What we measure, and what we refuse to claim
Every check behind our findings, why we don't lead with a single visibility score, and where measurement ends and judgement begins.
Every finding in Visibility Copilot comes from a specific pass/fail check with a specific ID, and every task links back to the finding that produced it. When one root cause blocks other checks, for example no sitemap means we can’t test sitemap URLs, those collapse into one task instead of five.
Here’s the whole list, and the honest caveats.
Crawl access
The hard gate. We read robots.txt and check the rules for search crawlers (Googlebot, Bingbot, OAI-SearchBot) and AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot, Applebot-Extended).
Then we make a live request to your homepage with an AI-crawler user agent. If robots.txt says Allow but that request gets a 401 or 403, you have a firewall or bot-protection rule nobody remembers adding. Fix these first, because nothing else matters to a crawler that gets a 403.
Indexability
HTTPS, http:// to https:// and www/apex redirects, redirect chains, canonical tags, and noindex in meta robots or X-Robots-Tag. Classic technical SEO that still decides whether a page can be cited.
Discoverability
Does sitemap.xml resolve and parse? Is there an llms.txt, and does it actually say who you serve and what you sell? Do a sample of the URLs listed in each actually return 200?
Broken maps mean crawlers never reach the pages you actually sell from.
Readable HTML and on-page basics
Whether your homepage and top sitemap pages carry real text in the raw HTML, not an empty JavaScript shell. Then titles, meta descriptions, a single H1, image alt text, structured data, Open Graph tags, security headers, and a sample of internal links.
We sample about ten URLs from your sitemap and a handful of homepage links. This isn’t a replacement for a full technical crawler. It’s enough to catch the obvious blockers that also affect classic search.
Prompt results
For each of your tracked questions (up to 10 per domain), we record whether each engine named you, which rival it named instead, and the sources behind the answer. A lost question becomes a suggested action, such as an answer-first page, a comparison page, or an outreach pitch, with a draft you edit before using.
This is the part that involves judgement, so we keep it to evidence: here’s the question, here’s who was named, here are the sources. What to write is your call.
Why we don’t lead with one score
We could compress all of the above into a single number and put it in the hero. Plenty of tools do.
The problem is that nobody outside the labs knows how an LLM weights its sources, which means any composite index is a weighting we chose. Presenting that as measurement would be dishonest, and we don’t have longitudinal data across enough sites yet to claim our weights predict anything.
So the app leads with countable things instead: how many prompts name you, how many checks pass, how many tasks are open. Those move for reasons you can inspect.
The weekly loop
Scans run weekly, not daily, because AI answers fluctuate enough that a daily line would mostly display noise. Each run compares against the last one, closes tasks that now pass, and opens new ones for anything that broke.
Visibility Copilot is in private beta. See pricing and request an invite to run your first report.