How we score agent visibility
An Agentfacing audit measures what an AI agent can see, say and do on a business website, the way agents actually read it: as text and markup, with no animations, no images and no phone calls. Every site is scored across six dimensions, ten points each, for a total out of 60.
Two rules every finding follows
The evidence rule. Every finding quotes a string an agent actually read on the site. No inference, no "likely". If we did not see it, we say we did not see it.
The absence rule. Text extraction alone cannot prove something is missing from the underlying code. Where a finding depends on absence (for example, no structured data), it is marked provisional until confirmed by raw-HTML inspection, which every paid audit includes.
The six dimensions
| Dimension | The question | How points are earned |
|---|---|---|
| 1. Discovery | Can an agent tell who the business is, what it does, where it operates, and whether it is licensed? | Clear identity, services, area and licence or registration numbers in crawlable text. Heavy deductions for contradictory addresses or another company's identity leaking into the copy. |
| 2. Price | Can an agent answer "what does this cost?" | Named dollar ranges in page text score 6 or more. Promotional teasers only: 2 at most. Nothing: 1. |
| 3. Transactability | Can an agent book, or at least formally request, a job? | Open online booking: 6 or more. Booking behind a login: 3 to 4. Contact form only: 2 to 3. Phone only: 1 to 2. Agents do not make phone calls. |
| 4. Availability | Can an agent find out when the business can actually do the work? | Opening hours per day in text: 3 or more. Live, queryable slots: 6 or more. Claims like "24/7" with no data: 2 at most. |
| 5. Machine trust | Is the machine-readable layer consistent and clean? | Structured data present, click-to-call links that match the displayed number, registration numbers published, and no placeholder text, template debris or dummy images. |
| 6. Action surface | What can an agent do here beyond reading? | Exposed actions, booking endpoints or APIs an agent can navigate. Almost every local business scores 1 to 3 today. |
What else we check
- Field-service software. We look for ServiceM8, Tradify, Fergus, Jobber, ServiceTitan, Housecall Pro, simPRO, AroFlo and similar, because an existing platform usually means booking can be exposed to agents quickly.
- Phone wiring. Every displayed number is compared with the number its click-to-call link actually dials.
- Franchise identity. For franchise branches, whether an agent can tell the local branch apart from the national network.
- Local registration. In New Zealand, PGDB registration and Master Plumbers membership are credited the same way as licence numbers elsewhere.
What a score is, and is not
A score describes what automated agents could read on the date of the audit. It is not a ranking guarantee, a marketing grade or legal advice, and it changes when the site or agent technology changes. That is why every fix is re-scored from the outside rather than self-graded.
See it applied
Read a complete sample report: a Phoenix-metro plumber that scored 18 out of 60, with the evidence for every point.
Our own score
Agentfacing scores itself with the same rubric: 44 out of 60 at our last self-audit in October 2026, up from 32 in July. We re-score after every material change to this site and publish the result in our footer.
Or email hello@agentfacing.co. Replies within one business day.