← Back to the AI-readiness check
Agoria · /assess methodology

Every check behind an AI-readiness score

The canonical rubric a website is measured against — 65 weighted checks across 13 categories, each graded by how strongly the evidence backs it. A business is only ever scored on the checks that apply to it. Alongside the score, we run live AI-visibility probes and a crawler-access audit.

65
weighted checks
13
scoring categories
3
hard gates
12
AI crawlers audited
3×5
live assistant queries

How the score is built

Readiness is the share of applicable weight a site earns — then capped if an AI crawler literally can't read or reach it.

The maths

Readiness = points earned ÷ applicable points, as a percentage. Every check carries a weight (0–5); only checks that apply to the business type and the evidence source count toward the total.

Hard gates

Three checks are gates Gate — if a bot is blocked, the content is JS-only, or the site is down, AI can't cite you at all. Each failed gate caps the ceiling: max(20, 55 − 15 × gates).

Grade bands

Excellent ≥ 90
Strong ≥ 75
Fair ≥ 55
Needs work ≥ 35
Critical < 35

Reading a finding

Every check resolves to one of three states: ✓ machine-readable — in structured data an AI reliably reads; ◐ on your page — present as text or an image but not yet in a form assistants can parse (earns no points yet, but these are the quick wins behind your achievable score); or ✕ not found — not detected on the site at all.

Head-to-head

We also score the websites of the competitors an assistant actually recommends — using this same rubric — so you see precisely who's out-reading you, and by how many points.

Proven Causal experiment or vendor-confirmed mechanism
Strong Large-N correlation studies
Moderate Multi-source practitioner evidence
Weak Plausible, effect size unverified
Site read from the live website
Profile from a stored Agoria profile

The 13 scoring categories

Weight reflects impact on AI citation — tuned to research, not classic SEO folklore. "Applies to" narrows a check to the business types it's meaningful for; non-applicable checks are excluded from the total, never counted against you.

