How the Autonomics AI Visibility Report Works
Complete documentation of the questions, platform checks, scores, competitor findings, completion rules and known limitations behind every Autonomics AI Visibility Report.
Purpose and audience
The Autonomics AI Visibility Report is a free diagnostic for service businesses. It measures how prominently a business appears across Google Search and customer-style questions tested through ChatGPT, Claude, Gemini and Google AI Mode.
Each report creates ten questions relevant to the business's services, customers and geographic market, then evaluates responses across four AI platforms—40 AI checks in total. It also examines Google visibility, website AI readiness, competitors recommended when the business is absent, and the improvements the business should prioritize first.
It is written for owners and marketing leaders of service businesses who need to know whether prospective customers can find them in AI answers and in Google, and what to fix first. The only inputs a visitor provides are their contact details and one business website address.
Inputs and website reading
We read up to ten publicly accessible pages on the submitted website, starting from the homepage and following internal links. Only content a normal visitor or crawler can reach is used; we do not attempt logins, forms or paywalled areas.
From that content we extract a structured business identity: business name and brand variants, primary and secondary categories, offerings, industry vocabulary and the geographic market the site claims to serve. Where a public Google Business Profile exists for the business, its details are used to corroborate the location and category.
Question creation and validation
Each report uses 10 commercial questions written from the extracted business identity. They are the kind of question a prospective customer asks when choosing a provider — mostly unbranded, so the AI platform has to decide who to recommend, with a small number of branded checks used to confirm the platform recognises the business at all.
- Questions must be supported by evidence from the website; the system does not invent services the site never mentions.
- Local intent (a city, region or “near me” framing) is applied only when proximity genuinely drives choice for that category and the site claims a service area.
- Questions are de-duplicated and validated for relevance before any paid search or AI-response work begins.
- If the website yields too little evidence to write reliable questions, the report stops rather than producing questions about a business we could not identify.
The four AI experiences
Every question is evaluated on ChatGPT, Claude, Gemini and Google AI Mode, so a complete report contains 40 AI-response checks (10 × 4). Each answer is stored individually with the platform, endpoint family, requested and resolved model, whether web search was used and the methodology version, so no two measurement surfaces are ever averaged together.
| Platform | Experience tested | Web search | Model selected | Model | Comparability caveat |
|---|---|---|---|---|---|
| ChatGPT | Model-based response with web search | Explicitly requested | Yes | Preference order: gpt-4o-mini → gpt-4.1-mini → gpt-4o | Comparable only with other reports produced on the same endpoint family, model and methodology version. |
| Claude | Model-based response with web search | Explicitly requested | Yes | claude-haiku-4-5-20251001, with the Haiku 4.5 family alias and then the newest Sonnet 4.5 as fallbacks. Opus builds are never used. | A model withdrawal changes the fallback, which is recorded per answer. |
| Gemini | Public Gemini interface | Inherent to the interface | No | Whatever the public interface serves at the time of the check | No model pinning is possible, so results move with the public product. |
| Google AI Mode | Google AI Mode search result | Inherent to the search experience | No | Not selectable | AI Mode is a search feature; results vary by location and rollout. |
The report analyzes web-enabled responses associated with ChatGPT and Claude, the public Gemini experience, and Google AI Mode search results. It is not intended to reproduce every personalized response that an individual user may receive in the consumer applications.
Autonomics uses third-party search and AI-response infrastructure to collect these answers, and stores the raw response for each check so a finding can always be traced back to what the platform actually said. Provider credentials, internal prompts and security controls are not published.
Valid answers versus provider failures
This distinction is the core of the methodology. An answer only counts once we have readable content from the platform.
- Valid: the platform answered and the business was cited, mentioned, or not mentioned. “Not mentioned” is a real zero-result observation and is counted in the denominator.
- Invalid: a timeout, provider error, malformed or unreadable response, unavailable platform or exhausted retries. These are not “not mentioned”.
- Invalid answers are never silently scored as zero. They are retried, and if they cannot be completed the affected score is withheld.
- An automatically delivered report must contain all 40 genuine answers under the current completion contract.
- A report that is manually reviewed in exceptional circumstances displays any unavailable information honestly, labelled as unavailable.
How each score is calculated
Per-platform AI visibility
round((valid answers where the audited business is mentioned or recommended ÷ 10) × 100)
Example: if ChatGPT produced ten valid answers and the audited business was mentioned or recommended in four of them, ChatGPT Visibility is 40. The score is published only when all 10 answers for that platform are valid; with nine valid answers the score is shown as not available rather than estimated from a smaller sample.
Google Visibility
Each relevant commercial keyword scores by ranking position (1–3: 100, 4–10: 70, 11–20: 35, 21–50: 10, below 50: 0). Those scores are averaged with a weight of log10 of the keyword's monthly search demand, clamped between 50 and 5,000 searches, so one very broad term cannot dominate. Branded, job-related and purely informational searches are excluded.
