Methodology

AI Visibility score

AI Visibility score is the headline score: the share of AI answers that name your business, (cited ÷ total queries) × 100. It’s calculated per scan and tracked as a trend, because the same model rarely answers the same prompt identically twice — a single scan is a sample, not a verdict.

A first score of 0 is normal. It means the models being queried don’t currently name your business for the prompts tested — not that something is broken.

Readiness score (sub-score)

The Readiness score measures the signals AI reads about you. It comes from an audit across four areas: technical infrastructure (can AI actually read your site — clean markup, structured data, crawlability), content (does it answer the questions customers actually ask), authority (does your business have a verifiable identity — e.g. a Google Knowledge Graph entry or directory profile), and reputation (what your reviews look like across the platforms AI can read).

Unlike the AI Visibility score, this sub-score reflects things you directly control, so it can move within hours of a fix — well before AI engines catch up and start naming you more often.

Businesses without a website skip every website-related check in this audit entirely and are scored only on the signals that apply to them — see our no-website handling below.

Presence score (sub-score)

The Presence score covers directories & listings: how consistently your business appears, with the same name, address and phone, across the directories AI reads for your kind of business. Like Readiness, it feeds the AI Visibility score over time but is never folded into it.

Platforms tested

Which engines a scan queries depends on the plan: ChatGPT and Gemini on Starter and the free tier; ChatGPT, Gemini, Claude, and Perplexity on Growth; and on Agency plans also Google AI Overviews and Bing Copilot where those integrations are enabled. Each engine is called through its API, fresh on every scan — Perplexity, Google AI Overviews and Bing Copilot search the live web for their answer, while ChatGPT, Gemini and Claude answer from what their models already know.

Prompt methodology

Each business gets a small set of prompts drawn from a category-specific library — the kind of question a real customer would ask when looking for a business like yours — plus any custom questions you add. The same prompts run across every engine on your plan on every scan, so results are comparable scan to scan.

Location handling

Prompts incorporate your business’s location where that’s relevant to the query (e.g. "best plumber in [city]"), so results reflect how AI actually answers location-specific questions rather than a generic, unlocalized version of the same prompt.

Competitor detection

You can name specific competitors to track. Findelle also auto-detects competitors from the scan results themselves — any business an AI platform names instead of you, on a prompt you’re tracked against, is surfaced as a competitor with its own mention count.

Repeatability

AI answers are not deterministic — the same prompt can get a different answer on two consecutive runs, even with nothing about your business changed. This is why the AI Visibility score is presented as a trend across scans rather than a single pass/fail number, and why one low scan isn’t cause for alarm on its own.

Data freshness

Monitor runs a full scan monthly, with a lighter weekly check-in on one or two engines in between (the AI Visibility score comes from the monthly scan). Reputation data (review ratings, counts and themes) refreshes monthly. Both are timestamped so you always know how current a given score is.

Limitations

Findelle can tell you whether AI platforms currently name your business, and can point at the technical, content, authority and reputation signals correlated with being named. It cannot guarantee placement in any AI answer — no one can, and any product that promises otherwise is overselling. Findelle also can’t see inside a model’s training process; for platforms that answer from training data rather than live retrieval, a fix you make today may not be reflected in that platform’s answers until it’s next updated, which is outside our control and usually outside the platform’s own public timeline.