🤖Crawler access & renderingWebsite17 pts
CheckWhat it verifiesEvidenceWtApplies to
AI search crawlers allowed Gate
robots.txt doesn't block OAI-SearchBot, PerplexityBot or Googlebot — blocked means uncitable, no matter how good the contentProven5
Site
Content readable without JavaScript Gate
Meaningful content is in the raw HTML; AI crawlers don't run JS, so JS-only content is invisibleProven5
Site
Live-fetch agents allowed
ChatGPT-User & Perplexity-User, which fetch your page live when a user clicks, aren't blockedProven2
Site
Indexable by Bing (powers ChatGPT)
Bingbot isn't blocked — ChatGPT search is ~87% dependent on Bing's indexStrong2
Site
Site reachable (200 OK) Gate
The homepage responds 200 and can actually be fetchedProven2
Site
Canonical URL set
A sane canonical link so your authority isn't split across duplicate URLsModerate1
Site
🪪Identity & clarityBoth13 pts
CheckWhat it verifiesEvidenceWtApplies to
Business name
A clear business name assistants can attach facts toStrong3
Both
Specific category
A precise category (not just 'Business') so assistants match you to the right questionsModerate2
Both
Tagline / slogan
A short positioning lineWeak1
Both
Clear description (40+ chars)
The 1–3 sentence description assistants quote about youStrong3
Both
3+ discovery keywords
Keywords that help agents match loose, natural queriesModerate2
Profile
Approved target personas
Positioning that tells assistants exactly who to recommend you toModerate2
Profile
🌐Off-site authority & entityBoth13 pts
CheckWhat it verifiesEvidenceWtApplies to
Links to authoritative profiles (sameAs)
schema sameAs to Wikipedia / Wikidata / LinkedIn — the strongest entity signal you controlStrong4
Site
Linked social profiles
Brand mentions across the web predict AI visibility ~3× better than backlinksStrong2
Both
Complete Organization entity
Organization schema with name + url + logo — the base of your knowledge-graph nodeModerate2
Site
Entity graph @id
An @id node that ties your schema into one referenceable entityWeak1
Site
Ratings / reviews
Reviews strongly influence which business an assistant recommendsStrong3
Both
Several reviews
Three or more reviews presentModerate1
Both
📞Contact & linksBoth4 pts
CheckWhat it verifiesEvidenceWtApplies to
Phone (machine-readable)
A tel: link or parseable phone numberModerate2
Both
Email
A contactable email addressWeak1
Both
Website
A website on the profileModerate1
Profile
📍Location & service areaBoth8 pts
CheckWhat it verifiesEvidenceWtApplies to
City / suburb
'Near me' queries need a city to match againstStrong3
BothLocal
Street address
A full street addressModerate1
BothLocal
Region / state
State or regionModerate1
BothLocal
Postcode
A postcodeWeak1
BothLocal
Map coordinates
Lat/long powers precise 'near me' matchingModerate1
ProfileLocal
Area served
The geographic area you serveModerate1
BothLocal
🕐Opening hoursBoth5 pts
CheckWhat it verifiesEvidenceWtApplies to
Hours published
'Open now?' is one of the most common assistant questionsStrong3
BothLocal
All 7 days specified
Hours given for every day of the weekModerate2
BothLocal
🛎️Offerings & pricingBoth24 pts
CheckWhat it verifiesEvidenceWtApplies to
Services listed
Services & prices are what agents quote and bookStrong4
BothServices
Services have descriptions
Each service explained, not just namedModerate2
BothServices
Services have prices
Prices attached to servicesModerate2
BothServices
Bookable durations set
Durations enable real appointment slotsWeak1
ProfileBookings
Product catalogue listed
A real product catalogue is presentStrong4
BothProducts
Products have prices
Prices attached to productsModerate2
BothProducts
Products have images
Product images presentWeak1
BothProducts
Plans listed
Comparable plans power the switch-&-save comparisonStrong4
BothProvider
3+ plans to compare
At least three plans to weigh upModerate2
BothProvider
Plans list their features
Each plan's features spelled outModerate2
BothProvider
Trust & mediaBoth3 pts
CheckWhat it verifiesEvidenceWtApplies to
Logo
A brand logoModerate1
Both
Photos / gallery
Real photography or a galleryWeak1
BothNon-provider
Credentials / licences
Licences or credentials that build trustModerate1
BothBookings / Provider
Answers (FAQs)Both5 pts
CheckWhat it verifiesEvidenceWtApplies to
FAQ present
At least one FAQModerate2
Both
3+ FAQs
Question→answer pairs map exactly to how assistants retrieve and quote passagesModerate3
Both
✍️Content GEO leversWebsite9 pts
CheckWhat it verifiesEvidenceWtApplies to
Statistics / data present
Adding statistics measurably lifts AI visibility (+6.6pts, GEO paper, KDD 2024)Proven2
Site
Quotations present
Quotations are the single highest-impact content lever (+8.5pts)Proven2
Site
Cites authoritative sources
Citing sources lifts AI visibility (+5.6pts)Proven2
Site
Question-style headings
Headings phrased as questions mirror how users prompt assistantsModerate1
Site
Comparison table
Real HTML tables are among the most-cited content patternsModerate1
Site
No keyword stuffing
Keyword stuffing actively hurts in generative engines (−1.5pts)Proven1
Site
📅FreshnessWebsite2 pts
CheckWhat it verifiesEvidenceWtApplies to
Visible published/updated date
AI cites content ~25% fresher than classic search; ChatGPT favours content updated within 90 daysModerate1
Site
Fresh dateModified in schema
A recent dateModified in structured dataModerate1
Site
Agent actionsProfile7 pts
CheckWhat it verifiesEvidenceWtApplies to
Transactional actions enabled
Actions let agents transact (book, order, quote), not just answerModerate3
Profile
Booking backed by services
Real services behind the booking action so it actually worksModerate2
ProfileBookings
Reachable (callback / message)
A callback or message path an agent can triggerModerate2
Profile
🔧Structured data & deliveryBoth16 pts
CheckWhat it verifiesEvidenceWtApplies to
schema.org structured data
65–71% of AI-cited pages carry structured data — it feeds the knowledge graphs AI readsStrong4
Both
Multiple schema types
Depth — Organization + Product/Service + Reviews + FAQ — improves citation candidacyModerate2
Site
Server-rendered content
Content that loads via JavaScript is often invisible to AI crawlersProven3
Both
sitemap.xml
A valid sitemap for discoveryModerate2
Both
Meta description
A meta descriptionWeak1
Both
Page title
A page titleModerate1
Both
Open Graph / social tags
Open Graph / social meta tagsWeak1
Both
HTTPS
Served over HTTPSModerate1
Site
robots.txt present
A robots.txt file existsWeak1
Both
llms.txt for AI assistants
Forward-looking hygiene — but ~0 AI crawlers read llms.txt today, so it doesn't yet move visibilityWeak0
Both

Live AI visibility — not scored into readiness

Separate from the 0–100 readiness score, we ask the assistants your customers actually use five real, intent-varied buying questions built from your category and region — then read back who they recommend.

GChatGPT OpenAI
  • 5 real buying questions
  • Built from your category + region
  • Read back who it recommends
CClaude Anthropic
  • 5 real buying questions
  • Built from your category + region
  • Read back who it recommends
Gemini Google
  • 5 real buying questions
  • Built from your category + region
  • Read back who it recommends
Surfaced? — are you named at all Hit rate — answers out of 5 Visibility % — across all queries Best rank — your top position Share of voice — you vs rivals Competitors — who's recommended instead

Crawler access audit

Your robots.txt is parsed for all 12 known AI crawlers. Search & live-fetch bots decide whether you can be cited; training crawlers decide whether you inform future models. Blocking a search bot is a hard gate.

Search / answer

Cite you in AI results — blocking one is a hard gate

OAI-SearchBotChatGPT
PerplexityBotPerplexity
GooglebotGoogle AI Overviews
BingbotCopilot / ChatGPT
Claude-SearchBotClaude
ApplebotApple / Siri

Live fetch

Fetch your page the moment a user clicks a citation

ChatGPT-UserChatGPT
Perplexity-UserPerplexity
Claude-UserClaude

Training

Whether you inform the next generation of models

GPTBotChatGPT
Google-ExtendedGemini
ClaudeBotClaude