Agent-Ready Scorecard

What actually gets you recommended.

When someone asks an assistant for a good local spa or auto shop, the answer is driven mostly by your off-site reputation, not your website markup. We lead with the real levers that win the recommendation. The on-site files that make you machine-readable are necessary plumbing, so we keep that part as a scoped sub-point and say so plainly.

Off-site levers first No self-graded vanity number No faked checkmarks
The real test

What an AI actually checks before recommending you

For a local business, the recommendation is driven mostly by off-site signals an AI can corroborate from sources it already trusts. These are the levers that move it, in plain terms. They are the work that wins, so they go first on this page.

  • Google Business Profile completeness

    Hours, categories, services, photos and Q&A filled in. AI assistants lean heavily on map and profile data when naming a local business.

  • Recent, verifiable reviews

    Enough recent reviews to look active and trusted. A stale or thin review history reads as risk to a model deciding who to recommend.

  • Name, address and phone consistency

    The same details across your site, your profile and the major directories. Conflicting listings make an agent unsure which record is real.

  • Credible third-party mentions

    Being named on sites the model already trusts. Outside corroboration is what turns a clean profile into a confident recommendation.

We do not measure these live, and we will not fake a green checkmark on any of them. This is a checklist of what an AI weighs, not a score we have run on you. The free Ops Scan is where we look at your own off-site gaps and hand back a plain result.

Scan your site: see what an AI checks before recommending you →
Why we frame it this way (the research behind it)

Two findings shaped this page. First, a 2026 analysis of roughly 500 million AI-bot requests by SE Ranking found that a /llms.txt file was fetched only a handful of times across that whole sample, and Google has publicly said it has no plans to support llms.txt for its AI features. So the file is worth publishing for clean machine reads, but it is not a citation lever. Second, the established local-search guidance (Princeton GEO research on what generative engines surface, plus mainstream local-AEO writing) points at off-site reputation: profile completeness, review signals, listing consistency and trusted mentions. We split the page along that real line, and we put the off-site levers first because that is the part that wins the recommendation.

What "agent-ready" actually means

Think of it as a second front door. The first one is for people. The second is for the AI agents that read the web on a person's behalf and try to act on what they find. If that door is locked, an agent has to guess at who you are and what you sell. Being agent-ready is a small set of published files that hold that door open and make you machine-actionable. It is table stakes, not a magic trick. It earns a clean read, not an automatic recommendation.

A Markdown facts page

An llms.txt file an assistant can read in one fetch, so it quotes you correctly instead of guessing.

Structured data

Schema markup that labels your services, hours and FAQ in a format machines parse without ambiguity.

A clean FAQ

Plain answers to the questions buyers ask, which is exactly what an answer engine wants to cite.

A sensible robots file

One that welcomes the helpful crawlers rather than blocking the very tools deciding who to recommend.

Discovery files (/.well-known/)

Standard identity and capability files agents look for first, so they know who you are and what you offer.

What it does and does not do

A clean read removes guesswork so an agent can act on your site. It does not, on its own, make an AI recommend you. That is decided off-site, by the levers at the top.

A scoped sub-point, not a headline

On-site machine-readability layer: complete.

This is the necessary plumbing, not the recommendation. We hold ourselves to an open, published checklist for the on-site machine-readability job only, where every item points at a real file you can open right now. That layer is complete because each file is live and correct. We built a real client site, a wellness business (HRC), to the same checklist, so this is something we practice, not something we only preach. It says nothing about whether an AI will recommend you. That is the off-site work above.

Every link above opens a live artifact on this domain. A real score is one you can check, not one you have to trust. Want to see the engine behind it? Our build pipeline fails if a page lies. Read the honesty firewall for the gates that run on every deploy.

Open the full machine-readability checklist (the supplementary detail)

The checklist below is the long-form version of the six machine-readability checks scored above. It is reference material, kept collapsed so the page stays calm to read. This layer is scoped on purpose: it is plumbing, not the thing that wins the recommendation.

  1. llms.txt present and current. A Markdown facts file at the site root, summarizing who you are, what you sell, pricing and proof status, refreshed when the facts change.
  2. JSON-LD structured data. WebPage and FAQPage schema (plus Organization and Service where relevant) emitted on every key page, with the visible FAQ and the schema FAQ kept identical.
  3. Human FAQ that matches the schema. The questions a real buyer asks, answered plainly, so the answer engine has trustworthy text to quote.
  4. robots.txt that allows answer-engine crawlers. Helpful agents are permitted; only genuinely abusive bots are limited. A robots file should not lock out the assistants you want recommending you.
  5. /.well-known/ discovery files. An agent.json identity file, an agent-card, a security.txt and a published API catalog, so an agent can discover your capabilities the standard way.
  6. A machine-readable API or service catalog. An openapi.json (or equivalent) listing the endpoints an agent can call, so automated callers find the right path without scraping.
  7. Clean, crawlable HTML. Content rendered as real text and headings, not trapped inside scripts an agent cannot read.
  8. Honest, verifiable claims. Numbers an assistant can trace to a source. Faked metrics get a business flagged, not cited.
The part we do not fake

The payment and agent-commerce layer is still pre-standard.

There is a real, exciting frontier here: AI agents that pay, authenticate and transact on your behalf. Today that whole cluster, the payment rails, the agent authentication, the agent-to-agent commerce handshake, has no settled standard across the industry. There is no agreed spec to pass yet.

So we do not put a green checkmark on it, and no honest scanner should either. A tool that claims your site already passes AI-commerce interop is selling you a checkmark for a test that has not been written. We leave that row open and say why. That restraint is the point. A score is only worth anything if it tells the truth about what it cannot measure.

Proof status, stated plainly: this scorecard reports $0 recovered and a Sample posture on anything we have not measured. We do not invent results to fill a row.

Find your own off-site gaps in a few minutes.

Version 1 is a showcase of the checks we hold ourselves to. A live URL-scanning version is a documented next step. For now, the free Ops Scan reads your operations and visibility, then hands back a plain result with the one fix to make first. No signup, no card for the scan. If you go further, entry is a refundable $500 First Recovery Setup, then flat $199/mo only if you keep it.

Run the free Ops Scan →

Prefer to see how the gates work? Read the honesty firewall.

View Markdown facts