01
Website / first-party information
A limited same-site HTML crawl builds a business profile from pages Avantagram can retrieve.
01 / MeasurementMake the
Make the
unknown
measurable.
Avantagram measures what a configured AI model can find, mention, cite, and support about a business — then keeps evidence separate from visibility.
- Pages crawled
- Unmeasured
- Prompts executed
- —
- AI responses
- —
- Brands detected
- —
- Visibility
- —
- Evidence
- Not collected
The question
Your business exists online. What can AI actually observe about it?
A website is not a claim. A claim is not evidence. A mention in an answer is not proof. The product keeps those layers separate.
Content—
Information present on the site or profile.
Evidence—
A sourced, non-puffery snippet for a typed claim.
Visibility—
Brand observed in evaluated answers.
Content coverage
Not collected
Information found on the site or profile.
Evidence coverage
Not collected
Sourced, non-puffery support for a claim.
AI visibility
—
Brand observed in evaluated answers.
Observed mentions are not third-party verification. Not collected until a workspace has been analyzed.
The system
A measurement layer between your business and AI.
02
Retrieval
Headings, metadata, language, and observed JSON-LD are extracted. JavaScript is not executed.
03
AI prompt observations
Buyer-style prompts are sent through a configured model. Direct ChatGPT, Gemini, and Claude apps are not live integrations.
04
Structured interpretation
Answers are parsed deterministically for brand mentions and URLs. No second model is required to score them.
05
Evidence
First-party sources are classified as proofs of typed claims. Observed citations are not treated as verified.
06
Visibility
A 0–100 score is computed from stored executions — presence, ranking, competition, citations, sentiment.
07
Recommendations
Actions are proposed from telemetry. They are not automated outcomes.
Observation
First-party HTML, measured on our crawler.
Retrievability here means: can Avantagram fetch your public HTML? That is a different question from whether an external AI crawler visited the site.
Off-domain and private-network destinations are not fetched. This crawler is not a proxy for GPTBot, ClaudeBot, Perplexity, Google, or Bing.
- User-Agent
- Avantagram-WebsiteIntelligence/1.0
- Scope
- Same-site HTML only
- Depth / pages
- 2 / 30
- Time limit
- 60 seconds
- JavaScript
- Not rendered
- Sitemap / noindex
- Sitemap unused; noindex not honored
AI visibility
What a configured model says when asked about you.
Prompt families run against a configured model. Mentions and citations are read from stored replies. Evidence Coverage is a different measurement and is not this score.
A citation in an answer is observed text. It is not independent verification that the cited page supports the claim.
Workspace score
—
0–100 after stored executions exist
Prompt families
category · service · geographic · destination · program · comparison · decision · long-tail
Brand mention
Observed in a stored model answer.
Citation
URL emitted in that answer. Observed, not verified.
Score components
- Presence
- Ranking
- Competition
- Citations
- Sentiment
Evidence
Being mentioned is not the same as being proven.
Content
The site or profile contains information about a topic.
Claim
A typed statement: service, geography, quantity, identity, structure.
Evidence
A sourced, non-puffery snippet that can support that claim.
Observation
A mention or citation seen in an evaluated answer.
Verification
Not assigned today. Observed is not verified.
Claim
SERVICE · GEOGRAPHIC · QUANTITATIVE · STRUCTURED ENTITY
First-party
Website or profile snippet with a URL. Can support a typed claim if it is not puffery.
Third-party observed
Independent citation host seen in an answer. The page was not fetched. Not verified.
Business knowledge
A profile is material for prompts — not a certificate.
A workspace can exist without a website. Analysis begins only when there is a crawlable site and remaining capacity.
Metadata is not service proof. JSON-LD is not a publisher. Partnerships, ratings, and priced offers are not invented.
- Services / products
- Supported terms, copied from the profile
- Locations
- Market geography for prompts
- Topics / audiences
- Fill context, not derived slogans
- JSON-LD
- Observed structured fields only
Monitoring
Each analysis is a recorded run.
If capacity is exhausted, the product returns HTTP 429 and does not pretend the job is still running. That limit is an application safety cap, not a purchased quota.
- 01Understanding your businessstage
- 02Preparing AI research promptsstage
- 03Asking AI modelsstage
- 04Analyzing responsesstage
- 05Calculating visibility scorestage
- 06Preparing reportstage
Recommendations
Measure first. Then decide what to change.
Observe what can be retrieved. Understand what was mentioned. Decide whether the evidence supports the claim. Act only after that. Avantagram does not promise that acting will raise a score.
What can be retrieved and what was stored.
Mentions, citations, and coverage layers.
Whether evidence supports the claim.
Proposed from telemetry — not guaranteed outcomes.
01
Observe
02
Understand
03
Decide
04
Act
Discipline
Avantagram philosophy
Observed ≠ verified
Content ≠ evidence
Visibility ≠ certainty
Make the unknown measurable.
Create a workspace. Add a crawlable site when you are ready. Visibility is scored from stored prompt tests.