 8 to 12 November: no guide published or revised. Product work only: a monthly report for every site (11 November) and brand names in reports (12 November).
- 11 November: the ChatGPT lane moved to a newer OpenAI model, because OpenAI retires the previous one on 11 December, and gained a cap on web searches per answer. Reports mark the change on that date.
What this does not show
It does not show that the guides worked. The guides went up four to five weeks before the first reading, so if they mattered, nothing in these two readings says when. Nothing we published between the readings targets these questions.
It does not show a real change either. The intervals overlap; the difference, +12.5 points, has a 95% interval of about −6 to +32 points, which includes zero. The same question asked twice can get a different answer, and one of the four is a ChatGPT answer taken the day after that lane changed model. And this is one site: our own.
What it does show is the shape the guide on how many prompts to track describes: on 24 questions, one answer moves the count by four points. A claim we would stand behind needs the reading to hold across several checks, and the scoreboard keeps publishing every one of them.
Read it yourself
The live scoreboard shows every tracked question against every AI assistant, the sources cited instead of us, and when it was last measured. This post is dated and stays as written, whatever a later number says; if the next readings differ, the scoreboard shows it first.
Frequently asked questions
Did publishing guides get Promvia into AI answers?
We cannot tell yet. Each of the four questions naming us on 12 November is one an October guide of ours answers, but the guides were published weeks before either reading, the two readings' intervals overlap, and the ChatGPT lane changed model in between. We report the correspondence and make no claim about the cause.
Why count only questions that do not name Promvia?
Because an answer that names us after a question that named us is an echo. Our scoreboard tracks 4 branded questions and shows them, but leaves them out of the count: 4 of 24 is over the questions a buyer asks without knowing us.
Can I check these numbers?
Yes. The live scoreboard at promvia.app/live shows each tracked question against each AI assistant and the date of the last measurement; the 7 November reading is in our guide published that day.