Position points are 100 for positions 1–3, 70 for 4–10, 35 for 11–20, 10 for 21–50 and 0 below position 50. Keyword-level opportunities are grouped into quick wins (positions 11–20) and declining terms (lost rankings, or a drop of five or more positions).
Website AI Readiness (readiness methodology v2)
Six weighted components: AI crawler access 15%, business clarity 20%, offering coverage 20%, customer-answer coverage 25%, trust and citability 15%, technical clarity 5%. Each component scores linearly as the share of its checks passed. A component that could not be checked earns no credit, and ceilings prevent a thin or inaccessible site from appearing fully ready.
Customer-answer coverage requires at least 60% of a question's distinctive terms to appear on the site, including at least one term unique to that question. Generic commercial vocabulary, the business name, the category and city names are excluded so a page cannot pass on filler words. When fewer than four pages could be read, the metric is reported as Homepage AI Readiness and flagged as a limited sample.
Local Visibility (shown only when proximity drives choice)
For categories where customers choose by proximity, a separate Local Visibility score combines Google Business Profile completeness, review volume and rating, published hours, website-to-profile consistency and the measured map position. It is omitted entirely when local proximity is not a real buying factor.
Overall AI Visibility (internal)
An aggregate AI figure is retained internally as the share of completed unbranded provider searches that mention the business, excluding unavailable answers. The current public report presents the four per-platform scores rather than a single blended number, so the aggregate is not published as a seventh metric. Reports produced under earlier methodologies keep their original calculation and label.
Competitors and recommendations
- A competitor is recorded only when a business is named in the sentence of an answer where the audited business was absent.
- Asset, infrastructure and tooling links are excluded, so the platforms' own plumbing can never appear as a competitor.
- Names are normalized (legal suffixes, domains and punctuation removed) so obvious variants merge into one entity, and the cleanest human-readable name is kept.
- Competitors are ranked by the number of recommendations made while the audited business was absent, then by platform and question coverage.
- Three prioritized recommendations are generated from the weakest measured evidence — never from unmeasured assumptions — and are ordered by expected impact.
Completion, verification and delivery
- Work targets 20 minutes and is abandoned as failed after 60 minutes of processing.
- Every stage and every individual check is saved as it completes, so an interrupted report resumes rather than restarting, and only missing or failed work is retried.
- Automatic delivery requires the full completion contract: all 40 genuine answers, saved scores, a generated public token and a verified public report link.
- Only then is the unlisted report link emailed to the requester.
- Reports that fall short can be reviewed manually by an Autonomics administrator; missing information stays visible as missing and the report is marked as a manual exception.
- A completed report is reusable and publicly readable for 14 days, after which the link expires and a fresh request produces a new report.
Data handling
- You submit your contact information and your business website address.
- The report reads publicly accessible content on that website.
- Third-party search and AI-response infrastructure processes the report's search and AI queries.
- No credit card is required at any point.
- Submitted information is not sold for third-party marketing.
- Autonomics does not intentionally use confidential client data to train its own general-purpose AI model.
- Report links are unlisted rather than authenticated: anyone holding the link can open the report, and it stays available for 14 days after completion.
- Privacy requests can be submitted to the address published in our Privacy Policy.
See the Privacy Policy for the full statement.
Known limitations
AI responses can change based on the model, prompt wording, search availability, location, personalization, sources and date. The report is a point-in-time diagnostic and does not guarantee future mentions, citations, rankings, traffic, leads, sales or revenue.
- Gemini and Google AI Mode expose no model selection, so their results move with the public products.
- Location and personalization affect real user answers in ways a single measurement cannot reproduce.
- Only publicly readable website content is assessed; content behind logins or rendered exclusively by client-side scripts may be missed.
- Ten questions is a sample of customer demand, not the whole of it.
- Scores from different methodology versions, endpoint families or models are not comparable and are never averaged together.
Autonomics Agency is not affiliated with, sponsored by or endorsed by OpenAI, Anthropic or Google. ChatGPT, Claude, Gemini and Google AI Mode are the trademarks of their respective owners and are named here only to describe what the report tests.
Versioning and changelog
A material change to a provider, endpoint, model, scoring rule or classification rule produces a new methodology version. Reports keep the version they were produced under; older labels are never rewritten.
v4 · effective 2026-09-14
ChatGPT moved from LLM Scraper to LLM Responses with an explicit model and forced web search, plus deterministic Claude model selection. Claude, Gemini and Google AI Mode measurement surfaces unchanged.
v3 · effective 2026-09-01
Commercial prompt methodology: ten unbranded provider questions; ChatGPT measured through the LLM Scraper (public ChatGPT interface).
Author
Written and reviewed by Nikolai Gerdzhikov, Founder and AI Search Strategist at Autonomics Agency. Autonomics Agency, a DBA of Epic Webcrafts LLC.
See it applied to your business
Request the free AI Visibility Report, or read a complete example first.