We are publishing the zero
Launch posts usually begin after the graph turns upward. We do not have that graph. Prefometry's new public pages went live on August 25, while the latest Bing AI Performance report ended on August 24. For that window, the citation count was zero.
Zero tells us nothing about the August 25 changes because they were not in the report yet. It does give us a clean starting point. We are recording it now, before there is a result to make us look clever.
What we measured on August 25
We opened the public site without signing in, then repeated the homepage request with six crawler user agents. This was an access check, nothing more. It shows that the pages can be fetched; it does not prove that any provider visited, indexed, or cited them.
- User agents tested: GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended, and bingbot
- Public marketing pages are allowed in robots.txt; app, API, authentication, claim, scan, invite, and unsubscribe routes are blocked
- The sitemap was submitted to Google and Bing, and the same public URL set was sent through IndexNow
| Check | Baseline | What it tells us |
|---|---|---|
| Public URLs in the sitemap | 41 | The number of public pages available for discovery |
| Crawler user-agent requests | 6 of 6 returned HTTP 200 | The homepage was reachable in each test |
| llms.txt | HTTP 200 · 3,757 bytes | A plain-text directory of the product, pricing, comparisons, tools, and guides is live |
| Bing AI Performance citations | 0 through August 24 | No Prefometry citation was reported before the new pages went live |
| Average cited pages | 0 through August 24 | No Prefometry page had become a reported Copilot source in that window |
A 200 response is not a citation
The path from published to cited has several separate steps. The page must be reachable and indexed, but it also has to answer the question better than the other sources available when the answer is generated. Passing the first step says very little about the last one.
Search Console can show that a page is indexed. It cannot tell us that ChatGPT, Perplexity, Gemini, an AI Overview, or Copilot will quote it. We are keeping those measurements separate because combining them would make an encouraging chart and a bad conclusion.
The product scan found the same zero from another angle
Later on August 25, we used Prefometry to scan Prefometry. Five commercial-intent questions were scheduled across ChatGPT, Perplexity, Gemini, Google AI Overviews, and Microsoft Copilot with two repetitions each. Forty of the 50 planned answers returned. All ten Perplexity attempts timed out, so the result remained partial instead of being presented as a complete five-engine sample.
Prefometry appeared in zero returned answers and received zero citations. The engines still produced 240 external citation occurrences across 37 answers with citations. That distinction matters: the category had plenty of available evidence, but none of it pointed back to Prefometry.
The first version of our own report also exposed quality problems: a generic navigation label had been tracked as a competitor, and some answers interpreted AI visibility as video surveillance. We excluded those patterns from future scans, added entity-boundary checks and topic relevance filtering, and now reduce both score and confidence when providers or relevant samples are missing.
| Measure | Result | Reporting rule |
|---|---|---|
| Planned provider runs | 50 | Five prompts × five providers × two repetitions |
| Returned answers | 40 | Missing samples remain missing |
| Failed runs | 10 Perplexity timeouts | Run status stays partial |
| Prefometry mentions | 0 | No inferred or synthetic mentions |
| Prefometry citations | 0 | Only native observed citations count |
What changed on launch day
We rewrote the homepage so a visitor can tell what Prefometry does without decoding product language. Pricing is public. The comparison pages name the products and show the criteria. The methodology explains samples, confidence, and rechecks. The guides stick to narrow questions instead of treating one model response as a permanent ranking.
We also added canonical metadata, social previews, structured data, an XML sitemap, and llms.txt. That is useful plumbing. It makes the site easier to read, but it does not make the site worth citing. The evidence still has to do that job.
The three questions we will keep fixed
Changing the test every week would make the updates meaningless. For the first 30 days, we will keep the wording, provider, surface, locale, and sampling plan on record so the later runs remain comparable.
- "What is the best AI visibility monitoring software for a small team?"
- "Which GEO tool is suitable for an agency managing several client brands?"
- "How does Prefometry compare with Otterly.AI?"
What we will publish next
We will check again after 7, 14, and 30 days: September 1, September 8, and September 24. Each update will report whether Prefometry was mentioned, whether one of its pages was cited, which source was chosen instead, and whether the result repeated across samples.
If the answer is still zero, we will publish zero. A clean no-change result is more useful than moving the goalposts. It tells us that technical access was not enough and that the site needs stronger evidence, more time, or authority beyond its own domain.
Limits of this baseline
The user-agent checks were our own HTTP requests, not visits confirmed by the providers. Bing AI Performance covers citation activity from Microsoft Copilots and partners and describes its figures as a sample of overall activity. Its August 24 cutoff also means the report does not measure the August 25 release.
Later movement will not prove that one page or one technical change caused an answer engine to behave differently. We can record what changed and when it changed. The causation claim stops there.