We pointed our third Actor at our own site. It found three things.

The third Actor asks a question the first two didn't: not "what did a company file?" but "can the AI engines see you, and what would you fix first?" This week it ran against our own site. Here is what it does, what it found, and how the private test went.

What it checks

Three tiers, in the order to act on them.

Eligibility is the technical layer, and it is request-only — no headless browser. It reads robots.txt once per AI crawler and tells the search bots (OAI-SearchBot, PerplexityBot, Claude-SearchBot — the ones that fetch pages to answer questions) apart from the training bots (GPTBot, ClaudeBot, Google-Extended and the rest). Blocking a training bot is a policy choice; blocking a search bot removes you from answers. It then fetches your pages as a browser and as each search bot and compares what came back — a 403, a challenge page, a body a fifth the size. It reads the raw HTML the way a crawler does: word count, noindex, canonicals, JSON-LD, sameAs links, the sitemap. A page whose content only exists after JavaScript runs is reported as such; the crawlers don't run it.

Observed inclusion asks the engines your buyers' questions. Perplexity and OpenAI, on your own keys, prompt by prompt: was the brand named, was the domain cited, which competitors were named, and which sites the engine trusted instead. An engine you didn't give a key for is absent from the report, never zero.

The plan is a fixed rule table over the findings and the metrics — no model writes it. Every recommendation carries the ids of the findings it rests on, and the Actor refuses to write a plan where one doesn't. It ranks by tier, then impact, then effort.

What it found on buildonmarketplaces.com

The report is on the product page, unedited. The short version:

• All three search bots fetch us exactly as a browser does. Nothing blocked, no noindex.
• Three warnings, three fixes: no Organization JSON-LD, no sameAs links to our profiles, no sitemap.
• Fifteen prompts on two engines, thirty answers: neither engine named BuildOn Marketplaces once for "Apify actors for GTM teams", "SEC filing event data" or "AI search visibility audit". Perplexity's sources were apify.com and sec.gov; OpenAI's were filedfeed.com, gtmcli.com, scrutar.com and veniara.com. (The report on the product page has since been re-run on the Standard tier — the bigger models — with the same result, 0 of 30; the sources shifted to apify.com, sec-api.io and docs.intrinio.com.)

That last line is the honest baseline for a two-week-old site, and it is exactly what the tool is for: the plan's inclusion tier says which prompts to publish pages for, and the opportunity tier says where the engines are already looking.

How the private test went

Twenty private-test runs on our two domains, all succeeded, every plan grounded, no key in any output. The price is on the product page: US$4.95 per report on the buyer's own keys, twenty questions included, US$0.10 a question beyond.

Revenue and contribution are shown to members. The method →

One thing broke on the way: Perplexity retired its chat-completions endpoint for new keys the day we tested ("Sonar is now the Agent API"). The adapter moved the same afternoon, and the fixture now records the new shape.

What happens next

The private test is done and the ledger says PRIVATE_TEST. The listing gate, the Store prices and the public listing follow on the portfolio's schedule — one measurement interval after the SEC pair — and the Monday numbers will show up here as they do for the others. In the meantime we have three fixes to make on our own site, and the next post will say whether they moved anything.