Artifact Machine

Search engines found you. Answer engines can’t read you.

We measure whether a machine sent to answer a question about a company can actually read its website — and whether that company has been paying for search visibility it no longer gets. One page, two seconds, no account.

Only 1 in 4

of the companies we measured had the problem worth acting on. Almost everyone scores badly on answer-engine readiness — that part is the weather. The finding is the mismatch: businesses that clearly paid for search and get nothing back from machines.

57 companies · homepage each · full method and data published · read the study

Check one site

Paste an address. You get search readiness, answer-engine readiness, and the distance between them, measured the same way as every row in the study.

If we can’t read the site — bot protection, a dead server — you get no scores at all and a reason. A zero would be a measurement, and nothing was measured.

Check a site

Score a whole list

One HTTP call, one row per domain. It drops into Clay, n8n, Make, Zapier or a spreadsheet without anything to install or approve.

POST /api/batch
{"domains": ["example.com"]}

Read the API docs

Things you can buy

Both run on Apify, per result, with no subscription. A run that returns nothing costs nothing. Neither line has meaningful usage yet, and this paragraph will say so until that stops being true.

Enrichment columns — $0.001 to $0.003 a row

Built for a list. You give them ten thousand domains and get one column back, priced so that scoring a whole list is a rounding error rather than a decision. Rows we could not read come back empty with a reason, never as a zero.

Single-purpose audits — $0.01 to $0.02 a result

Built for one site at a time, and the older of the two lines. Each answers one question about one page in more depth than a column can carry.

What we won’t tell you

Whether an AI assistant mentions you today. That needs repeated sampling of the models themselves, several companies sell it, and this is not it. This is the cheaper question underneath: if a model does come looking, is there anything here it can use?

We also won’t claim the gap is a signal nobody else has. Across 3,000 randomly sampled domains it correlates 0.84 with search readiness alone — a genuinely different ranking, about half the top 200 differs, but not an independent one. That number is published in the study rather than left out of it.

One person, measuring in public.