AG AgentReady Research
State of AI Customer Readiness
A live benchmark of how prepared scanned business websites are to be discovered, understood, trusted and acted on by AI-driven customers and agents.
This page reports the broader AG AgentReady research benchmark. Its four Website Readiness scores are aggregate research measures and are distinct from page-specific runtime WebMCP verification in current individual reports.
As of 16 October 2026 · 103 distinct websites
Scans included: 31 August 2026 to 16 October 2026 · methodology v1.0 · checks v1.0
Average scores
Mean and median across the corpus. Every score runs 0–100.
Overall readiness
49.3
median 54
Agent readiness
52.2
median 60
Revenue readiness
46.7
median 46
Trust & clarity
50.7
median 55
Overall score distribution
How the corpus splits across the four score bands.
| Band | Share of corpus | Count |
|---|---|---|
| Excellent (90–100) | 0% | 0 of 103 |
| Good (70–89) | 14% | 14 of 103 |
| Fair (50–69) | 49% | 50 of 103 |
| Needs Improvement (0–49) | 38% | 39 of 103 |
Category comparison
The three pillars side by side, using the mean of each.
| Pillar | Mean score | Count |
|---|---|---|
| Revenue Readiness | 47% | n = 103 |
| Agent Readiness | 52% | n = 103 |
| Trust & Clarity | 51% | n = 103 |
Most common problem categories
Share of websites with at least one unresolved finding in each category.
| Category | Websites affected | Count |
|---|---|---|
| Trust | 32% | 33 of 103 |
| Agent Readiness | 31% | 32 of 103 |
| Conversion | 31% | 32 of 103 |
| Structured Data | 30% | 31 of 103 |
| Revenue | 29% | 30 of 103 |
| Clarity | 19% | 20 of 103 |
| WebMCP | 19% | 20 of 103 |
| Technical | 1% | 1 of 103 |
Most commonly failed checks
Failure rate per stable check. Only checks observed on at least 5 websites are listed, and the denominator is shown for each.
| Check | Failure rate | Count |
|---|---|---|
| Direct contact methods are reachable | 97% | 31 of 32 |
| Contact details are machine-readable | 97% | 31 of 32 |
| Main action is visible early on the page | 97% | 31 of 32 |
| Forms are labelled and parseable | 97% | 31 of 32 |
| Actions are unambiguous to an agent | 94% | 29 of 31 |
| Service structured data present | 91% | 29 of 32 |
| Reviews or ratings are referenced | 90% | 28 of 31 |
| Complete contact details are published | 88% | 28 of 32 |
| LocalBusiness structured data present | 88% | 28 of 32 |
| A clear primary call to action exists | 87% | 27 of 31 |
| A quote can be requested | 78% | 25 of 32 |
| The next step is spelled out | 74% | 23 of 31 |
What the data shows so far
Within this corpus, the mean Overall Website Readiness score is 49.3 and the median is 54. Of the three pillars, the lowest mean is Revenue Readiness at 46.7, which is where the scanned sites lose the most ground.
The most widespread problem category is Trust, present on 33 of 103 websites (32%). These are observations about the websites in this corpus only. We deliberately avoid extrapolating to any wider population, and we do not describe movement over time until the corpus is large and stable enough to support it.
Methodology summary
Each website is scanned for up to five publicly accessible pages. Deterministic checks in code produce Revenue Readiness, Agent Readiness and Trust & Clarity, weighted 45/30/25 into Overall Website Readiness. Scores are never generated by a language model.
The corpus includes completed scans with all four scores present, using only the most recent completed scan per website. Domains are lowercased with a leading “www.” removed, and internal or test domains are excluded. Full detail is in the methodology.
Limitations
- This is a self-selected corpus of websites whose owners or users chose to scan them. It should be read as a directional benchmark, not a census of the entire web.
- Figures are recalculated live, so they change as new websites are scanned. Cite the “as of” date above with any figure.
- Only publicly reachable pages are analysed. Sites that block automated access or render entirely in the browser may score lower than a human reviewer would judge them.
- Checks measure observable signals in published pages, not business outcomes. A high score is not a guarantee of traffic, ranking or revenue.
- Per-check failure rates only cover findings carrying a stable check identity.
Assess your own website
Run a free Agent Readiness scan to inspect your website’s current agent-facing capabilities and evidence. Qualifying scans also contribute anonymised measurements to the broader research benchmark.
Run Free Website Scan