How every number on a report is measured

Engines

A free scan asks 5 engines. Paid reports (Audit, Deep and every paid re-scan) add a sixth.

• ChatGPT — third-party engine, asked through its API.
• Claude — third-party engine, asked through its API.
• Gemini — third-party engine, asked through its API.
• Perplexity — third-party engine, asked through its API.
• Lenscore Retrieval Engine — ours. Lenscore Retrieval Engine: a search over an independent web index followed by our own synthesis, so the frontier engines are not the only witnesses.
• Google AI Mode — third-party engine, asked through its API. Paid tiers only.

Each question is asked once per engine, on the day of the scan. Where an engine can be told to search the live web, it is; every answer records whether it actually searched and the report shows “Answers from live search — N of M”. An engine that answers from memory is shown as such, not hidden. An answer we cannot read — another script, a refusal, or no list of names — is recorded as not observed with the reason, excluded from both sides of every ratio, and counted in the report.

Questions

Every scan builds a benchmark of 25 buying questions: 10 core questions fixed per business type (so scans compare like for like over time) and 15 adaptive questions generated from what your site actually sells and where. They are spread evenly over five buyer-journey angles: discovery, comparison, use-case, trust, purchase. The free scan asks 8 of the 25, at least one per angle; the Audit unlock tops up to all 25.

A buying question is what someone types when they are ready to choose, not a brand search. Creators get questions from both the audience side and the brand-buyer side; local businesses get city-scoped questions.

KPIs

• Named in N of M — questions in which any engine named you. The headline number.
• Recommendation rate — questions in which an engine placed you first or second.
• Recommendation share of voice — your recommendations as a share of all recommendations in your category questions.
• Ordinal prominence — where in each answer you appear, weighted (first counts most).
• Citation rate / share / authority — how often your own pages are the cited source and how strong the sources that feed your category are.
• Entity resolution / consistency — whether engines describe you accurately and whether your name and description agree across your site, schema and Wikidata.
• Volatility — only measured when a question is asked more than once per engine; otherwise shown as not measured.
• Answers from live search — how many answers came from a real web search rather than the model’s memory.
• Answers we could not read — how many replies came back in another script, as a refusal, or without a list; they are excluded, never counted as zero.

Anything we could not measure on a given scan is shown as not measured, never as zero.

The composite score

The LensCore Discovery Index combines five factors (presence, prominence, recommendation, citation, entity) into one number out of 100. It is labelled a weighted blendand carries a formula version because the five inputs are measured but the weighting of them is our judgement; the raw KPIs above it are the measurement. It is called a blend rather than “experimental” because a weighting is always a judgement — that will not change with time, so a word implying it might would be misleading. Same inputs, same score: the index moves when you or the engines change, never on its own.

Versions

Every report and every monitoring snapshot is stamped, so two reports are only compared when they were measured the same way.

• Methodology 2026.09.s70
• Benchmark core-25.v1 · question packs packs-5x5.v1
• Prominence weights ordinal.v1 · Discovery Index weights geo-5factor.v1
• Citation metrics citation.v1 · volatility volatility.v1 · entity metrics entity.v1

Reports measured before versioning existed say “pre-versioning”. Changes are listed in the changelog.

Tiers

• Free scan — 8 of 25 questions, 5 engines, score, top rival, every issue named, one fix. No email needed to see it.
• Audit (₹2,499 / $49) — all 25 questions, six engines, every rival, every fix, the PDF. Does not include verbatim per-engine answers, the source-authority map or a follow-up re-scan.
• Deep (₹5,999 / $129) — everything in Audit (the full Audit report is included), plus per-engine verbatim answers, the citation/source-authority map, entity gap, wrong-answer log, a 90-day roadmap and one follow-up re-scan.
• Founder Bundle (₹7,999 / $165) — Audit + Deep + one month of monitoring.
• Monitoring (₹1,999 / $39 per month) — weekly re-scans, alerts by email, wins/losses digest. Creators: ₹499 / $9 per month.
• Agency workspace — white-label reports and PDFs, teammates, one client dashboard, a monthly allowance of prospect scans (10 on 5 client slots, 20 on 10, custom above).

Monitoring

Every Sunday the same 25 questions are asked again on the same engines with the same methodology version. The digest shows what moved, which rivals gained and which fixes are still open. Alerts go by email. Cancel any time from your account.

Data

We read your public website and public profiles, the same pages any visitor can open. We keep the report, the raw engine answers behind it (so a number can always be traced to an answer), the evidence screenshots and your email if you gave one. Report screenshots are served from our asset store. Reports are private to you unless you create a share link. We do not sell data and do not use one brand’s report to inform another’s. Details in the privacy policy.

Limits and guarantees

• One pass per engine is a sample, not a census. Each of the 25 questions is asked once per engine. That is enough to see whether an engine reaches for you at all, and not enough to measure a stable percentage. Read a 3-of-25 as "rarely", not as 12.0%.
• AI answers move on their own. The same question can be answered differently next week without you or your rivals changing anything, because the engines change. A re-scan measures the engines on that day, which is why a single audit is a starting position rather than a permanent verdict.
• We audit the questions buyers ask, not every question. The 25 questions are generated for your category and market. They are the commercial ones — what someone asks when they are ready to buy. A brand can be well known and still score low here, because being famous and being recommended at the point of purchase are different things.
• Same formula, sampled answers. The score formula never changes: the same answers always produce the same number. The answers it reads are one day's sample from engines that change on their own. So two scans a week apart can differ, and the difference is the engines moving, not the method.
• The category behind your questions may be read by a model. When our rules cannot place what you sell, a model reads your site text and picks one label from a fixed list. The report says so where it happens, and the confidence is capped at medium. A re-scan can resolve it differently; telling us your category makes it a declared input instead.
• When we could not check something, we say so. A check that fails to run is recorded as not observed, and appears that way in the report. It is never rounded down to a zero, because a zero is a finding and a failure to look is not.

• Deterministic scoring: the same inputs always produce the same score. No black box, no mood.
• We label estimates as estimates. A live AI measurement is never dressed up from a guess.
• Every question is unbranded. We ask what a buyer asks — "best X for Y" — never "is <your brand> good?". Being named is earned in the answer, not planted in the question.
• No guaranteed rankings or revenue. We report your current standing, not a promise.
• No fabricated testimonials or reviews, ever. The sample report is our proof, not invented quotes.
• No fake countdowns or pressure. The only urgency is real: AI answers change as you and rivals publish.

More on the trust page.