Methodology

Data sources

Every report is built from publicly available signals only. We do not access private data, direct messages, or information behind authentication walls. Specifically:

• Platform metrics: follower counts, reach, impressions, engagement rates, posting frequency, and content categorisation data retrieved via platform APIs where available.
• Public web signals: domain authority, backlink profiles, search presence, and audience geography where public web intelligence is available for the profile under review.
• Cross-platform comparison: where signals are available across multiple platforms, discrepancies and consistencies are flagged explicitly.
• Historical records: prior analysis runs on the same profile are compared to surface material changes over time.

We do not scrape private profiles, purchase third-party data feeds, or use data that the subject has not voluntarily published in a public commercial context.

How confidence levels are decided

Confidence is not a guess — it is a function of data availability, cross-sign corroboration, and the specific type of finding being assessed.

Confidence is calibrated continuously through adjudicated accuracy review — when an admin reviews a prior analysis and confirms or disputes a finding, that feedback lowers the confidence shown on the same type of finding in future reports. Confidence can only be lowered by review feedback, never raised by it.

Claims we will never make

Conservatism is written into the product. These boundaries are not optional — they are enforced at the report generation stage:

• No fraud judgments. We will not declare a profile "fake", "fraudulent", or "inauthentic" as a definitive conclusion. We flag suspicious signals and assign them a confidence level.
• No identity verification. We will not claim to confirm or deny the real-world identity of any individual. That is outside the scope of this tool.
• No legally binding determinations. Reports are informational and advisory. They do not constitute legal, financial, or compliance opinions.
• No credit scoring or employment screening. Our outputs are not used for, and must not be used for, any rights-affecting decision — no lending, hiring, housing, or insurance underwriting.
• No AI training on subject data. Profile data submitted for analysis is never used to train any AI or machine learning model. Data sent to LLM providers during analysis is used solely to process that specific request, and providers are contractually prohibited from using it for model training.
• No data resale. Intelligence reports are delivered to the requesting customer only. Profile data and report content are never re-sold as raw data or made publicly searchable.

Why we stay conservative

The influencer marketing industry faces an estimated $4.8 billion annual problem with fraudulent reach and misrepresentation. The tools used to fight it often make the problem worse — burying buyers in dashboards full of confident-sounding numbers that cannot be defended when challenged.

Pinkk was built in response to that. We started in iGaming and gaming communities — where inflated signals and inauthentic traffic are most aggressive — and built the methodology to resist overclaiming. Every report uses cautious, qualified language throughout: "may indicate", "suggests", "could be". Strong language is reserved for findings that earn it with strong evidence.

A result that says "we're not sure" is more valuable than one that sounds certain but isn't. We'd rather under-promise and be right.

What a report contains

Every intelligence report follows the same structured format, regardless of profile size or vertical:

• Executive summary with verdict and confidence level — stated up front so you know what you are reading before you dig in.
• Key findings table with evidence for each finding and its confidence rating — no buried caveats.
• Traffic source analysis — where audience signals come from and what that means for reliability.
• Audience quality assessment — composition, engagement patterns, and anomaly flags.
• Compliance context — relevant signals for regulated sectors including iGaming, finance, and health-adjacent verticals.
• Recommended next steps — what to do with the findings, including what additional research is needed before a decision is safe.

Reports are versioned and stored. Every run on the same profile produces a dated, verifiable record — so you can build a documented trail rather than relying on a single snapshot.

AI assistance and human review

Pinkk uses AI as an analysis assist — to surface patterns, flag anomalies, and structure findings. It does not make the final call. Human review is applied at the adjudication stage, where findings are assessed against the raw evidence before they reach a report.

When human reviewers disagree with an AI-generated confidence level, that feedback is recorded and used to recalibrate confidence on the same type of finding in future analyses. Confidence is a measured output, not a marketing label.

All outputs are informational and advisory. Customers remain solely responsible for any downstream decisions they make based on a report. This is stated explicitly in every report and reinforced in our Terms of Service.