Evidence first. Limitations included.

One profile is not the whole product.

Profiles are independent and are not affiliations or endorsements. The provider/model used for a saved scan is shown in its report; configurations may change for future scans.

Perplexity and AI Overviews profiles imitate answer format using other models. They do not query those products or retrieve their results. Evidence views default to API model samples, with simulations available separately.

How share of voice differs from mention rate

Share of voice = your brand appearances ÷ all identified brand appearances in the selected sample × 100. Each named brand counts at most once per answer; an answer can contain multiple brands. Competitor aliases are not automatically merged beyond case and whitespace normalization. Names are model-assessed and should be reviewed.

Mention rate = answers that mention your business ÷ total sampled answers × 100. If no brands appear, share of voice is undefined, not zero. These metrics describe only the selected questions and source filter.

Sample size, bias and before/after comparisons

A free audit currently tests two questions against two API models: four answers, with one answer per question/model pair. Paid scans use tracked questions and the profiles included in the plan. Questions naming your business can raise mention rates; include unbranded discovery questions for a more balanced diagnostic.

These questions are not a random sample of customer demand. Answers generated in batches can be correlated, and models vary between runs. We do not report statistical confidence intervals or claim that a single score shift is a real market change.

The evidence and action views show a before/after delta only when query text, model IDs, profile types, sample counts and stored methodology match. An implementation date helps distinguish a later scan from a pre-change baseline. Even a matched comparison does not establish causality. Historical overview charts can contain different samples.

Actions and AI-generated content

Recommendations are hypotheses grounded in the sampled answers. They are not proof that a website is missing content or that competitors have stronger authority. The Action Center uses saved evidence and owner-supplied facts to create editable drafts. Missing facts remain TBD, or are omitted from structured data.

Checks are model-assisted review notes, not fact-checking or originality certification. No live web verification or plagiarism search is performed. You must review claims and publish manually. We do not support fake reviews, fabricated citations, impersonation, spam or guaranteed results.

Data handling & security

We store account details, business profiles, queries, scan answers, preferences and saved actions to operate your workspace. Passwords are hashed; paid data routes check account ownership. Reporting API keys are stored as hashes. Model output is treated as untrusted content rather than executable HTML.

Business context, sampled responses, drafting facts and chat messages may be sent to AI processing services to produce results. Do not submit passwords, payment details, patient information or other sensitive personal data. This is not a confidential-document processing service.

Anyone with a free-audit link can view its score and sampled answers. Paid workspace evidence and drafts are account-restricted. Report email consent is collected separately and does not enroll you in marketing.

The configured retention policy is 365 days for scans and lead requests, and 90 days for free audits and alerts. Cleanup is tied to the monitoring scheduler, which is not yet activated, so automatic deletion on those dates is not currently guaranteed. Saved actions and their evidence snapshots remain until deleted with the associated business/account by an authorized administrator.

Reports can be exported as JSON or printable HTML; action drafts as text or JSON-LD. Self-service account deletion and full-account export are not yet implemented. Contact us to request access or deletion; no automated fulfillment deadline is promised. Independent security certification, penetration testing and storage-encryption guarantees are not claimed here.