Grade C: exploratory
A small sample, a single model, a one-time observation, or incomplete evidence.
How it works
An AI answer can change from one run to the next. One run is not a measurement. Everything below is there so that a result can be traced back to the evidence behind it.
See it work
When AI answers, does it say your name? Here is how we check.
Every picture is labelled. Where a result is an example, it says so. Where it is real, it says that too.
Ask the AI what a buyer would ask.
We put buyer-style questions to answer engines through their official APIs and record who is named, and which sources are cited. Below: a real Gemini answer with Google Search on, and the same question put to Perplexity, both on 19 October 2026. Neither is a customer result.
What are the best AI search visibility monitoring tools?
Named in the answer
- Otterly.AI
- Profound
- Peec AI
- ZipTie.dev
- Semrush
- Ahrefs
- Similarweb
- Radarkit
- Orchly
- Scrunch AI
- AthenaHQ
- Vindhya Security (fictional) — not mentioned
Cited sources · 11 across 10 publishers
- thatmarketingbuddy.com
- otterly.ai
- zapier.com
- semrush.com
- nexscope.ai
- reddit.com
- +4 more
Google Search grounding on: a live-retrieval answer.
One real Google Gemini answer (gemini-3.8-flash), asked on 19 October 2026. Search grounding on. Vindhya Security is fictional. Kept as history: Google Gemini gemini-3.8-flash without search, 19 October 2026; Google Gemini gemini-3.8-flash without search, 19 October 2026; Perplexity Sonar, 3 August 2026. See the full sample result.
Named: Profound, Peec AI, Ahrefs Brand Radar, Conductor, Rankscale, Otterly.ai and others.
Cited publishers: frase.io, seo.com, blog.hubspot.com, 01net.com, searchengineland.com, +5 more.
Turn it into a score you can defend.
Code, not a model, works out the score from the evidence. Each number says who saw the evidence: you, our checks, or a person who verified it.
FOUND Score
Nothing leaves without your approval.
Agents prepare drafts and check them. You approve anything published or sent.
Waiting for your approval
- FOUND-FStrategist approves decision-grade findings and anything published or sent externally.Review
- FOUND-OReview requests, analyst briefings and outreach are sent only by a human or after explicit approval.Queued
- DRONICS-ONamed customers, quotes and figures appear only with written permission.Queued
Nothing goes out until a person approves it.
Later: connect the tools you already use.
The harness would let agents read from your CRM, analytics and content systems. It is a plan. It is not built.
Confidence is calculated, not simply stated
Grade B: directional
Enough coverage of platforms and prompts, with the gaps written down.
Grade A: decision-grade
Many models, many runs, place controlled, current evidence, checked by people, repeatable, and with corrections on record.
A reviewer may lower a grade and give the reason. A reviewer may not raise it above what the system allows.
The score card’s High/Medium/Low label is different: the share of points from direct reads (provenance), not statistical confidence, completeness or human verification.
LiveShare of Answer is reported in parts
Inclusion rate, recommendation rate, citation rate, accuracy rate and competitive-win rate are reported separately. We never fold them quietly into one visibility number.
Engine coverage, as it stands today
Our sampling method needs three engines. Three adapters are built. Each runs only when its key is set. Here is the true state of each.
Answers differ a great deal between engines. A company present in one can be missing from another for the same question. So, until all three run, limited coverage is a limit on every result.
- Perplexity SonarAdapter built and tested against a real call: 20 citations parsed from one response on 2026-08-03, cost reported by the provider at $0.005670. That observation is what the sample assessment is built from; it was captured in Syntax, our measurement workspace. The adapter is now connected to the live assessment here; Perplexity Sonar is the only live answer engine in production today.
- Live
- Google GeminiThe citations come in the grounding metadata. Adapter built; it ran live on 19 October 2026. Gemini is deliberately paused in production: its key is not set for the application.
- Demonstration
- OpenAI ResponsesA stand-in for the ChatGPT answer that consumers see. It is not an exact copy of it. Adapter built, but not run live. It runs only when an OpenAI key is set.
- Demonstration
What we do not claim
We do not control what models answer. We do not manufacture reviews, testimonials, analyst opinions or awards. A finding is an association we observed within a stated period and sample. It is never a promise of cause.
LivePut the method to work on your own site.
Free, open and no login.