Search stopped being ten blue links. Buyers ask an assistant, read one paragraph, and shortlist from it. If the engines can't crawl your site or can't tell what your company is, you are not in that paragraph. I check whether the door is open, then I check what the engines actually say when it is.
Two numbers from other people's data, both from 2026. I am quoting them because they are the clearest public evidence I have found, not because they are mine.
Across more than a trillion visits to US retail sites, visitors arriving from AI assistants converted 42% better than the rest. A year earlier the same cohort converted 38% worse. That reversal happened inside twelve months.
Most AI referrals arrive with no referrer, so analytics files them under Direct. In that study the hidden slice converted at 10.21% against 2.46% for everything else. Companies are reading a report that does not contain their best channel.
I will not turn either of those into a forecast for your business. They are industry context. What I can tell you is whether the engines can currently read you at all, and that part is measurable today.
Four public files, read straight off your server, scored out of 100. It is deterministic. There is no language model anywhere in it, which is why every line on the page names the artifact it came from and you can go verify it yourself in about a minute.
Whether you are blocking the AI crawlers. Plenty of sites block GPTBot or Google-Extended by accident, usually inherited from a template or a plugin default, and nobody notices because nothing breaks.
The plain-text file that tells an assistant what your company is and where the real pages live. Newer convention, cheap to add, and most B2B sites do not have one.
Structured markup that states, in machine terms, what you sell and who you are. Without it an engine has to infer your category from prose, and it frequently infers wrong.
Whether there is a clean index of your pages to crawl at all, and whether it is actually reachable rather than pointing at a stale path.
You get a page like this one, which is my own site scored by the same code. I fix my own findings before I send anyone else theirs.
The first real run of this check was against my own product, AcquisitionIQ. It came back with no llms.txt, no schema, and a homepage Google had never indexed.
Worse, when I asked the engines whether the company was legitimate, the answer was that they could not verify it. That is what an invisible entity looks like from the buyer's side. I shipped the fixes the same day, verified them in Search Console, and only then started offering the check to anyone else.
The structural check tells you the door is open. It does not tell you what happens when a buyer walks through it and asks a real question. That is a different measurement and it is the one worth money.
Written against how someone actually shops for your category, not keyword phrases. Each one is run five times, because these systems are non-deterministic and a single run tells you almost nothing.
Claude, ChatGPT, Perplexity, and Google AI Overviews. You get the raw answers with dates attached, so the finding is auditable and you can re-run it later against the same baseline.
How often you appear, how often they do, and what the engines say you are when they get it wrong. Plus the fixes, ordered by what actually moves the answer.
Turnaround is about a week. Re-run it monthly if you want a trendline, though I would wait until you have shipped the first round of fixes before paying for a second look.
Every vendor in this category should publish this section and almost none of them do.
There is no submission form and no ranking API for any of these engines. Anyone promising you a position in AI answers is selling something they do not control. I can remove the blockers and improve how clearly you are described. I cannot buy you a slot.
I will not multiply your traffic gap by a conversion rate and call it lost revenue. The industry numbers on this page are attributed to the people who measured them, and they are not a forecast for you.
Four files. It is a structural read, not an answer-quality read. A perfect 100 means nothing is blocking the crawlers, which is necessary and not sufficient.
These models change under you. A finding is dated for that reason, and a single audit is a snapshot rather than a permanent state of the world.
Scope and rate go in writing before any work starts.
The four-file read, scored, on a page you can share internally. Same day. No call required and no obligation attached to it.
The prompt set across four engines, competitor comparison, verbatim dated answers, and a prioritised fix list. About a week. Price moves with how many competitors and prompts the scope needs.
If you want the fixes implemented rather than documented, that is a separate conversation and it lives over on Build With Ron.
One line with your URL is enough. I run the check, send you the page, and if the first fix is a one-liner I will just tell you what it is rather than making you book anything. If you want to walk through it, that is ten minutes and I will have the fix drafted before we start.